See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/344680984

The use of machine learning and deep learning algorithms in functional

magnetic resonance imaging—A systematic review

Article in Expert Systems · October 2020

DOI: 10.1111/exsy.12644

CITATIONS READS

52 1,630

3 authors:

Mamoon Rashid Harjeet Singh

School of ICT gisma business school germany
181 PUBLICATIONS 2,968 CITATIONS 13 PUBLICATIONS 165 CITATIONS

SEE PROFILE SEE PROFILE

Vishal Goyal
Punjabi University
63 PUBLICATIONS 648 CITATIONS

SEE PROFILE

All content following this page was uploaded by Mamoon Rashid on 17 October 2020.

The user has requested enhancement of the downloaded file.

Received: 19 January 2020 Revised: 12 July 2020 Accepted: 21 September 2020
DOI: 10.1111/exsy.12644

INVITED REVIEW

The use of machine learning and deep learning algorithms in

functional magnetic resonance imaging—A systematic review

Mamoon Rashid1 | Harjeet Singh2 | Vishal Goyal3

1
Department of Computer Science &
Engineering, Punjabi University, Patiala, India Abstract
2
Department of Computer Science, Mata Gujri Functional Magnetic Resonance Imaging (fMRI) is presently one of the most popular
College, Fatehgarh Sahib, India
techniques for analysing the dynamic states in brain images using various kinds of
3
Department of Computer Science, Punjabi
University, Patiala, India algorithms. From the last decade, there is an exponential rise in the use of the
machine and deep learning algorithms of artificial intelligence for analysing fMRI data.
Correspondence
Mamoon Rashid, Department of Computer However, it is a big challenge for every researcher to choose a suitable machine or
Science & Engineering, Punjabi University, deep learning algorithm for analysing fMRI data due to the availability of a large num-
Patiala, India.
Email: [email protected] ber of algorithms in the literature. It takes much time for each researcher to know
about the various approaches and algorithms which are in use for fMRI data. This
paper provides a review in a systematic manner for the present literature of fMRI
data that makes use of the machine and deep learning algorithms. The major goals of
this review paper are to (a) identify machine learning and deep learning research
trends for the implementation of fMRI; (b) identify usage of Machine Learning Algo-
rithms and deep learning in fMRI, and (c) help new researchers based on fMRI to put
their new findings appropriately in existing domain of fMRI research. The results of
this systematic review identified various fMRI studies and classified them based on
fMRI types, mental diseases, use of machine learning and deep learning algorithms.
The authors have provided the studies with the best performance of machine learn-
ing and deep learning algorithms used in fMRI. The authors believe that this system-
atic review will help incoming researchers on fMRI in their future works.

KEYWORDS

algorithms, artificial intelligence, deep learning, functional MRI, machine learning, systematic
review

1 | I N T RO DU CT I O N

From the last couple of years, researchers have shown a lot of interest in using machine learning and deep learning algorithms for analysing data
with functional magnetic resonance imaging (fMRI). In every such research, an attempt is made to extract valuable information from fMRI data
(Norman, Polyn, Detre, & Haxby, 2006). After the evolution of fMRI, researchers started making use of machine learning and now deep learning
algorithms which are branches of artificial intelligence. This research provides a valuable contribution for validating and selecting machine learning
and deep learning models and outlining the practical evaluations required for various aspects of models related to brain imaging data (Lemm,
Blankertz, Dickhaus, & Müller, 2011). At present, there is an abundance of Machine Learning Algorithms like K-nearest neighbours (Patrick &
Fischer, 1970), Naive Bayes classifier (Rish, 2001), support vector machines (Tong & Koller, 2001) with applications on domains of spam filtering,
predictions in real-time and healthcare systems. In the last few years, there is an exponential rise in applying these Machine Learning Algorithms
for the study of brain images. While machine learning remains a popular choice for brain images, alternate methods of deep learning are gaining

Expert Systems. 2020;e12644. wileyonlinelibrary.com/journal/exsy © 2020 John Wiley & Sons Ltd 1 of 29
https://doi.org/10.1111/exsy.12644
2 of 29 RASHID ET AL.

considerable importance in the study of brain images (Bengio, 2009). Earlier it was quite possible to read all fMRI literature in a couple of days
whereas it is now not possible to read existing papers that are even published last fortnight. The reason for this growth is machine learning along
with deep learning applied to fMRI predicts the behaviour from brain images with excellent spatial resolution and acceptable temporal resolution
in comparison to the earlier methods like MRI.
This paper provides a review in a systematic manner about the usage and application of machine learning and deep learning algorithms in
fMRI. The study is drawn in such a way that it will provide a sequential way of understanding the various trends in these algorithms which are
used in research and development. The motive of this systematic review is to bring various studies of machine learning and deep learning on fMRI
together so that researchers will obtain more information ahead to make better implementations. The major goals of this review paper are to
(a) identify machine learning and deep learning research trends for the implementation of fMRI; (b) identify questions related to the usage of
Machine Learning Algorithms and deep learning in fMRI; and (c) help new researchers based on fMRI to put their new findings appropriately in
existing domain of fMRI research. This survey study tried to identify various existing works of fMRI analysis, classify various machine learning and
deep learning approaches used for fMRI, and discuss the technologies related to fMRI research and utility of various algorithms along with their
performance metrics on fMRI.
This review paper is outlined as follows: Theoretical background of fMRI, machine learning and deep learning are given in Section 2. A sys-
tematic review of all related and identified works is given in Section 3. Results and inferences drawn from various existing researches are given in
Section 4. Open issues in fMRI based on machine learning and deep learning are introduced in Section 5. Conclusion and future work are drawn
in Section 6.

2 | T H E O R E T I C A L BA C K G R O U N D

This section provides an overview of fMRI technique, machine learning, and deep learning.

2.1 | Functional magnetic resonance imaging

Functional MRI is a procedure used for measuring the activity of the brain by detecting low-frequency blood oxygen level dependent (BOLD) sig-
nals (Kwong, 2012). fMRI procedure observes the flow change in blood by mapping the activity of neurons of the brain related to the use of
energy. The advantage of methods used in fMRI is the increase in the blood through the area where neurons are getting activated in the brain.
The BOLD technique is the most commonly used practice for performing functional mapping of the brain (Ogawa, 2012). BOLD signals are com-
posite and reflect regional cerebral blood flow changes along with its oxygenation and volume. The magnitude of these signals measures the activ-
ity of neurons in indirect ways. Functional MRI technology became so popular in brain imaging as people do not need to go for surgery, to get
exposed to ionizing radiation, or to ingest any substances (Sörös & Witt, 2018). Several fMRI studies are available in terms of literature which have
extensively explained and reviewed its basic concepts and core principles (Amaro & Barker, 2006; Atenas, Díaz, Quiroga, Arancibia, &
Rodríguez, 2018; Deroy, 2019; Faro & Mohamed, 2006; Wig, Schlaggar, & Petersen, 2011). In all these kinds of studies, the fundamental principle
in fMRI experiments for measurements is an increase in blood flow and higher consumption of energy with an increase of neuronal activity. A sta-
tistical map is used for the representation of brain function which often reflects the activity of regions. Activation of signals in these maps can be
represented by the strength of colour-codes across the region of the brain which is under study. Apart from detecting BOLD responses from vari-
ous tasks due to their stimuli, the fMRI technique includes resting-state fMRI. Resting-state fMRI is responsible for measuring the BOLD variance.
Functional MRI finds its usage in the clinical and research world. Functional MRI can be useful in combination with other techniques of near-
infrared spectroscopy (NIRS) and electroencephalography (EEG) as well for brain neuronal activities. Research is being done to discover newer
methods that will improve both time and spatial resolutions. The fMRI technique has led to the expansion of cognitive neuroscience in many areas
significantly (Poldrack, 2012). The usage of fMRI lies in drawing conclusions related to neuronal mechanisms of cognitive states covering from
memory to recognition. Now there is a further possibility of viewing live brain activation states using real-time fMRI (DeCharms, 2008). fMRI also
finds its use in topics revolving around neuromarketing (Kuhn, Strelow, & Gallinat, 2016) and addiction behaviour (Kober et al., 2016).

2.2 | Machine Learning

Machine Learning is a branch of artificial intelligence where machines can acquire and identify intelligence based on knowledge from the real
world without explicitly being programmed and later performance of tasks can be improved based on this acquired knowledge. Logically machine
learning is defined in terms of three parameters: Experience (E), Task (T) and Performance (P) as follows: For any tasks T, a computer program is
going to learn from experience E with performance P, if the performance in the computer program at tasks in T, which is measured by P, improves
RASHID ET AL. 3 of 29

with experience E (Michalski, 1986). The idea of machine learning comes after training of data along with algorithms. There is rich literature pres-
ently available related to the usage of Machine Learning Algorithms. These algorithms are classified into four types based on their learning
process-supervised learning, unsupervised learning, reinforcement learning, and semi-supervised learning (Portugal, Alencar, & Cowan, 2018).
Supervised learning takes place when algorithms are provided data along with labels. Data will be used to train algorithms and labels are cor-
rect answers for each kind of trained observation. The main role of the machine learning algorithm is to learn and get intelligence based on the
real data which is fitted in terms of training data. The idea is to split data into training and testing parts. The training part of data is supplied to an
algorithm and prepares a machine learning model. Later the performance of the model is compared in terms of its output predictions to that of
the test part of data which contains correct answers (Taherkhani, Belatreche, Li, & Maguire, 2018). As an example, consider a machine learning
algorithm which is used for classifying houses based on house prices for some real estate agent. Data related to houses is required with different
features like number of bedrooms, size in square feet, neighbourhood, and a sale price of the house. Among all the given observations of data
related to houses, the training set is prepared to contain training input data with labels as output for each input. The machine learning algorithm is
trained with a training set to learn. When a new house comes to a real estate agent, the trained algorithm can easily predict the sales price of a
house based on the intelligence acquired while training the model.
Unsupervised learning-based algorithms contain some real data and do not have a training data set. These kinds of algorithms need to learn
from this data on their basis (Krotov & Hopfield, 2019). Unsupervised algorithms are known to find hidden patterns in data. As an example, if
someone is providing you data in terms of numbers who hardly have any understanding about these numbers. Chances are there about some pat-
tern or grouping may result when one will analyse these numbers. This is the rule on which unsupervised learning is based on. Taking back an
example of a real estate agent, if one knows features like number of bedrooms, size in square feet, neighbourhood but this time did not know
about the sales price of a house, then finding hidden patterns for predicting sale prices is unsupervised learning.
Semi-supervised learning-based algorithms are provided with data containing training set with missing information in observations and these
algorithms need to learn only out of this training data set (Zhu, 2017). As an example, unsupervised algorithms find their use recommender sys-
tems for prediction of movie ratings. For any particular movie, it is not necessary that every user is going to rate it and thus higher chances are
that machine learning will be applied to data with missing information. The beauty of semi-supervised algorithms lies in learning and provides
valuable conclusions even if the given data is not complete.
Reinforcement based learning algorithms gain or learn knowledge-based feedback given externally either by an environment or some thinking
based entity (Szepesvári, 2010). Reinforcement learning maximizes its performance by allowing software agents and machines to determine ideal
behaviour automatically in particular contexts. Reinforcement learning makes learning by using five states—Internal state which makes learning
about the environment, Reward function which is always used for training agents to show them how to behave, environment is a situation which
agent will face at any time, action is the activity performed by agent in environment and agent fulfils and performs all these tasks. In reinforce-
ment learning, if actions are performed correctly then it turns to be positive feedback and reward and if actions are not performed correctly, then
it turns to be negative feedback and results in punishment. As an example, the clicker makes a pet dog learn various actions in terms of sitting and
jumping. Every time this dog will respond correctly, a reward will result in pet and thus good behaviour and whenever action is not performed by
dog wrongly, then it will get punished and thus an act of negative behaviour.

2.3 | Deep Learning

Deep Learning is a machine learning technique that makes use of its algorithms for learning in similar ways of artificial neural networks. In other
terms, deep learning algorithms learn in ways as our brain functioning takes place. Deep Learning allows various computing models with
processing layers to learn data representations (LeCun, Bengio, & Hinton, 2015). Deep learning algorithms follow the work approach of nervous
systems where neurons are connected which are responsible for passing information. The models in deep learning follow a layered structure and
simple deep learning models contain three layers. In such layered models, the information is always is accepted by any particular layer from the
previous layer and then its output acts as an input for the next layer (Schmidhuber, 2015). The performance curve in deep learning algorithms is
marginally quite high in comparison to machine learning models which stop performing better after reaching to saturation point as shown in
Figure 1.
The best part of deep learning lies in no need for use of feature engineering before training deep network models. The input data is directly
passed to deep networks and models come out with good performances in terms of accuracy. This nullifies the feature selection stage in machine
learning models where it is required manually to select the best possible features for optimum results (Humphrey, Bello, & LeCun, 2013).
Most popular deep architectures for implementing deep learning algorithms are Multi-layer Perceptron, Convolutional Neural Networks, and
Recurrent Neural Networks. Each of these deep architectures will be explained with their main features and in context to data for which these
architectures are to be used (Bengio, 2009).
4 of 29 RASHID ET AL.

F I G U R E 1 Deep learning performance comparison with traditional

learning algorithms (Ng, 2015)

F I G U R E 2 Deep learning multi-layer

perceptron architecture (Ronaghan, 2018)

F I G U R E 3 Convolutional
neural network architecture
(Ronaghan, 2018)

Multi-layer perceptron (MLP) based deep architectures may contain many layers but minimum layers to be used are three: input layer, hidden
layer, and output layer (Bengio, 2009). In MLP, neurons at any particular layer are connected to all neurons in the previous layer and are also con-
nected to neurons of the next layer as shown in Figure 2.
MLP architectures are much suited for structured kinds of data. The key point about the MLP model is to make use of it as a baseline model
for accuracy before ensuring significant improvements in complicated architectures.
Convolutional neural network (CNN) is a neural network that is mainly used for image recognition and classifications. In the CNN model, an
input image is passed through several convolutional layers with pooling, filters and fully connected layers (Krizhevsky, Sutskever, & Hinton, 2012).
Convolution layer is the first layer that extracts features from a given image and always preserves relation among pixels by learning features of
the image in small squares of a given image. The parameters in large images are often reduced with the help of the pooling layer on CNN. After a
series of convolution and pooling layers on CNN, one final neural layer is flattened which is then passed to a fully connected network to target
the output layer as shown in Figure 3.
Recurrent Neural Network (RNN) is a neural network which makes use of an internal state of memory for processing various sequence of
events. RNN's are mostly used in speech and handwriting recognition. RNN works as states which act as memory and helps important aspects
which happened in the past and based on them predicts the aspects to be followed next (Yin, Zhu, Fei, & He, 2017). In RNN, if input data is in the
form of data series, then it usually considers features at time intervals t−5, t−4 ……t−1 to make predictions at time interval t as shown in Figure 4.
RASHID ET AL. 5 of 29

F I G U R E 4 Recurrent neural network

architecture (Ronaghan, 2018)

3 | SYSTEMATIC REVIEW

To do fMRI analysis for studying the functioning of a healthy brain and how it is getting affected by different kinds of mental diseases, various
health practitioners and researchers need to make use of the right algorithms for leading solutions to these problems. The use of these algorithms
in fMRI analysis is carried out and then resulting neuroimaging data is passed through the required steps of analyses to study the functioning of
the brain and to predict various diseases that are affecting the brain. The abundance of these algorithms and their combinations and variations in
literature makes it a challenging task. Moreover, these algorithms are continuously growing which further makes the study of fMRI analysis com-
plicated. Keeping all these issues under consideration, the authors have restricted studies related to Machine Learning Algorithms and deep learn-
ing architectures which are used in fMRI to study its functioning and prediction of various mental diseases. The authors have filtered studies in
fMRI domain covering algorithms used in the classification of fMRI types in terms of resting-state and task-based, algorithms related to mental
diseases and algorithms related to fMRI methods involved. The procedure followed in this systematic review is based on (Portugal et al., 2018)
and is addressing the following goals:

1. Identification of resting-state and task-based fMRI studies where Machine Learning Algorithms and deep learning architectures are used.
2. Identification of various studies related to mental diseases based on fMRI where Machine Learning Algorithms and deep learning architectures
are used.
3. Identification of various methods used in fMRI research based on Machine Learning Algorithms and deep learning architectures.
4. Identification of ways for assisting researchers in the future with proper categorization of studies used in fMRI analysis.

The approach used in this systematic review is based on fMRI studies with either the use of the fMRI method with the application of a
machine learning algorithm or deep learning architecture. All studies considered were first checked out for the presence of validation of results
and support of the experimental study. Based on this condition, only such studies were included in this systematic review where the application
of Machine Learning Algorithms and deep learning architectures are analysing the performance metrics of the fMRI technique based on a parame-
ter of accuracy.
The inclusion/exclusion of various papers in this systemic review is done based on three steps. In the first step, all publications related to
fMRI are gathered from most popular scientific databases like Science Direct, Scopus, Web of Science and IEEE Xplore. The authors analysed all
these publications on one to one basis and excluded papers based on first criteria of exclusion as given in Table 1. In the second step, papers that
are filtered based on criteria in the first step were analysed further by reading their abstracts and based on this, the second criterion of exclusion
was applied as given in Table 1 to eliminate survey kind of papers. In the third step, the papers left are analysed by checking their methods and
workflow and then the third criterion of exclusion is applied where papers and studies are excluded based on absence in terms of methods and
experimental study related to fMRI. For improving the quality of this systematic review, the authors applied these criteria of exclusion (COE) on
various studies based on the following proposed protocol.

COE1: Research articles that are not peer-reviewed are not considered for review.
COE2: Book chapters, White papers, and Patents are not considered for review.
COE3: Thesis and Dissertation works done at the graduate and post-graduate levels are not considered for review.
COE4: Studies written in languages other than English are not considered for review.
COE5: Survey papers where any fMRI method is not implemented is not considered for review.
COE6: Studies whose abstracts were available but the entire length of paper were not accessible were not considered for review.
COE7: Studies where datasets, methods, and algorithms were repeated and having similarities were not considered for review.
COE8: Studies not having the presence of fMRI methods with machine learning or deep learning algorithms were not considered for review.
COE9: Studies where the methods are proposed but experimentation and validation are not conducted are not considered for review.
6 of 29 RASHID ET AL.

TABLE 1 Results achieved for number of studies with criteria of exclusions

Number of
Parameters studies
Total number of studies 360
First criteria of COE1 Research articles that are not peer-reviewed are not considered for review. 3
exclusion COE2 Book chapters, communications, and Patents are not considered for review. 6
COE3 Thesis and Dissertation works done at the graduate and post-graduate levels are not considered 1
for review.
Second criteria of COE4 Studies written in languages other than English are not considered for review. 1
exclusion COE5 Survey papers where any fMRI method is not implemented is not considered for review. 15
COE6 Studies whose abstracts were available but the entire length of paper were not accessible were not 3
considered for review.
Third criteria of COE7 Studies, where datasets, methods, and algorithms were repeated and having similarities were not 83
exclusion considered for review.
COE8 Studies not having the presence of fMRI methods with machine learning or deep learning 4
algorithms were not considered for review.
COE9 Studies where the methods are proposed but experimentation and validation are not conducted 3
are not considered for review.
Total number of studies retained 241

The search queries SQ1 and SQ2 are used for gathering studies in this systematic review inspecting the title, abstract and keywords of the
searched study with a combination of fMRI and Machine Learning Algorithms or fMRI with deep learning algorithms. SQ1 is a search query that is
used for finding fMRI studies with Machine Learning Algorithms and SQ2 is a search query that is used for finding fMRI studies with deep learning
algorithms. The search queries were further improved with other synonyms used in existing research of fMRI which are represented in the form
of equations 1 and 2 as follows:

SQ1 = ½ð“Functional Magnetic Resonance Imaging” OR “fMRI” OR “fMRI Methods”Þ AND ð“Machine Learning Algorithms” OR “ML Algorithms”Þ ð1Þ

SQ2 = ½ð“Functional Magnetic Resonance Imaging” OR “fMRI” OR “fMRI Methods”Þ AND

ð2Þ
ð“Deep Learning Algorithms” OR “DL Algorithms” OR “Deep Learning Architectures” OR “DL Architectures”Þ

In SQ1, an attempt is made to search papers based on keywords of functional magnetic resonance imaging or fMRI or fMRI Methods with
Machine Learning Algorithms or ML Algorithms. The authors tried all combinations of these keywords based on logical “AND” between fMRI and
Machine Learning Algorithms. Similarly in SQ2, an attempt is made to search papers based on keywords of functional magnetic resonance imaging
or fMRI or fMRI Methods with deep learning Architectures or DL Architectures. The authors tried all combinations of these keywords based on
logical “AND” between fMRI and deep learning Architectures.
For every search in this systematic review, the authors used search queries based on equations 1 and 2 in popular scientific databases of Sci-
ence Direct, Scopus, Web of Science and IEEE Xplore. The total number of publications gathered for review was 360. Out of these, 127 publica-
tions were fetched from Science Direct, 130 publications from Scopus, 31 publications from Web of Science and 72 publications from IEEE
Xplore. These studies were validated for duplicate publications and later filtered with the help of the COEs already stated. The overall results
achieved are shown in Table 1.
The COE applied to the total number of publications initially gathered have reduced the number of studies from 360 to 241. Under the first
COE, the authors found three number of studies that did not clear the test of being peer-reviewed. 6 studies out of which five book chapters and
one short communication did not clear the test due to the absence of any experimental study. One study in the form of the graduate thesis is also
discarded due to the unavailability of test results related to fMRI methods. Under the second COE, one study has been excluded as the paper is
written in the Chinese language. Fifteen papers were excluded as they were containing only surveys or existing literature related to fMRI and
machine learning or deep learning. Further, under this criterion of exclusion, three papers were not accessible to authors despite seeking help
from other research labs and universities and thus were also excluded. Under the third COE, 83 studies were found with repeated content while
fetching from different databases and were therefore discarded. Next, the authors found 4 studies where the study was based on fMRI, however,
methods and algorithms required were not proposed and were thus discarded. Three papers were found where fMRI methods and machine learn-
ing or deep learning architectures were present, however, experimentation and validation were not conducted and were therefore discarded as
RASHID ET AL. 7 of 29

FIGURE 5 Flow diagram of PRISMA for studies included in systematic review

well. At last, 241 studies out of 360 were retained after application of all the criteria of exclusion as shown in flow diagram of PRISMA in Figure 5
The retained studies were deeply examined for various insights as given in Section 4 and were classified on the basis of types of fMRI with their
common methods, classification of fMRI studies on the basis of mental diseases, classification of fMRI studies on the basis of Machine Learning
Algorithms and finally classification of fMRI studies on the basis of deep learning algorithms. The results achieved from these systematic classifica-
tions are given in Section 4.

4 | SYSTEMATIC REVIEW RESULTS

The results in this study are classified based on four types of information related to fMRI: one that relates the various methods which apply to
fMRI in terms of its classification of resting-state fMRI and Task-based fMRI, second that covers the number of studies for various types of mental
diseases under fMRI along with their change in functional connectivity, third which relates to the use of Machine Learning Algorithms along with
its types and performance metrics to fMRI, and fourth which relates the use of deep learning architectures along with its types and performance
metrics to fMRI. While evaluating results, all papers used in this study were collected and then abstract, proposed technique description and con-
clusions were deeply examined. The insights gained from each paper were noted in worksheets of Microsoft Excel and later all gathered informa-
tion out of these papers were processed and organized in a presentable manner.
8 of 29 RASHID ET AL.

4.1 | Functional magnetic resonance imaging

fMRI technique can be classified into resting-state fMRI and task-based fMRI. Resting-state fMRI is used to evaluate the interaction among
regions of the brain in its resting state when the brain is not performing any task for some external subject (Buckner, Krienen, & Yeo, 2013). The
most common methods used for processing resting-state fMRI data are Seed based Functional Connectivity (SBFC), Regional Homogeneity
(ReHo), Amplitude of Low-frequency fluctuation (ALFF), Principal Components Analysis (PCA), and Independent Component Analysis (ICA). Task-
based fMRI is used to evaluate the time series data against neural functions for which some cognitive task is performed. A map is constructed
later for the activation of voxels that respond to the task shown in terms of subject (Chen & Glover, 2015). The most common methods which are
used for processing task-based fMRI are General Linear Model, psycho-physiological interaction, Dynamic Causal Modelling, Multivariate Pattern
Analysis, and Structural Equation Modelling. The classification of fMRI methods is shown in Figure 6.

4.1.1 | Resting state fMRI classification results

The number of studies related to resting-state fMRI that contain methods as given in Figure 6 is shown in Table 2. Each method provided in
Table 2 is supported by advantages and disadvantages based on the selected number of studies. From selected studies of resting-state fMRI, all
methods are used to characterize resting states in brain networks among patients. None out of all these methods is standard for examining pro-
cesses of resting-state and these methods are treated as complementary to each other. Analysis based on combining these methods is one oppor-
tunity that can provide the best characterization of resting-state brain connectivity. Results achieved in terms of the number of studies pointed a
significant number of studies available for the ReHo method when applied on resting-state fMRI. The results are shown in Figure 7 for resting-
state fMRI. It clearly confirms that ReHo is the method that is studied most for resting-state fMRI. The possible reason for the availability of more
research papers in ReHo for fMRI is its significance in terms of evaluating the similarity of synchronization for measuring brain activity among the
time series of voxel and is nearest neighbours.

4.1.2 | Task-based fMRI classification results

The number of studies related to task-based fMRI that contain methods as given in Figure 6 is shown in Table 3. Each method provided in Table 3
is supported by advantages and disadvantages based on the selected number of studies. From selected studies of task-based fMRI, it can be

F I G U R E 6 Classification of
fMRI methods
RASHID ET AL. 9 of 29

TABLE 2 Classification of studies based on resting state fMRI methods

Method
used Selected studies Advantages Disadvantages
SBFC Di Martino et al. (2011), Gandhi (2016), Grandjean, 1. Interpretation of results is easier. 1. The assumption is required to
Schroeter, Batata, and Rudin (2014), Joel, Caffo, Van Zijl, 2. Proves good in showing the select the seed area.
and Pekar (2011), Smith et al. (2014), Van Dijk, Sabuncu, strongly connected network of 2. Faces challenges while
and Buckner (2012), and Von dem Hagen, Stoyanova, regions with ROI. interpreting noise.
Baron-Cohen, and Calder (2012),
ReHo Chen et al. (2012), Dong, Huang, and Du (2012), Fang 1. No assumptions are required 1. Requires a high level of
et al. (2013), Gnanadas, Sathishbabu, and related to spatial independence sensitivity for spatial
Vijayakarthik (2017), Liang et al. (2011), Liu et al. (2010), of the map identified. smoothing.
Peng et al. (2011), Qin et al. (2017), Qing, Dong, Li, Zang, 2. Group Analysis is easy. 2. Difficult to draw inferences
and Liu (2015), Qiu et al. (2019), Shukla, Keehn, and between spatially remote
Müller (2010), Song, Zhang, and Liu (2014), Wang, Song, regions.
Jiang, Zhang, and Yu (2011), Wu et al. (2007), Wu
et al. (2009), Yang et al. (2018), Yuan et al. (2013), Zang,
Jiang, Lu, He, and Tian (2004), and Zeng, Pizarro, Nair, La,
and Prabhakaran (2013)
ALFF Di, Kim, Huang, Lin, and Biswal (2013), Dong, Guo, Zhang, 1. An effective method to measure 1. Sensitive for scaling of raw
Fu, and Shi (2010), Guo, Dong, Zhang, Zhang, and resting-state fMRI signals on signals.
Yin (2010), Hu et al. (2018), Khodaee, Hossein-Zadeh, and each voxel basis. 2. This method is prone to noise
Ananloo (2015), Liang, P. (2014), Liu et al. (2014), Meng, 2. The preferred method for from physiological sources,
Zhang, Fan, and Li (2018), Turner et al. (2012), Yan, Zhuo, measuring the level of neural especially near ventricles.
Wang, and Wang (2011), Yang et al. (2007), Yuan activity among functional regions.
et al. (2013), Yu-Feng et al. (2007), Zhang et al. (2015),
Zou et al. (2008), and Yu et al. (2014)
PCA Carbonell, Bellec, and Shmuel (2011), Du et al. (2012), 1. Performs better for multi-subject 1. Faces computational challenge
Hanson, Westlye, and Lundervold (2014), Leonardi datasets with data driven in high dimensional datasets.
et al. (2013), Wang, Jiao, and Lu (2011), Xie, Cao, and decompositions. 2. The impractical method in
Weng (2008), and Zhong et al. (2009), 2. The preferred method for cases where subjects increase
removing components carrying and require high memory.
noise in signals.
ICA Alizadeh, Fatemizadeh, and Deevband (2014), Binnewijzend 1. Supporting feature of avoiding 1. Requires Iterative
et al. (2012), Boubela et al. (2013), Calhoun, Kiehl, and spatial assumption. Optimization for
Pearlson (2008), De Luca, Beckmann, De Stefano, 2. A better method for comparing decomposition.
Matthews, and Smith (2006), Đuri
c et al. (2016), Fan, Yao, activity coherence in distributed 2. Requires estimation in
Wu, and Liu (2012), Feis et al. (2015), Gandhi (2016), voxels. components selected.
Griffanti et al. (2014), Kokkonen et al. (2009), Musso,
Brinkmeyer, Mobascher, Warbrick, and Winterer (2010),
Perlbarg et al. (2008), Pruim, Mennes, Buitelaar, and
Beckmann (2015), Pruim et al. (2015), and Rashid
et al. (2016)

Abbreviations: ALFF, amplitude of low-frequency fluctuation; ICA, Independent Component Analysis; PCA, Principal Components Analysis; ReHo, Regional
Homogeneity; SBFC, Seed based Functional Connectivity.

concluded that task-based fMRI methods are considered as a benchmark for identifying ROI or voxels based on external stimuli. However, the
studies also reveal that brain networks are involved on a large scale for various brain responses in response to visual stimulus. It becomes chal-
lenging to determine all these large functional networks in any form of task-based fMRI data due to the excessive costs and time-consuming
issues. These issues are still open to address while working with datasets of task-based fMRI. Results achieved in terms of the number of studies
pointed a significant number of studies available for the Multivariate Pattern Analysis (MVPA) method when applied on task-based fMRI. The
results are shown in Figure 8 for task-based fMRI. It clearly confirms that Multivariate Pattern Analysis (MVPA) is popular among neuroimaging
communities due to its comparison and analysis of distribution patterns inactivity by detecting differences between conditions which are having
higher sensitivity than that of univariate analysis.
The authors have plotted a timeline for the number of publications that are studied for types of fMRI in Figure 9. The plot clearly shows a
spike for both resting-state and task-based fMRI around the year 2014 which gives a clear indication about the popularity of this subject in recent
years.
10 of 29 RASHID ET AL.

F I G U R E 7 Number of studies for

resting state based fMRI methods

TABLE 3 Classification of studies based on task-based fMRI methods

Method
used Selected studies Advantages Disadvantages
GLM Johnstone et al. (2006), Kay, Rokem, Winawer, 1. This model supports hypothesis testing 1. This method is based on a
Dougherty, and Wandell (2013), Nakai, Bagarinao, for fitted parameters. single voxel.
Matsuo, Ohgami, and Kato (2006), Shahin, Shayegh, 2. Easy support of estimation of parameters. 2. Involves mass-univariate
Mortaheb, and Amirfattahi (2016), Soch, Haynes, and testing
Allefeld (2016), Wang, Childress, Wang, and
Detre (2007), Wang, Childress, and Detre (2006), Xu,
Potenza, and Calhoun (2013), and Zilverstand, Sorger,
Zimmermann, Kaas, and Goebel (2014),
PPI Boksman et al. (2005), Burnett and Blakemore (2009), 1. This method identifies regions in cases 1. This method suffers in
Bolstad et al. (2013), Osada et al. (2015), Lordier where activity depends on the interaction identifying similarities
et al. (2019), Harrison, McLaren, Moody, Feusner, and between task and time course of ROI. between physiological
Bookheimer (2017), and Tailby, Kowalczyk, and 2. PPI is helpful in identifying voxels where contrasts.
Jackson (2018). the activity is related to seed ROI for 2. No support in experiments
given stimuli. that require more number
of conditions.
SEM Chen et al. (2011), Karunanayaka et al. (2011), Kim, Zhu, 1. An efficient method for examining brain 1. No Distinction between
Chang, Bentler, and Ernst (2007), Marco, Fall, and connectivity disorders. hemodynamic and
Vrignaud (2007), Moore et al. (2018), and Zhuang, 2. A Good estimation of parameters in this neuronal levels.
Peltier, He, LaConte, and Hu (2008) model. 2. Applicable where samples
are independent
statistically.
DCM Sladky et al. (2013), Kasess et al. (2008), Grefkes 1. Interactions in this model take Structural 1. Better fitting models
et al. (2010), Bönstrup, Schulz, Feldheim, Hummel, Equation Modelling at the neuronal level. require high computations.
and Gerloff (2016), Seghier, Zeidman, Neufeld, Leff, 2. This method results in better fitting 2. Slow estimation of
and Price (2010), Mane, Nagori, and Agrawal (2012), models that are biologically realistic. parameters in this model.
Friston, Moran, and Seth (2013), Goulden
et al. (2012), Osório, Rosa, Silvestre, and
Figueiredo (2015), and Penny, Iglesias-Fuster, Quiroz,
Lopera, and Bobes (2018)
MVPA Bleich-Cohen et al. (2014), Brodersen et al. (2012), 1. Supports fMRI response patterns in a 1. This method requires a
Charest, Kriegeskorte, and Kay (2018), De Martino large number of voxels. priori knowledge about the
et al. (2008), Dosenbach et al. (2010), Gardumi 2. This method is having more sensitivity regions of interest to test.
et al. (2016), Hosseini and Kesler (2014), Mack, than univariate hypothesis testing. 2. Requires a large number of
Preston, and Love (2013), Wang, Li, and Gu (2017), features for multivariate
Yan, Yang, Wu, and Guo (2014), and Zhang statistics.
et al. (2012)

Abbreviations: DCM, Dynamic Causal Modelling; GLM, General Linear Model; MVPA, multivariate pattern analysis; PPI, psycho-physiological interaction;
SEM, Structural Equation Modelling.
RASHID ET AL. 11 of 29

F I G U R E 8 Number of studies for

task-based fMRI methods

FIGURE 9 Timeline of types of fMRI

studies

4.2 | Classification of fMRI studies on the basis of mental diseases

The amount of research studies where fMRI is used for the assessment of different mental diseases is shown in Table 4. The authors have done
this classification based on the fMRI method to be used subject to the condition of mental disease covered along with its functional connectivity.
The most common diseases where fMRI techniques and methods are applied include aging brain, Alzheimer's disease, autism, attention deficit
hyperactivity disorder, bipolar disorder, depression, epilepsy, mild cognitive impairment, obsessive compulsive disorder, Parkinson's disease, pain
disorder, social anxiety disorder, and Schizophrenia. Aging brain is prone to functional and structural changes in the brain and generates the neural
connectivity of brain activation. Functional MRI is one of the choices in the present research to measure this functional connectivity of the brain.
Alzheimer's disease is another chronic disease that degenerates neurons over a period of time. Functional MRI is currently acting as one of the
promising biomarkers in the application of Alzheimer's disease. Autism is a type of mental disorder that occurs due to the problems of communica-
tion and social interactions. Functional MRI techniques can greatly help researchers in discovering the changes in development in the brain for
monitoring disorder. Attention deficit hyperactivity disorder is a mental disease where individuals are prone to excessive activities which results
in inappropriate behaviour and it gets difficult for individuals to focus and seek attention. Functional MRI techniques are playing a vital role in the
diagnosis and treatment of attention deficit hyperactivity disorder patients. Bipolar Disorder is one of the forms of mental disease which occurs
due to the abnormal elevation in mood. Functional MRI techniques are used in research for studying bipolar disorders where cognitive tasks
related to emotions in the brain are examined. Depression is a kind of mental disease which occurs due to the presence of low mood in individuals
stretching across multiple weeks. Functional MRI networks are one of the favourable choices currently used for the classification of depression
disorders. Epilepsy is a mental disorder that occurs due to the recurrence in epileptic seizures which is a symptom of the abnormal neural activity
of the brain. Functional MRI plays a vital role in individuals with epilepsy for evaluating epileptic seizures and possible measures for its therapy.
12 of 29 RASHID ET AL.

TABLE 4 Classification of fMRI studies based on the basis of common mental diseases

Change in functional
Type of mental disease connectivity Selected studies
Aging brain Brain system Dennis and Thompson (2014) and Vergun et al. (2013)
disruption
Alzheimer's disease Decreased Barros and Silveira (2017), Basaia (2019), Binnewijzend et al. (2012), Brier et al. (2014), Challis
connectivity et al. (2015), Dennis and Thompson (2014), Harrison et al. (2017), Hojjati (2017), Hosseini-Asl,
Gimel'farb, & El-Baz, 2016, Hu, Ju, Shen, Zhou, and Li (2016), Khazaee, Ebrahimzadeh, and
Babajani-Feremi (2015), Khazaee, Ebrahimzadeh, and Babajani-Feremi (2016), Khazaee (2017),
Koch et al. (2012), Liang (2014), Liu, Liu, Cai, Che, et al. (2014), Liu, Liu, Cai, Pujol, et al. (2014),
Morra et al. (2009), Penny et al. (2018), Sarraf and Tofighi (2016), Shi, (2017), Suk (2017),
Suk, (2014, 2016), Wee (2016), and Yetkin, Rosenberg, Weiner, Purdy, and Cullum (2006)
Autism Altered connectivity Aghdam, Sharifi, and Pedram (2018), Alizadeh et al. (2014), Bi, Wang, Shu, Sun, and Xu (2018),
Chanel et al. (2016), Chen et al. (2017), Di Martino et al. (2011), Koshino et al. (2005), Shukla
et al. (2010), and Von dem Hagen et al. (2012)
Attention deficit Altered small Aron, Dowson, Sahakian, and Robbins (2003), Burgess et al. (2006), Bush et al. (2009), Tian
hyperactivity networks et al. (2006), and Wang, Jiao, and Lu (2011)
disorder
Bipolar disorder Abnormal Adler, Holland, Schmithorst, Tuchfarber, and Strakowski (2004), Argyelan et al. (2013), Arribas,
connectivity Calhoun, and Adali (2010), Deng et al. (2018), Gao et al. (2017), Li et al. (2017), Liang
et al. (2013), Rashid, Damaraju, Pearlson, and Calhoun (2014), Rubin-Falcone et al. (2018),
Yoshida et al. (2017), and Yurgelun-Todd et al. (2004)
Depression Abnormal Fan et al. (2012), Guo et al. (2011), Liang et al. (2013), Peng et al. (2011), and Sacchet, Prasad,
connectivity Foland-Ross, Thompson, and Gotlib (2015)
Epilepsy Decrease/increase in Gnanadas et al. (2017), Munsell et al. (2015), Tailby et al. (2018), Woermann et al. (2003), and
connectivity Zeng et al. (2013)
Mild cognitive Abnormal Binnewijzend et al. (2012), Challis et al. (2015), Deng et al. (2018), Gao et al. (2017), Johnston,
impairment connectivity Steele, Tolomeo, Christmas, and Matthews (2015), Johnston, Tolomeo, et al. (2015), Khazaee
et al. (2016), Li et al. (2017), Liang (2014), Rubin-Falcone et al. (2018), Sankar et al. (2016),
Wang, Ren, and Zhang (2017), Yetkin et al. (2006), and Yoshida et al. (2017)
Obsessive compulsive Increase/decrease in Endrass et al. (2010) and Meng et al. (2018)
disorder connectivity
Parkinson's disease Altered connectivity Su et al. (2015), Wu et al. (2009)
Pain disorder Altered connectivity Brodersen et al. (2012) and Sundermann et al. (2014)
Social anxiety disorder Disrupted Månsson et al. (2015), Sladky et al. (2013), and Tang, Jiang, Liao, Wang, and Luo (2013)
connectivity
Schizophrenia Disrupted networks Arbabshirani, Castro, and Calhoun (2014), Argyelan et al. (2013), Arribas et al. (2010),
Bleich-Cohen et al. (2014), Boksman et al. (2005), Chen et al. (2017), Damaraju et al. (2015),
Du et al. (2012), Harrison, Yücel, Pujol, and Pantelis (2007), Khodaee et al. (2015), Kim,
Calhoun, Shim, and Lee (2016), Lu et al. (2016), Matsubara, Tashiro, and Uehara (2019),
Qureshi, Oh, Cho, Jo, and Lee (2017), Rashid et al. (2014), Sui et al. (2014), Turner et al. (2012),
Yu et al. (2014), and Zhou et al. (2007)

Mild Cognitive Impairment is a mental disease that occurs in aged adults due to the barriers in cognitive processes without having ill effects on a
daily routine. Functional MRI techniques are popularly used for the identification and investigation of brain structures for the recognition of cogni-
tive impairments. Obsessive-compulsive disorder is a kind of mental disease where individuals are repeatedly performing certain routines and are
prone to obsessions where it is difficult to control the repetition of thoughts. Functional MRI acts as a popular tool for targeting brain regions with
obsessive-compulsive disorders. Parkinson's disease is a mental disorder that mainly affects the neural connections associated with muscle tis-
sues. Functional MRI is quite useful in individuals suffering from Parkinson's disease for studying the patterns of brain connectivity which are
abnormal. Pain Disorder is a kind of disorder that disables any individual from the proper functioning of the brain and is caused mainly by psycho-
logical stress. Functional MRI is considered one of the reliable tool for studying the areas which often get affected by pain disorder. Social anxiety
disorder is one of the most common mental disorder which acts due to the social phobia where individuals under subject feel high anxiety and get
fear of any rejections or being judged negatively. Functional MRI is quite useful in investigating individuals with problems of Social Anxiety Disor-
ders. Schizophrenia is a mental disease that occurs in individuals whose behaviour remains abnormal and has less ability for understanding. There
are several studies that are carried out on this disease involving fMRI where an individual's brain system is examined to study the abnormal behav-
iours and connectivity in brain regions.
RASHID ET AL. 13 of 29

Results achieved in terms of the number of studies including fMRI methods or techniques with mental diseases pointed out a significant num-
ber of studies available for Alzheimer's disease followed by Schizophrenia. The results are shown in Figure 10. The figure confirms the most num-
ber of research studies done on Alzheimer's disease when using fMRI methods or techniques. The possible reason for the availability of more use
of fMRI in Alzheimer's disease is its means to investigate and detect changes in brain functions at earlier stages in terms of symptoms before the
development of any structural damage of the brain. Moreover, Functional MRI has shown great promise in terms of visualization for modulating
specific circuits using therapeutic interventions.

4.3 | Machine Learning Algorithms

Machine Learning algorithms belonging to supervised, unsupervised, semi-supervised, and reinforcement learning are classified into types with
common machine learning approach subject to the use of fMRI methods. The authors grouped the number of studies having a common machine
learning algorithm with the use of fMRI methods as per the search query SQ1 given in Section 3. The classification is supported by advantages
and disadvantages of each machine learning algorithm with the use of fMRI methods based on studies considered in this systematic review. While
making this classification based on Machine Learning Algorithms, the authors found several papers where some changes were done in core
Machine Learning Algorithms. For avoiding the complexity, the authors have grouped them in the same category while making a classification.
Also, some studies were found in literature where a combination of algorithms is used along with fMRI. The authors have grouped such studies
under the “Ensemble” category. In the machine learning field, an ensemble method is implemented in ways of random forest, bagging and boo-
sting, however, the authors have not considered the difference in this systematic review. The results achieved while classifying fMRI studies on
the Machine Learning Algorithms are shown in Table 5. Results achieved in terms of the number of studies including fMRI methods or techniques
common machine learning algorithm pointed out a significant number of studies available for support vector machine followed by Ensemble. The
possible reason for the availability of more use of support vector machine in fMRI is due to its higher classification accuracy even in noisy
datasets. Another prime reason which resulted from studies is the ability of support vector machine classifier to perform better even in cases of
generalization with high dimensions in terms of data and a small number of patterns in the training phase. The overall results for each machine
learning classifier used along with the fMRI method are shown in Figure 11.
Figure 11 suggests that K means and neural networks despite being popular Machine Learning Algorithms are having the least number of
studies available in fMRI research. However, at the same time, it opens a window for researchers to target these algorithms along with fMRI
methods in the future.

4.3.1 | Studies with best performance of Machine Learning Algorithms

This section of systematic review provides the studies with the best performance of Machine Learning Algorithms used in fMRI in terms of vari-
ous studies. The authors believe that studies with best performances will help incoming researchers in their future works. The performance of
Machine Learning Algorithms in these studies is evaluated based on accuracy. The classification is done in terms of Table 6. Machine learning

F I G U R E 1 0 Number of fMRI studies

for various mental diseases
14 of 29 RASHID ET AL.

TABLE 5 Classification of fMRI studies based on Machine Learning Algorithms

Type of ML
algorithm Selected studies Advantages Disadvantages
Support vector Altmann et al. (2016), Arbabshirani et al. (2014), 1. A multivariate method for 1. For noisy datasets, SVM does
machine Bahrami and Shamsi (2017), Bi et al. (2018), providing efficient prediction not yield good results.
Brodersen et al. (2012), Chanel et al. (2016), Chen of brain responses in fMRI 2. This method must be avoided in
et al. (2017), Cox and Savoy (2003), De Martino data. problems where datasets are not
et al. (2007, 2008), Deng et al. (2018), Dosenbach 2. Performs better when the large.
et al. (2010) Gao et al. (2017), Garner et al. (2019), separation between the
Guo, Xie, Cheng, and Zhao (2009), Ji, Liu, Wang, and classes is not ambiguous.
Tang (2004), Johnston, Steele, et al. (2015),
Johnston, Tolomeo, et al. (2015), Khazaee A (2017),
LaConte, Strother, Cherkassky, Anderson, and
Hu (2005), Li et al. (2017), Linn et al. (2016), Lorbert
and Ramadge (2013), Lu et al. (2016), Månsson
et al. (2015), Matsubara et al. (2019), Meier
et al. (2012), Misaki, Luh, and Bandettini (2013),
Mourao-Miranda, Bokde, Born, Hampel, and
Stetter (2005), Peltier, Lisinski, Noll, and
LaConte (2009), Qureshi et al. (2017), Rubin-Falcone
et al. (2018), Sacchet et al. (2015), Sankar
et al. (2016), Schrouff, Kussé, Wehenkel, Maquet,
and Phillips (2012), Tan et al. (2015), Tang
et al. (2013), Vergun et al. (2013), Wang et al. (2019),
Wang, Ren, and Zhang (2017), Wang et al. (2007),
Wang et al. (2009), Wetherill et al. (2019), Wu
et al. (2015), and Zafar, Malik, Shuaibu, ur Rehman,
and Dass (2017)
Ensemble Cabral, Silveira, and Figueiredo (2012), Chyzhyk, Savio Ensemble classifiers provide The complexity of computations in
and Graña (2015), Genuer, Michel, Eger, and better results than single ensemble classifiers for fMRI
Thirion (2010), Kesler et al. (2017), Plumpton, classifier in individual voxel data is higher than individual
Kuncheva, Linden, and Johnston (2010), Sui selection methods. classifiers.
et al. (2014), Yuan, Mei, Huang, & Yang (2014),
Zhuang et al. (2018) and Zhuang et al. (2018)
Logistic Barros and Silveira (2017), Koike, Hatakeyama, and This classifier is computationally The accuracy of predictions is
regression Yoshida (2014), Lindquist and McKeague (2009), efficient in the process of limited due to a large number of
Ryali and Menon (2009), Ryali, Supekar, Abrams, and identifying brain regions in features in comparison to several
Menon (2010), Turek, Willke, Chen, and fMRI data. observations.
Ramadge (2017), Wang, Zhang, Qian, Wang, and
Davidson (2014), Yoshida et al. (2017), and Zhang
et al. (2017)
Naïve Bayes Do, Yang, Kim, Lee, and Kim (2015), Huang, Mei, and Better classifier for smoothing This classifier works on the
Wang (2018), Khazaee, A (2017), Parida, Dehuri, and fMRI images in the spatial assumption of independence in
Cho (2014), Seo et al. (2015), Tahmassebi et domain. attributes of fMRI data.
al. (2018), and Yargholi and Hossein-Zadeh (2016),
J48 decision Alamdari and Fatemizadeh (2013), Douglas, Harris, This classifier efficiently This classifier is having
tree/C4.5 Yuille, and Cohen (2011), Richiardi, Eryilmaz, searches a subset of voxels in computational complexity and
Schwartz, Vuilleumier, and Van De Ville (2010), and fMRI data to maximize the takes more time.
Tahmassebi et al. (2018) gap in classes.
AdaBoost Alamdari and Fatemizadeh (2013), Martínez-Ramón, This classifier is having high This classifier provides poor results
Koltchinskii, Heileman, and Posse (2006), and Morra computational speed and in noisy fMRI datasets.
et al. (2009) suits real-time fMRI.
k nearest Michel, Damon, and Thirion (2008), Onal, Ozay, Firat, This classifier provides better This classifier is not suitable for
neighbours Öztekin, and Vural (2013), and Yang and results in the segmentation of large fMRI datasets.
Shahabi (2007) ROI in fMRI data.
Gaussian Challis et al. (2015), Marquand et al. (2010), and This classifier is suited in models Efficiency in this model suffers
processes Schrouff et al. (2012) for predictions of variables from fMRI data of high
that are continuous. dimensional spaces.
RASHID ET AL. 15 of 29

TABLE 5 (Continued)

Type of ML
algorithm Selected studies Advantages Disadvantages
K means Thirion, Varoquaux, Dohmatob, and Poline (2014) and This classifier performs better in This classifier fails to fit data in a
Vijay and Selvakumar (2015) fMRI data where several balanced way in brain
parcels are low. parcellations across fMRI
datasets.
Neural network Huynh and Won (2008) and Qureshi et al. (2017) Efficient classifier for extracting This classifier is expensive in terms
functional connectivity in ROI of computational costs for
of fMRI data. processing fMRI data.

F I G U R E 1 1 Number of fMRI studies

based on Machine Learning Algorithms

studies with highest accuracies for health controls versus Schizophrenia, health controls versus bipolar disorder, health controls versus Alzheimer's
disease, health controls versus major depressive disorder, health controls versus autism spectrum disorder, bipolar disorder versus major depres-
sive disorder with their sample sizes and methods are given in Table 6. It can be seen from Table 6 that the highest accuracy achieved is 99.3% by
(Qureshi et al., 2017) when using extreme learning machine. Other studies achieving the highest accuracies are most using support vector
machine classifier with leave one out cross validation. All studies with their accuracies are shown in terms of Figure 12.

4.4 | Deep learning algorithms

Our studies revealed that most common deep learning architectures used for fMRI research are CNN, Deep Belief Network (DBN), Deep Bayesian
Network (DBaN), Deep Autoencoder (DAE), Deep Boltzman Machine (DBM), Deep Weighted Subclass-based sparse multi-task learning (DW-S2
MTL), Deep Multilayer Perceptron (DMP) and Stacked Autoencoder (SAE). The authors grouped the number of studies having common deep learn-
ing architecture with the use of fMRI methods as per the search query SQ2 given in Section 3. The classification is supported by the advantages and
disadvantages of each deep learning architecture with the use of fMRI methods based on studies considered in this systematic review. The results
achieved while classifying fMRI studies on the deep learning architectures are shown in Table 7. Results achieved in terms of several studies includ-
ing fMRI methods or techniques with common deep learning architecture pointed out a significant number of studies available for CNN followed by
Stacked Autoencoder. The possible reason for the availability of more use of CNN in fMRI is its ability to automatically extract the required features
for fMRI data processing. The overall results for each deep learning architecture used along with the fMRI method are shown in Figure 13.

4.4.1 | Studies with best performance of Deep Learning Algorithms

This section of systematic review provides the performance of deep learning algorithms used in fMRI in terms of various studies that will help
incoming researchers in their future works. The performance of deep learning algorithms in these studies is evaluated based on accuracy aimed to
16 of 29 RASHID ET AL.

TABLE 6 Machine learning studies with highest accuracies with subjects and methods used

Name of study Subjects & sample size Machine learning method/classifier Accuracy achieved
Qureshi et al. (2017) SCZ = 72, HC = 72 ELM 99.3%
Johnston, Tolomeo, et al. (2015) HC = 21, MDD = 19 SVM with leave one out cross validation 97.00%
Wang, Ren, and Zhang (2017) HC = 29, MDD = 31 SVM with leave one out cross validation 95.00%
Khazaee, A (2017) HC = 45, AD = 34 Naïve bayes 93.29 %
Gao et al. (2017) BD = 37, MDD = 37 SVM with 10-fold cross validation 93.00%
Matsubara et al. (2019) SCZ = 54, HC = 67 SVM with 10-fold cross validation 92.00%
Arbabshirani et al. (2014) SCZ = 195, HC = 175 Functional connectivity and SVM 88.21%
Li et al. (2017) BD = 22, MDD = 22 SVM with leave one out cross validation 86.00%
Johnston, Steele, et al. (2015) HC = 21, MDD = 25 SVM with leave one out cross validation 85.00%
Chen et al. (2017) HC = 31, ASD = 22 SVM 83.33%
Yoshida et al. (2017) HC = 65, MDD = 58 PLSR with leave one out cross validation 80.00%
Rubin-Falcone et al. (2018) BD = 26, MDD = 26 SVM with leave-two out cross validation 75.00%
Khazaee A (2017) 45 HC, AD = 34 SVM with radial base function 71.95%
Sankar et al. (2016) HC = 20, MDD = 23 SVM with 5-fold cross validation 70.00%
Deng et al. (2018) BD = 31, MDD = 36 SVM with leave one out cross validation 68.30%

Abbreviations: AD, Alzheimer's disease; ASD, autism spectrum disorder; BD, bipolar disorder; ELM, extreme learning machine; HC, health controls; MDD,
major depressive disorder; SCZ, Schizophrenia; SVM, support vector machine.

F I G U R E 1 2 Machine learning studies

with highest accuracies with subjects

classify healthy control based patients. Most of the studies on deep learning architectures were found on Alzheimer's disease Neuroimaging Initia-
tive dataset carried out on Alzheimer's disease and its related stage of mild cognitive impairment. The authors have taken initiative to group the
studies found in this systematic review with the highest accuracies. The classification is done in terms of Tables 8–10. Deep learning studies with
the highest accuracies for health controls versus Alzheimer's disease are given in Table 8. In Table 8, all studies are carried out for Alzheimer's dis-
ease Neuroimaging Initiative dataset with the use of different deep learning architectures. It can be seen from Table 8 that the highest accuracy
achieved is 99.2% by (Basaia, 2019) when using CNN. Table 8 reveals that deep learning architectures with the highest accuracies are CNN, Sta-
cked Autoencoder Deep Weighted Subclass-based sparse multi-task learning, DBM, Sparse Auto-encoder with CNN, DMP and Multiple Sparse
Regression with CNN. All studies with the highest accuracy are shown in terms of Figure 14.
Deep learning studies with the highest accuracies for healthy controls versus mild cognitive impairment are given in Table 9. In Table 9, all
studies are carried out for Alzheimer's disease Neuroimaging Initiative dataset with the use of different deep learning architectures. It can be seen
from Table 9 that the highest accuracy achieved is 92.1% by (Hosseini-Asl et al., 2016) when using CNN. Table 9 reveals that deep learning
RASHID ET AL. 17 of 29

TABLE 7 Classification of fMRI studies based on deep learning architectures

Type of DL
architecture Selected studies Advantages Disadvantages
CNN Basaia (2019), Gao and Hui (2016), Gupta, Ayhan, This architecture is useful in Fails to test the presence of redundant
and Maida (2013), Zhao et al. (2017), Hosseini- fMRI data processing and features while performing feature
Asl et al. (2016), Huang et al. (2017), Khosla, extracting valid features extraction.
Jamison, Kuceyeski, and Sabuncu (2019), Mao automatically.
et al. (2019), Sarraf and Tofighi (2016),
Suk (2017), and Zafar et al. (2017)
DBN Aghdam et al. (2018), Jang, Plis, Calhoun, and Efficient architecture for This architecture takes more time in
Lee (2017), Kuang and He (2014), Plis parameter reduction and calculations of fMRI feature extractions
et al. (2018), and Zhang et al. (2019) minimizes the degree of over-
fitting.
DBaN Hao, He, and Yin (2015) This architecture provides an This architecture is computationally more
efficient approach to handle expensive.
uncertainties in fMRI data.
DAE Chen et al. (2016), Huang et al. (2017), and Suk, This architecture learns data This architecture loses its power once the
Lee, et al. (2016) and Suk, Wee, Lee, and efficiently with proper image complexity in fMRI data
Shen (2016) filtration of noise in signals. increases.
DBM Suk (2014) This architecture is useful in The main challenge in this architecture is
data where there is an to examine the functional relationship
increase in computational that is existing between different brain
capacity. regions.
DW-S2 Suk, Lee, et al. (2016) and Suk, Wee, et al. (2016) Efficient architecture for This architecture is computationally more
MTL discarding non-informative expensive.
features recursively in fMRI
dataset.
DMP Li et al. (2014) Efficient architecture for This architecture can lead to under-fitting
classification in high or over-fitting due to the varying use of
dimensional spaces of fMRI hidden neurons by the user.
data.
SAE Hu et al. (2016), Kim et al. (2016), Liu, Liu, Cai, Che, This architecture improves This architecture is computationally more
et al. (2014), Liu, Liu, Cai, Pujol, et al. (2014), performance in fMRI data by expensive.
Munsell et al. (2015), Shi (2017), Suk and providing promising feature
Shen (2013), and Suk (2015) information.

Abbreviation: CNN, convolutional neural network; DAE, deep autoencoder; DBaN, deep Bayesian network; DBM, deep Boltzman machine; DBN, deep
belief network; DMP, deep multilayer perceptron; DW-S2 MTL, deep weighted subclass-based sparse multi-task learning; SAE, Stacked Autoencoder.

architectures with the highest accuracies are CNN, DMP, Stacked Autoencoder, DBM, Deep Weighted Subclass-based sparse multi-task learning
and Multiple Sparse Regression with CNN, Multiple Sparse Regression with CNN and Deep Autoencoder. All studies with the highest accuracy
are shown in terms of Figure 15.
Deep learning studies with the highest accuracies for in Alzheimer's disease versus mild cognitive impairment are given in Table 10. In
Table 10, all studies are carried out for Alzheimer's disease Neuroimaging Initiative dataset with the use of different deep learning architectures. It
can be seen from Table 10 that the highest accuracy achieved is 95% by (Chen et al., 2016) when using Stacked Autoencoder. Table 10 reveals
that deep learning architectures with the highest accuracies are Stacked Autoencoder, CNN, and DMP. All studies with the highest accuracy are
shown in terms of Figure 16.

5 | OPEN ISSUES

Based on our systematic review, there are some of the challenges in fMRI analysis with the use of machine learning and deep learning algorithms.
In this section, we are introducing them.

1. For single-subject based resting-state fMRI (rs-fMRI), it is difficult is to make predictions due to the variable ways of representing features
(Qiu et al., 2019). From existing literature, there is not even a single presence of standard atlas that will help in the extraction of time series in
18 of 29 RASHID ET AL.

F I G U R E 1 3 Number of fMRI studies

based on deep learning architectures

TABLE 8 Deep learning studies with highest accuracies in healthy controls versus Alzheimer's disease

Name of study Dataset used Deep learning architecture Accuracy achieved

Basaia (2019) ADNI Convolutional neural network 99.2
Suk (2015) ADNI Stacked Autoencoder 98.8
Hosseini-Asl et al. (2016) ADNI Convolutional neural network 97.6
Sarraf and Tofighi (2016) ADNI Convolutional neural network 96.9
Suk and Shen (2013) ADNI Stacked Autoencoder 95.9
Suk (2015) ADNI Deep weighted subclass-based sparse multi-task learning 95.1
Suk (2014) ADNI Deep Boltzman Machine 94.9
Gupta et al. (2013) ADNI Sparse auto-encoder with convolutional neural network 94.7
Li et al. (2014) ADNI Deep multilayer perceptron 91.4
Suk (2017) ADNI Multiple sparse regression with convolutional neural network 91.02
Liu, Liu, Cai, Che, et al. (2014) ADNI Stacked Autoencoder 87.8

F I G U R E 1 4 Deep learning studies

with highest accuracies in healthy
controls versus Alzheimer's disease

such data. Therefore, it is one of the important objectives to address this challenge for the easy exploration of such data with advanced
machine learning models that take a lot of time in training models.
2. Different methods are discussed in this systematic review under subsections 4.1.1 and 4.1.2 for analysing fMRI images. However, most of
these methods suffer from computational bottlenecks (Hanson et al., 2014). It is a challenge to reduce the complexity of these fMRI based
algorithms with the use of efficient numerical based methods and mathematical transformations.
RASHID ET AL. 19 of 29

TABLE 9 Deep learning studies with highest accuracies in healthy controls versus mild cognitive impairment

Name of Study Dataset used Deep learning architecture Accuracy achieved

Hosseini-Asl et al. (2016) ADNI Convolutional neural network 92.1
Li et al. (2014) ADNI Deep multilayer perceptron 90.7
Hu et al. (2016) ADNI Stacked Autoencoder 87.5
Gupta et al. (2013) ADNI Convolutional neural network 86.4
Suk and Shen (2013) ADNI Stacked Autoencoder 85
Suk (2014) ADNI Deep Boltzman machine 80.6
Suk (2015) ADNI Deep weighted subclass-based sparse multi-task learning 80.1
Suk (2017) ADNI Multiple sparse regression with convolutional neural network 73.2
Suk, Lee, et al. (2016) and Suk, ADNI Deep autoencoder 72.6
Wee, et al. (2016)

F I G U R E 1 5 Deep learning studies

with highest accuracies in healthy
controls versus mild cognitive impairment

T A B L E 1 0 Deep learning studies

Name of study Dataset used Deep learning architecture Accuracy achieved
with highest accuracies in Alzheimer's
disease versus mild cognitive impairment Chen et al. (2016) ADNI Stacked Autoencoder 95.0
Gupta et al. (2013) ADNI Convolutional neural network 88.1
Hosseini-Asl et al. (2016) ADNI Convolutional neural network 86.8
Basaia (2019) ADNI Convolutional neural network 85.9
Li et al. (2014) ADNI Deep multilayer perceptron 83.7

3. In fMRI based studies, Machine Learning Algorithms are used for training models and evaluating the performance of the model with accuracies.
However, in such models more emphasis is required while training such models as several variables like head motion can affect these models
and induce biases while estimating their functional connectivity (Garner et al., 2019).
4. Based on our systematic review, there are a good number of studies available where machine learning and deep learning algorithms are used
in fMRI methods. However, these algorithms not always prove efficient in terms of results (Hosseini-Asl et al., 2016). Using a small number of
data samples will lead to the over-fitting of these models and thus predicted results are quite far from real ones. Therefore it is a challenge to
address this issue by coming up with new approaches of regularization strategies to prevent over-fitting in these models.
20 of 29 RASHID ET AL.

F I G U R E 1 6 Deep learning studies

with highest accuracies in Alzheimer's
disease versus mild cognitive impairment

5. In neuroimaging studies based on brain disorder, there is a challenge to handle the problem of dimensionality (Gupta et al., 2013; Hosseini-Asl
et al., 2016). The inclusion of transfer learning, where a sample of large images is used for learning previously learned features, can be a good
advantage.
6. In the analysis of fMRI studies, the use of augmentation technique is an option for addressing the issue of a smaller sample of data. This tech-
nique of computer vision will apply several transformations to the data in terms of scaling, rotation, etc. and thus increase the sample size. This
will eventually result in lesser pre-processing times and help in terms of achieving better computational efficiency of models.

6 | CONCLUSION AND FUTURE WORK

This systematic review covers studies in terms of fMRI types and the usage and application of various machine and deep learning algorithms for
analysing mental diseases. An attempt is made to provide a sequential way of understanding the various trends in these algorithms which are used
in research and development of fMRI along with their advantages and disadvantages. Results were drawn out of various studies to identify
machine learning and deep learning research trends for the implementation of fMRI and to provide studies with better performance in these
algorithms.
In this systematic review, 360 studies were collected from different knowledge databases and were screened for the eligibility of inclusion in
this paper based on various COE given in Section 3. 241 number of studies were finally selected for consideration. The study reveals that a signif-
icant number of studies are available for support vector machine followed by Ensemble in terms of Machine Learning Algorithms with fMRI and a
significant number of studies are available for CNN followed by Stacked Autoencoder in terms of deep learning with Functional fMRI. The results
further reveal that Alzheimer's disease and Schizophrenia are the mental diseases where the highest number of studies are available for fMRI. The
results are further supported by evaluating the best studies where machine learning and deep learning are used in achieving the highest accura-
cies. Finally, this systematic review is ended with open issues to help future researchers to work on these challenges.
As part of future work, the authors expect a number of studies on fMRI to use deep learning architectures due to their ability to automatically
extract the required features for fMRI data processing and thus addressing the major challenge in fMRI high dimensional datasets where thou-
sands of feature exist on single instances. Moreover, the authors believe that the number of machine and deep learning algorithms must be used
for analysing other kinds of mental diseases with high spatial and temporal resolutions.

CONF LICT OF IN TE RE ST
The authors have no conflict of interest to declare.

ORCID
Mamoon Rashid https://orcid.org/0000-0002-8302-4571

RE FE R ENC E S
Adler, C. M., Holland, S. K., Schmithorst, V., Tuchfarber, M. J., & Strakowski, S. M. (2004). Changes in neuronal activation in patients with bipolar disorder
during performance of a working memory task. Bipolar Disorders, 6(6), 540–549.
Aghdam, M. A., Sharifi, A., & Pedram, M. M. (2018). Combination of rs-fMRI and sMRI data to discriminate autism spectrum disorders in young children
using deep belief network. Journal of Digital Imaging, 31(6), 895–903.
RASHID ET AL. 21 of 29

Alamdari, N. T., & Fatemizadeh, E. (2013, September). Comparison of classification and dimensionality reduction methods used in fMRI decoding. Paper
presented at 2013 8th Iranian Conference on Machine Vision and Image Processing (MVIP), IEEE. pp. 175–179.
Alizadeh, A., Fatemizadeh, E., & Deevband, M. R. (2014, November). Investigation of Brain Default Network's activation in autism spectrum disorders using
Group Independent Component Analysis. In 2014 21th Iranian Conference on Biomedical Engineering (ICBME), IEEE. pp. 177–180.
Altmann, A., Schröter, M. S., Spoormaker, V. I., Kiem, S. A., Jordan, D., Ilg, R., … Sämann, P. G. (2016). Validation of non-REM sleep stage decoding from
resting state fMRI using linear support vector machines. NeuroImage, 125, 544–555.
Amaro, E., Jr., & Barker, G. J. (2006). Study design in fMRI: Basic principles. Brain and Cognition, 60(3), 220–232.
Arbabshirani, M. R., Castro, E., & Calhoun, V. D. (2014, August). Accurate classification of schizophrenia patients based on novel resting-state fMRI fea-
tures. Paper presented at 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, IEEE. pp. 6691–6694.
Argyelan, M., Ikuta, T., DeRosse, P., Braga, R. J., Burdick, K. E., John, M., & Szeszko, P. R. (2013). Resting-state fMRI connectivity impairment in schizophre-
nia and bipolar disorder. Schizophrenia Bulletin, 40(1), 100–110.
Aron, A. R., Dowson, J. H., Sahakian, B. J., & Robbins, T. W. (2003). Methylphenidate improves response inhibition in adults with attention-defi-
cit/hyperactivity disorder. Biological Psychiatry, 54(12), 1465–1468.
Arribas, J. I., Calhoun, V. D., & Adali, T. (2010). Automatic Bayesian classification of healthy controls, bipolar disorder, and schizophrenia using intrinsic con-
nectivity maps from FMRI data. IEEE Transactions on Biomedical Engineering, 57(12), 2850–2860.
Atenas, T. L., Díaz, E. C., Quiroga, J. C., Arancibia, S. U., & Rodríguez, C. C. (2018). Functional magnetic resonance imaging: Basic principles and application
in the neurosciences. Radiología, 60(5), 368–377.
Bahrami, S., & Shamsi, M. (2017). A non-parametric approach for the activation detection of block design fMRI simulated data using self-organizing maps
and support vector machine. Journal of Medical Signals and Sensors, 7(3), 153.
Barros, H., & Silveira, M. (2017, July). Atlas based sparse logistic regression for Alzheimer's disease classification. Paper presente at 2017 39th Annual Inter-
national Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), IEEE. pp. 501–504.
Basaia, S., Agosta, F., Wagner, L., Canu, E., Magnani, G., Santangelo, R., & Alzheimer's Disease Neuroimaging Initiative. (2019). Automated classification of
Alzheimer's disease and mild cognitive impairment using a single MRI and deep neural networks. NeuroImage: Clinical, 21, 101645.
Bengio, Y. (2009). Learning deep architectures for AI. Foundations and Trends in Machine Learning, 2(1), 1–127.
Bi, X. A., Wang, Y., Shu, Q., Sun, Q., & Xu, Q. (2018). Classification of autism spectrum disorder using random support vector machine cluster. Frontiers in
Genetics, 9, 18.
Binnewijzend, M. A., Schoonheim, M. M., Sanz-Arigita, E., Wink, A. M., van der Flier, W. M., Tolboom, N., & Barkhof, F. (2012). Resting-state fMRI changes
in Alzheimer's disease and mild cognitive impairment. Neurobiology of Aging, 33(9), 2018–2028.
Bleich-Cohen, M., Jamshy, S., Sharon, H., Weizman, R., Intrator, N., Poyurovsky, M., & Hendler, T. (2014). Machine learning fMRI classifier delineates sub-
groups of schizophrenia patients. Schizophrenia Research, 160(1-3), 196–200.
Boksman, K., Théberge, J., Williamson, P., Drost, D. J., Malla, A., Densmore, M., & Neufeld, R. W. (2005). A 4.0-T fMRI study of brain connectivity during
word fluency in first-episode schizophrenia. Schizophrenia Research, 75(2-3), 247–263.
Bolstad, I., Andreassen, O. A., Reckless, G. E., Sigvartsen, N. P., Server, A., & Jensen, J. (2013). Aversive event anticipation affects connectivity between the
ventral striatum and the orbitofrontal cortex in an fMRI avoidance task. PLoS One, 8(6), e68494.
Bönstrup, M., Schulz, R., Feldheim, J., Hummel, F. C., & Gerloff, C. (2016). Dynamic causal modelling of EEG and fMRI to characterize network architectures
in a simple motor task. NeuroImage, 124, 498–508.
Boubela, R. N., Kalcher, K., Huf, W., Kronnerwetter, C., Filzmoser, P., & Moser, E. (2013). Beyond noise: Using temporal ICA to extract meaningful informa-
tion from high-frequency fMRI signal fluctuations during rest. Frontiers in Human Neuroscience, 7, 168.
Brier, M. R., Thomas, J. B., Fagan, A. M., Hassenstab, J., Holtzman, D. M., Benzinger, T. L., & Ances, B. M. (2014). Functional connectivity and graph theory
in preclinical Alzheimer's disease. Neurobiology of Aging, 35(4), 757–768.
Brodersen, K. H., Wiech, K., Lomakina, E. I., Lin, C. S., Buhmann, J. M., Bingel, U., … Tracey, I. (2012). Decoding the perception of pain from fMRI using mul-
tivariate pattern analysis. NeuroImage, 63(3), 1162–1170.
Buckner, R. L., Krienen, F. M., & Yeo, B. T. (2013). Opportunities and limitations of intrinsic functional connectivity MRI. Nature Neuroscience, 16(7), 832.
Burgess, G. C., Depue, B. E., Ruzic, L., Willcutt, E. G., Du, Y. P., & Banich, M. T. (2010). Attentional control activation relates to working memory in
attention-deficit/hyperactivity disorder. Biological Psychiatry, 67(7), 632–640.
Burnett, S., & Blakemore, S. J. (2009). Functional connectivity during a social emotion task in adolescents and in adults. european. Journal of Neuroscience,
29(6), 1294–1301.
Bush, G., Frazier, J. A., Rauch, S. L., Seidman, L. J., Whalen, P. J., Jenike, M. A., & Biederman, J. (1999). Anterior cingulate cortex dysfunction in attention-
deficit/hyperactivity disorder revealed by fMRI and the Counting Stroop. Biological Psychiatry, 45(12), 1542–1552.
Calhoun, V. D., Kiehl, K. A., & Pearlson, G. D. (2008). Modulation of temporally coherent brain networks estimated using ICA at rest and during cognitive
tasks. Human Brain Mapping, 29(7), 828–838.
Carbonell, F., Bellec, P., & Shmuel, A. (2011). Global and system-specific resting-state fMRI fluctuations are uncorrelated: Principal component analysis
reveals anti-correlated networks. Brain Connectivity, 1(6), 496–510.
Cabral, C., Silveira, M., & Figueiredo, P. (2012). Decoding visual brain states from fMRI using an ensemble of classifiers. Pattern Recognition, 45(6),
2064–2074.
Chyzhyk, D., Savio, A., & Graña, M. (2015). Computer aided diagnosis of schizophrenia on resting state fMRI data by ensembles of ELM. Neural Networks,
68, 23–33.
Challis, E., Hurley, P., Serra, L., Bozzali, M., Oliver, S., & Cercignani, M. (2015). Gaussian process classification of Alzheimer's disease and mild cognitive
impairment from resting-state fMRI. NeuroImage, 112, 232–243.
Chanel, G., Pichon, S. J. A., Conty, L., Berthoz, S., Chevalier, C., & Grèzes, J. (2016). Classification of autistic individuals by merging information from multi-
ple fMRI experiments.
Charest, I., Kriegeskorte, N., & Kay, K. N. (2018). GLM denoise improves multivariate pattern analysis of fMRI data. NeuroImage, 183, 606–616.
Chen, G., Glen, D. R., Saad, Z. S., Hamilton, J. P., Thomason, M. E., Gotlib, I. H., & Cox, R. W. (2011). Vector autoregression, structural equation modeling,
and their synthesis in neuroimaging data analysis. Computers in Biology and Medicine, 41(12), 1142–1155.
22 of 29 RASHID ET AL.

Chen, H., Uddin, L. Q., Duan, X., Zheng, J., Long, Z., Zhang, Y., & Chen, H. (2017). Shared atypical default mode and salience network functional connectivity
between autism and schizophrenia. Autism Research, 10(11), 1776–1786.
Chen, H. J., Zhu, X. Q., Yang, M., Liu, B., Zhang, Y., Wang, Y., & Teng, G. J. (2012). Changes in the regional homogeneity of resting-state brain activity in
minimal hepatic encephalopathy. Neuroscience Letters, 507(1), 5–9.
Chen, J. E., & Glover, G. H. (2015). Functional magnetic resonance imaging methods. Neuropsychology Review, 25(3), 289–313.
Chen, P. H., Zhu, X., Zhang, H., Turek, J. S., Chen, J., Willke, T. L. & Ramadge, P. J. (2016). A convolutional autoencoder for multi-subject fMRI data aggrega-
tion. arXiv preprint arXiv:1608.04846.
Cox, D. D., & Savoy, R. L. (2003). Functional magnetic resonance imaging (fMRI) “brain reading”: Detecting and classifying distributed patterns of fMRI
activity in human visual cortex. NeuroImage, 19(2), 261–270.
Damaraju, E., Allen, E. A., Belger, A., Ford, J. M., McEwen, S., Mathalon, D. H., & Turner, J. A. (2014). Dynamic functional connectivity analysis reveals tran-
sient states of dysconnectivity in Schizophrenia. NeuroImage: Clinical, 5, 298–308.
De Luca, M., Beckmann, C. F., De Stefano, N., Matthews, P. M., & Smith, S. M. (2006). fMRI resting state networks define distinct modes of long-distance
interactions in the human brain. NeuroImage, 29(4), 1359–1367.
De Martino, F., Gentile, F., Esposito, F., Balsi, M., Di Salle, F., Goebel, R., & Formisano, E. (2007). Classification of fMRI independent components using IC-
fingerprints and support vector machine classifiers. NeuroImage, 34(1), 177–194.
De Martino, F., Valente, G., Staeren, N., Ashburner, J., Goebel, R., & Formisano, E. (2008). Combining multivariate voxel selection and support vector
machines for mapping and classification of fMRI spatial patterns. NeuroImage, 43(1), 44–58.
DeCharms, R. C. (2008). Applications of real-time fMRI. Nature Reviews Neuroscience, 9(9), 720.
Deng, F., Wang, Y., Huang, H., Niu, M., Zhong, S., Zhao, L., & He, Y. (2018). Abnormal segments of right uncinate fasciculus and left anterior thalamic radia-
tion in major and bipolar depression. Progress in Neuro-Psychopharmacology and Biological Psychiatry, 81, 340–349.
Dennis, E. L., & Thompson, P. M. (2014). Functional brain connectivity using fMRI in aging and Alzheimer's disease. Neuropsychology Review, 24(1), 49–62.
Deroy, O. (2019). Categorising without Concepts. Review of Philosophy and Psychology, 10(3), 1–14.
Di Martino, A., Kelly, C., Grzadzinski, R., Zuo, X. N., Mennes, M., Mairena, M. A., & Milham, M. P. (2011). Aberrant striatal functional connectivity in children
with autism. Biological Psychiatry, 69(9), 847–856.
Di, X., Kim, E. H., Huang, C. C., Lin, C. P., & Biswal, B. B. (2013). The influence of the amplitude of low-frequency fluctuations on resting-state functional
connectivity. Frontiers in Human Neuroscience, 7, 118.
Do, L. N., Yang, H. J., Kim, S. H., Lee, G. S., & Kim, S. H. (2015). A multi-voxel-activity-based feature selection method for human cognitive states classifica-
tion by functional magnetic resonance imaging data. Cluster Computing, 18(1), 199–208.
Dong, G., Huang, J., & Du, X. (2012). Alterations in regional homogeneity of resting-state brain activity in internet gaming addicts. Behavioral and Brain
Functions, 8(1), 41.
Dong, H. H., Guo, M. X., Zhang, Y. T., Fu, Y., & Shi, H. L. (2010, October). Sex differences in brain gray and white matter in healthy young adults: correla-
tions with resting state ALFF. Paper presented at 2010 3rd International Conference on Biomedical Engineering and Informatics (Vol. 2), IEEE.
pp. 560–563.
Dosenbach, N. U., Nardos, B., Cohen, A. L., Fair, D. A., Power, J. D., Church, J. A., & Barnes, K. A. (2010). Prediction of individual brain maturity using fMRI.
Science, 329(5997), 1358–1361.
Douglas, P. K., Harris, S., Yuille, A., & Cohen, M. S. (2011). Performance comparison of machine learning algorithms and number of independent compo-
nents used in fMRI decoding of belief vs. disbelief. NeuroImage, 56(2), 544–553.
Du, W., Calhoun, V. D., Li, H., Ma, S., Eichele, T., Kiehl, K. A., & Adali, T. (2012). High classification accuracy for schizophrenia with rest and task fMRI data.
Frontiers in Human Neuroscience, 6, 145.
Đuri
c, S., Lončar-Turukalo, T., Dabi
c, D., Koprivšek, K., Luči
c, M., & Šveljo, O. (2016, November). Temporal pattern based classification of independent com-
ponents in resting state fMRI. Paper presented at 2016 13th Symposium on Neural Networks and Applications (NEUREL), IEEE. pp. 1–4.
Endrass, T., Schuermann, B., Kaufmann, C., Spielberg, R., Kniesche, R., & Kathmann, N. (2010). Performance monitoring and error significance in patients
with obsessive-compulsive disorder. Biological Psychology, 84(2), 257–263.
Fan, T., Yao, L., Wu, X., & Liu, C. (2012, July). Independent component analysis of the resting-state brain functional MRI study in adults with bipolar depres-
sion. Paper presented at 2012 ICME International Conference on Complex Medical Engineering (CME), IEEE. pp. 38–42.
Fang, W., Lv, F., Luo, T., Cheng, O., Liao, W., Sheng, K., & Yang, Q. X. (2013). Abnormal regional homogeneity in patients with essential tremor revealed by
resting-state functional MRI. PloS One, 8(7), e69199.
Faro, S. H., & Mohamed, F. B. (2006). Functional MRI: Basic principles and clinical applications. Springer-Verlag New York: Springer Science & Business
Media.
Feis, R. A., Smith, S. M., Filippini, N., Douaud, G., Dopper, E. G., Heise, V., & Mackay, C. E. (2015). ICA-based artifact removal diminishes scan site differ-
ences in multi-center resting-state fMRI. Frontiers in Neuroscience, 9, 395.
Friston, K., Moran, R., & Seth, A. K. (2013). Analysing connectivity with Granger causality and dynamic causal modelling. Current Opinion in Neurobiology, 23
(2), 172–178.
Gandhi, T. K. (2016, March). Resting state fMRI analysis using seed based and ICA methods. Paper presented at 2016 3rd International Conference on
Computing for Sustainable Global Development (INDIACom), IEEE. pp. 2551–2554.
Gao, S., Osuch, E. A., Wammes, M., Théberge, J., Jiang, T. Z., Calhoun, V. D., & Sui, J. (2017, September). Discriminating bipolar disorder from major depres-
sion based on kernel SVM using functional independent components. Paper presented at 2017 IEEE 27th International Workshop on Machine Learning
for Signal Processing (MLSP), IEEE. pp. 1–6.
Gao, X. W., & Hui, R. (2016, July). A deep learning based approach to classification of CT brain images. In 2016 SAI Computing Conference (SAI), IEEE.
pp. 28–31.
Gardumi, A., Ivanov, D., Hausfeld, L., Valente, G., Formisano, E., & Uluda g, K. (2016). The effect of spatial resolution on decoding accuracy in fMRI multivar-
iate pattern analysis. NeuroImage, 132, 32–42.
Garner, R., La Rocca, M., Barisano, G., Toga, A. W., Duncan, D., & Vespa, P. (2019, April). A machine learning model to predict seizure susceptibility from
resting-state fMRI connectivity. Proceedings of the Modeling and Simulation in Medicine Symposium. Society for Computer Simulation Interna-
tional. p. 14.
RASHID ET AL. 23 of 29

Genuer, R., Michel, V., Eger, E., & Thirion, B. (2010, August). Random forests based feature selection for decoding fMRI data. In Proceedings Compstat (Vol.
267). pp. 1–8.
Gnanadas, A. A., Sathishbabu, S., & Vijayakarthik, M. (2017, February). Identification of epileptic seizure using ReHo from Rs-fMRI. Paper presented at
2017 Second International Conference on Recent Trends and Challenges in Computational Models (ICRTCCM), IEEE. pp. 292–296.
Goulden, N., Elliott, R., Suckling, J., Williams, S. R., Deakin, J. F. W., & McKie, S. (2012). Sample size estimation for comparing parameters using dynamic
causal modeling. Brain Connectivity, 2(2), 80–90.
Grandjean, J., Schroeter, A., Batata, I., & Rudin, M. (2014). Optimization of anesthesia protocol for resting-state fMRI in mice based on differential effects
of anesthetics on functional connectivity patterns. NeuroImage, 102, 838–847.
Grefkes, C., Nowak, D. A., Wang, L. E., Dafotakis, M., Eickhoff, S. B., & Fink, G. R. (2010). Modulating cortical connectivity in stroke patients by rTMS
assessed with fMRI and dynamic causal modeling. NeuroImage, 50(1), 233–242.
Griffanti, L., Salimi-Khorshidi, G., Beckmann, C. F., Auerbach, E. J., Douaud, G., Sexton, C. E., & Moeller, S. (2014). ICA-based artefact removal and acceler-
ated fMRI acquisition for improved resting state network imaging. NeuroImage, 95, 232–247.
Guo, M. X., Dong, H. H., Zhang, Y. T., Zhang, Q., & Yin, X. H. (2010, October). ALFF changes in brain areas of human with high myopia revealed by resting-
state functional MRI. Paper presented at 2010 3rd International Conference on Biomedical Engineering and Informatics (Vol. 1), IEEE. pp. 91–94.
Guo, R., Xie, S. Y., Cheng, X. N., & Zhao, H. (2009, June). Combined SVM and PCA to Recognize the Brain Function from fMRI Images. Paper presented at
2009 3rd International Conference on Bioinformatics and Biomedical Engineering, IEEE. pp. 1-3.
Guo, W. B., Liu, F., Xue, Z. M., Yu, Y., Ma, C. Q., Tan, C. L., & Chen, H. F. (2011). Abnormal neural activities in first-episode, treatment-naive, short-illness-
duration, and treatment-response patients with major depressive disorder: A resting-state fMRI study. Journal of Affective Disorders, 135(1-3),
326–331.
Gupta, A., Ayhan, M., & Maida, A. (2013, February). Natural image bases to represent neuroimaging data. Paper presented at International conference on
machine learning. pp. 987–994.
Hanson, E. A., Westlye, E., & Lundervold, A. (2014, April). A PCA-based thresholding strategy for group studies of brain connectivity-with applications to
resting state fMRI. Paper presented at 2014 Southwest Symposium on Image Analysis and Interpretation, IEEE. pp. 61–64.
Hao, A. J., He, B. L., & Yin, C. H. (2015). Discrimination of ADHD children based on Deep Bayesian Network.
Harrison, B. J., Yücel, M., Pujol, J., & Pantelis, C. (2007). Task-induced deactivation of midline cortical regions in schizophrenia assessed with fMRI. Schizo-
phrenia Research, 91(1-3), 82–86.
Harrison, T. M., McLaren, D. G., Moody, T. D., Feusner, J. D., & Bookheimer, S. Y. (2017). Generalized Psychophysiological Interaction (PPI) analysis of
memory related connectivity in individuals at genetic risk for alzheimer's disease. Journal of Visualized Experiments, 129, e55394.
Hojjati, S. H., Ebrahimzadeh, A., Khazaee, A., Babajani-Feremi, A., & Alzheimer's disease Neuroimaging Initiative. (2017). Predicting conversion from MCI to
AD using resting-state fMRI, graph theoretical approach and SVM. Journal of Neuroscience Methods, 282, 69–80.
Hosseini, S. H., & Kesler, S. R. (2014). Multivariate pattern analysis of FMRI in breast cancer survivors and healthy women. Journal of the International Neu-
ropsychological Society, 20(4), 391–401.
Hosseini-Asl, E., Gimel'farb, G., & El-Baz, A. (2016). Alzheimer's disease diagnostics by a deeply supervised adaptable 3D convolutional network. arXiv pre-
print arXiv:1607.00556.
Hu, C., Ju, R., Shen, Y., Zhou, P., & Li, Q. (2016, May). Clinical decision support for Alzheimer's disease based on deep learning and brain network. In 2016
IEEE International Conference on Communications (ICC), IEEE. pp. 1–6.
Hu, M., Ji, F., Lu, Z., Huang, W., Khosrowabadi, R., Zhao, L. & Stephenson, M. C. (2018, July). Differential amplitude of low-frequency fluctuations in brain
networks after BCI training with and without tDCS in stroke. Paper presented at 2018 40th Annual International Conference of the IEEE Engineering
in Medicine and Biology Society (EMBC), IEEE. pp. 1050–1053.
Huang, H., Hu, X., Zhao, Y., Makkie, M., Dong, Q., Zhao, S., & Liu, T. (2017). Modeling task fMRI data via deep convolutional autoencoder. IEEE Transactions
on Medical Imaging, 37(7), 1551–1561.
Huang, J., Mei, X., & Wang, X. (2018, June). De-noising and Feature Enhancement of fMRI Data Based on Naive Bayesian Link Prediction. In 2018 IEEE 3rd
International Conference on Image, Vision and Computing (ICIVC), IEEE. pp. 394–399.
Humphrey, E. J., Bello, J. P., & LeCun, Y. (2013). Feature learning and deep architectures: New directions for music informatics. Journal of Intelligent Informa-
tion Systems, 41(3), 461–481.
Huynh, H. T., & Won, Y. (2008, September). Decoding cognitive states from fMRI data using single hidden-layer feedforward neural networks. Paper pres-
ented at 2008 Fourth International Conference on Networked Computing and Advanced Information Management (Vol. 1), IEEE. pp. 256–260.
Jang, H., Plis, S. M., Calhoun, V. D., & Lee, J. H. (2017). Task-specific feature extraction and classification of fMRI volumes using a deep neural network ini-
tialized with a deep belief network: Evaluation using sensorimotor tasks. NeuroImage, 145, 314–328.
Ji, Y., Liu, H. B., Wang, X. K., & Tang, Y. Y. (2004, August). Cognitive states classification from fMRI data using support vector machines. Proceedings of
2004 International Conference on Machine Learning and Cybernetics (IEEE Cat. No. 04EX826) (Vol. 5). pp. 2919–2923.
Joel, S. E., Caffo, B. S., Van Zijl, P. C., & Pekar, J. J. (2011). On the relationship between seed-based and ICA-based measures of functional connectivity.
Magnetic Resonance in Medicine, 66(3), 644–657.
Johnston, B. A., Steele, J. D., Tolomeo, S., Christmas, D., & Matthews, K. (2015). Structural MRI-based predictions in patients with treatment-refractory
depression (TRD). PLoS One, 10(7), e0132958.
Johnston, B. A., Tolomeo, S., Gradin, V., Christmas, D., Matthews, K., & Douglas Steele, J. (2015). Failure of hippocampal deactivation during loss events in
treatment-resistant depression. Brain, 138(9), 2766–2776.
Johnstone, T., Ores Walsh, K. S., Greischar, L. L., Alexander, A. L., Fox, A. S., Davidson, R. J., & Oakes, T. R. (2006). Motion correction and the use of motion
covariates in multiple-subject fMRI analysis. Human Brain Mapping, 27(10), 779–788.
Karunanayaka, P., Schmithorst, V. J., Vannest, J., Szaflarski, J. P., Plante, E., & Holland, S. K. (2011). A linear structural equation model for covert verb gener-
ation based on independent component analysis of fMRI data from children and adolescents. Frontiers in Systems Neuroscience, 5, 29.
Kasess, C. H., Windischberger, C., Cunnington, R., Lanzenberger, R., Pezawas, L., & Moser, E. (2008). The suppressive influence of SMA on M1 in motor
imagery revealed by fMRI and dynamic causal modeling. NeuroImage, 40(2), 828–837.
Kay, K., Rokem, A., Winawer, J., Dougherty, R., & Wandell, B. (2013). GLMdenoise: A fast, automated technique for denoising task-based fMRI data. Fron-
tiers in Neuroscience, 7, 247.
24 of 29 RASHID ET AL.

Kesler, S. R., Rao, A., Blayney, D. W., Oakley-Girvan, I. A., Karuturi, M., & Palesh, O. (2017). Predicting long-term cognitive outcome following breast cancer
with pre-treatment resting state fMRI and random forest machine learning. Frontiers in Human Neuroscience, 11, 555.
Khazaee, A., Ebrahimzadeh, A., & Babajani-Feremi, A. (2015). Identifying patients with Alzheimer's disease using resting-state fMRI and graph theory. Clini-
cal Neurophysiology, 126(11), 2132–2141.
Khazaee, A., Ebrahimzadeh, A., & Babajani-Feremi, A. (2016). Application of advanced machine learning methods on resting-state fMRI network for identifi-
cation of mild cognitive impairment and Alzheimer's disease. Brain Imaging and Behavior, 10(3), 799–817.
Khazaee, A., Ebrahimzadeh, A., Babajani-Feremi, A., & Alzheimer's disease Neuroimaging Initiative. (2017). Classification of patients with MCI and AD from
healthy controls using directed graph measures of resting-state fMRI. Behavioural Brain Research, 322, 339–350.
Khodaee, A., Hossein-Zadeh, G. A., & Ananloo, E. S. (2015, May). Fractional amplitude of low-frequency fluctuations of BOLD signal in schizophrenia. Paper
presented at 2015 23rd Iranian Conference on Electrical Engineering, IEEE. pp. 104–108.
Khosla, M., Jamison, K., Kuceyeski, A., & Sabuncu, M. R. (2019). Ensemble learning with 3D convolutional neural networks for functional connectome-
based prediction. NeuroImage, 199, 651–662.
Kim, J., Calhoun, V. D., Shim, E., & Lee, J. H. (2016). Deep neural network with weight sparsity control and pre-training extracts hierarchical features and
enhances classification performance: Evidence from whole-brain resting-state functional connectivity patterns of schizophrenia. NeuroImage, 124,
127–146.
Kim, J., Zhu, W., Chang, L., Bentler, P. M., & Ernst, T. (2007). Unified structural equation modeling approach for the analysis of multisubject, multivariate
functional MRI data. Human Brain Mapping, 28(2), 85–93.
Kober, H., Lacadie, C. M., Wexler, B. E., Malison, R. T., Sinha, R., & Potenza, M. N. (2016). Brain activity during cocaine craving and gambling urges: an fMRI
study. Neuropsychopharmacology, 41, 628–637. https://doi.org/10.1038/npp.2015.193
Koch, W., Teipel, S., Mueller, S., Benninghoff, J., Wagner, M., Bokde, A. L., & Meindl, T. (2012). Diagnostic power of default mode network resting state
fMRI in the detection of Alzheimer's disease. Neurobiology of Aging, 33(3), 466–478.
Koike, N., Hatakeyama, Y., & Yoshida, S. (2014, August). Decoding colour of stimuli given to a human subject from functional magnetic resonance imaging
voxel patterns using machine learning algorithm. Paper presented at 2014 World Automation Congress (WAC), IEEE. pp. 681–686.
Kokkonen, S. M., Nikkinen, J., Remes, J., Kantola, J., Starck, T., Haapea, M., & Kiviniemi, V. (2009). Preoperative localization of the sensorimotor area using
independent component analysis of resting-state fMRI. Magnetic Resonance Imaging, 27(6), 733–740.
Koshino, H., Carpenter, P. A., Minshew, N. J., Cherkassky, V. L., Keller, T. A., & Just, M. A. (2005). Functional connectivity in an fMRI working memory task
in high-functioning autism. NeuroImage, 24(3), 810–821.
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. Paper presented at Advances in neural
information processing systems. pp. 1097–1105.
Krotov, D., & Hopfield, J. J. (2019). Unsupervised learning by competing hidden units. Proceedings of the National Academy of Sciences, 116(16),
7723–7731.
Kuang, D., & He, L. (2014, November). Classification on ADHD with deep learning. Paper presented at 2014 International Conference on Cloud Computing
and Big Data. IEEE. pp. 27–32.
Kuhn, S., Strelow, E., & Gallinat, J. (2016). Multiple “buy buttons” in the brain: Forecasting chocolate sales at point-of-sale based on functional brain activa-
tion using fMRI. NeuroImage, 136, 122–128. https://doi.org/10.1016/j.neuroimage.2016.05.021
Kwong, K. K. (2012). Record of a single fMRI experiment in May of 1991. NeuroImage, 62(2), 610–612.
LaConte, S., Strother, S., Cherkassky, V., Anderson, J., & Hu, X. (2005). Support vector machines for temporal classification of block design fMRI data.
NeuroImage, 26(2), 317–329.
LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436.
Lemm, S., Blankertz, B., Dickhaus, T., & Müller, K. R. (2011). Introduction to machine learning for brain imaging. NeuroImage, 56(2), 387–399.
Leonardi, N., Richiardi, J., Gschwind, M., Simioni, S., Annoni, J. M., Schluep, M., & Van De Ville, D. (2013). Principal components of functional connectivity:
A new approach to study dynamic brain connectivity during rest. NeuroImage, 83, 937–950.
Li, F., Tran, L., Thung, K. H., Ji, S., Shen, D., & Li, J. (2014, September). Robust deep learning for improved classification of AD/MCI patients. Paper pres-
ented at International Workshop on Machine Learning in Medical Imaging, Springer, Cham. pp. 240–247.
Li, M., Das, T., Deng, W., Wang, Q., Li, Y., Zhao, L., & Meng, Y. (2017). Clinical utility of a short resting-state MRI scan in differentiating bipolar from unipolar
depression. Acta psychiatrica Scandinavica, 136(3), 288–299.
Liang, M. J., Zhou, Q., Yang, K. R., Yang, X. L., Fang, J., Chen, W. L., & Huang, Z. (2013). Identify changes of brain regional homogeneity in bipolar disorder
and unipolar depression using resting-state FMRI. PloS One, 8(12), e79999.
Liang, P., Liu, Y., Jia, X., Duan, Y., Yu, C., Qin, W., & Li, K. (2011). Regional homogeneity changes in patients with neuromyelitis optica revealed by resting-
state functional MRI. Clinical Neurophysiology, 122(1), 121–127.
Liang, P., Xiang, J., Liang, H., Qi, Z., Li, K., & Alzheimer's disease NeuroImaging Initiative. (2014). Altered amplitude of low-frequency fluctuations in early
and late mild cognitive impairment and Alzheimer's disease. Current Alzheimer Research, 11(4), 389–398.
Lindquist, M. A., & McKeague, I. W. (2009). Logistic regression with Brownian-like predictors. Journal of the American Statistical Association, 104(488),
1575–1585.
Linn, K. A., Gaonkar, B., Satterthwaite, T. D., Doshi, J., Davatzikos, C., & Shinohara, R. T. (2016). Control-group feature normalization for multivariate pat-
tern analysis of structural MRI data using the support vector machine. NeuroImage, 132, 157–166.
Liu, D., Yan, C., Ren, J., Yao, L., Kiviniemi, V. J., & Zang, Y. (2010). Using coherence to measure regional homogeneity of resting-state FMRI signal. Frontiers
in Systems Neuroscience, 4, 24.
Liu, S., Liu, S., Cai, W., Che, H., Pujol, S., Kikinis, R., & Fulham, M. J. (2014). Multimodal neuroimaging feature learning for multiclass diagnosis of Alzheimer's
disease. IEEE Transactions on Biomedical Engineering, 62(4), 1132–1140.
Liu, S., Liu, S., Cai, W., Pujol, S., Kikinis, R., & Feng, D. (2014, April) Early diagnosis of Alzheimer's disease with deep learning. Paper presented at 2014 IEEE
11th international symposium on biomedical imaging (ISBI), IEEE. pp. 1015-1018.
Liu, X., Wang, S., Zhang, X., Wang, Z., Tian, X., & He, Y. (2014). Abnormal amplitude of low-frequency fluctuations of intrinsic brain activity in Alzheimer's
disease. Journal of Alzheimer's Disease, 40(2), 387–397.
RASHID ET AL. 25 of 29

Lorbert, A., & Ramadge, P. J. (2013, May). The pairwise elastic net support vector machine for automatic fMRI feature selection. Paper presented at 2013
IEEE International Conference on Acoustics, Speech and Signal Processing, IEEE. pp. 1036–1040.
Lordier, L., Loukas, S., Grouiller, F., Vollenweider, A., Vasung, L., Meskaldij, D. E., & Grandjean, D. (2019). Music processing in preterm and full-term new-
borns: A psychophysiological interaction (PPI) approach in neonatal fMRI. NeuroImage, 185, 857–864.
Lu, X., Yang, Y., Wu, F., Gao, M., Xu, Y., Zhang, Y., & Zhong, X. (2016). Discriminative analysis of schizophrenia using support vector machine and recursive
feature elimination on structural MRI images. Medicine, 95(30), e3973.
Mack, M. L., Preston, A. R., & Love, B. C. (2013). Decoding the brain's algorithm for categorization from its neural implementation. Current Biology, 23(20),
2023–2027.
Mane, T. N., Nagori, M. B., & Agrawal, S. A. (2012). fMRI data analysis with dynamic causal modeling and bayesian networks. Advanced Materials Research,
433, 5303–5307.
Månsson, K. N., Frick, A., Boraxbekk, C. J., Marquand, A. F., Williams, S. C. R., Carlbring, P., & Furmark, T. (2015). Predicting long-term outcome of Internet-
delivered cognitive behavior therapy for social anxiety disorder using fMRI and support vector machine learning. Translational Psychiatry, 5(3), e530.
Mao, Z., Su, Y., Xu, G., Wang, X., Huang, Y., Yue, W., & Xiong, N. (2019). Spatio-temporal deep learning method for ADHD fMRI classification. Information
Sciences, 499, 1–11.
Marco, G. D., Fall, S., & Vrignaud, P. (2007). Use of fMRI and structural equation modeling for studying interconnected brain areas within a hypothetical
network. Current Medical Imaging Reviews, 3(4), 249–261.
Marquand, A., Howard, M., Brammer, M., Chu, C., Coen, S., & Mour~ao-Miranda, J. (2010). Quantitative prediction of subjective pain intensity from whole-
brain fMRI data using Gaussian processes. NeuroImage, 49(3), 2178–2189.
Martínez-Ramón, M., Koltchinskii, V., Heileman, G. L., & Posse, S. (2006). fMRI pattern classification using neuroanatomically constrained boosting.
NeuroImage, 31(3), 1129–1141.
Matsubara, T., Tashiro, T., & Uehara, K. (2019). Deep neural generative model of functional MRI images for psychiatric disorder diagnosis. IEEE Transactions
on Biomedical Engineering, 66(10), 2768–2779.
Meier, T. B., Desphande, A. S., Vergun, S., Nair, V. A., Song, J., Biswal, B. B., & Prabhakaran, V. (2012). Support vector machine classification and characteri-
zation of age-related reorganization of functional brain networks. NeuroImage, 60(1), 601–613.
Meng, Z., Zhang, Z., Fan, Q., & Li, Y. (2018, July). Altered fractional amplitude of low frequency fluctuations in unmedicated female patients with obsessive-
compulsive disorder. Paper presented at 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC),
IEEE. pp. 1144–1147.
Michalski, R. S. (1986). Understanding the nature of learning: Issues and research directions. Machine Learning: An Artificial Intelligence Approach, 2(1), 3–25.
Michel, V., Damon, C., & Thirion, B. (2008, May). Mutual information-based feature selection enhances fMRI brain activity classification. Paper presented
at 2008 5th IEEE international symposium on biomedical imaging: From nano to macro, IEEE. pp. 592–595.
Misaki, M., Luh, W. M., & Bandettini, P. A. (2013). The effect of spatial smoothing on fMRI decoding of columnar-level organization with linear support vec-
tor machine. Journal of Neuroscience Methods, 212(2), 355–361.
Moore, M., Culpepper, S., Phan, K. L., Strauman, T. J., Dolcos, F., & Dolcos, S. (2018). Neurobehavioral mechanisms of resilience against emotional distress:
An integrative brain-personality-symptom approach using structural equation modeling. Personality Neuroscience, 1(e8), 1–10.
Morra, J. H., Tu, Z., Apostolova, L. G., Green, A. E., Toga, A. W., & Thompson, P. M. (2009). Comparison of AdaBoost and support vector machines for
detecting Alzheimer's disease through automated hippocampal segmentation. IEEE Transactions on Medical Imaging, 29(1), 30–43.
Mourao-Miranda, J., Bokde, A. L., Born, C., Hampel, H., & Stetter, M. (2005). Classifying brain states and determining the discriminating activation patterns:
Support vector machine on functional MRI data. NeuroImage, 28(4), 980–995.
Munsell, B. C., Wee, C. Y., Keller, S. S., Weber, B., Elger, C., da Silva, L. A. T., … Bonilha, L. (2015). Evaluation of machine learning algorithms for treatment
outcome prediction in patients with epilepsy based on structural connectome data. NeuroImage, 118, 219–230.
Musso, F., Brinkmeyer, J., Mobascher, A., Warbrick, T., & Winterer, G. (2010). Spontaneous brain activity and EEG microstates. A novel EEG/fMRI analysis
approach to explore resting-state networks. NeuroImage, 52(4), 1149–1161.
Nakai, T., Bagarinao, E., Matsuo, K., Ohgami, Y., & Kato, C. (2006). Dynamic monitoring of brain activation under visual stimulation using fMRI—the advan-
tage of real-time fMRI with sliding window GLM analysis. Journal of Neuroscience Methods, 157(1), 158–167.
Ng, A. (2015). What data scientists should know about deep learning?, 44. Retrieved from https://www.slideshare.net/ExtractConf
Norman, K. A., Polyn, S. M., Detre, G. J., & Haxby, J. V. (2006). Beyond mind-reading: Multi-voxel pattern analysis of fMRI data. Trends in Cognitive Sciences,
10(9), 424–430.
Ogawa, S. (2012). Finding the BOLD effect in brain images. NeuroImage, 62(2), 608–609.
Onal, I., Ozay, M., Firat, O., Öztekin, _I., & Vural, F. T. Y. (2013, July). Analyzing the information distribution in the fmri measurements by estimating the
degree of locality. Paper presented at 2013 35th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC),
IEEE. pp. 6772–6775.
Osada, T., Adachi, Y., Miyamoto, K., Jimura, K., Setsuie, R., & Miyashita, Y. (2015). Dynamically allocated hub in task-evoked network predicts the vulnera-
ble prefrontal locus for contextual memory retrieval in macaques. PLoS Biology, 13(6), e1002177.
Osório, P., Rosa, P., Silvestre, C., & Figueiredo, P. (2015). Stochastic dynamic causal modelling of fMRI data with multiple-model kalman filters. Methods of
Information in Medicine, 54(03), 232–239.
Parida, S., Dehuri, S., & Cho, S. B. (2014, February). Application of genetic algorithms and gaussian naïve bayesian approach in pipeline for cognitive state
classification. Paper presented at 2014 IEEE International Advance Computing Conference (IACC), IEEE. pp. 1237–1242.
Patrick, E., & Fischer, F., III. (1970). A generalized k-nearest neighbor rule. Information and Control, 16(2), 128–152. https://doi.org/10.1016/S0019-9958
(70)90081-1
Peltier, S. J., Lisinski, J. M., Noll, D. C., & LaConte, S. M. (2009, September). Support vector machine classification of complex fMRI data. Paper presented at
2009 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, IEEE. pp. 5381–5384.
Peng, D. H., Jiang, K. D., Fang, Y. R., Xu, Y. F., Ting, S., Long, X. Y., & Zang, Y. F. (2011). Decreased regional homogeneity in major depression as revealed by
resting-state functional magnetic resonance imaging. Chinese Medical Journal, 124(3), 369–373.
Penny, W., Iglesias-Fuster, J., Quiroz, Y. T., Lopera, F. J., & Bobes, M. A. (2018). Dynamic causal modeling of preclinical autosomal-dominant Alzheimer's dis-
ease. Journal of Alzheimer's Disease, 65(3), 697–711.
26 of 29 RASHID ET AL.

Perlbarg, V., Marrelec, G., Doyon, J., Pélégrini-Issac, M., Lehéricy, S., & Benali, H. (2008, May). NEDICA: Detection of group functional networks in fMRI
using spatial independent component analysis. Paper presented at 2008 5th IEEE International Symposium on Biomedical Imaging: From Nano to
Macro, IEEE. pp. 1247–1250.
Plis, S. M., Amin, M. F., Chekroud, A., Hjelm, D., Damaraju, E., Lee, H. J., & Calhoun, V. D. (2018). Reading the (functional) writing on the (structural) wall:
Multimodal fusion of brain structure and function via a deep neural network based translation approach reveals novel impairments in schizophrenia.
NeuroImage, 181, 734–747.
Plumpton, C. O., Kuncheva, L. I., Linden, D. E., & Johnston, S. J. (2010, August). On-line fmri data classification using linear and ensemble classifiers. Paper
presented at 2010 20th International Conference on Pattern Recognition, IEEE. pp. 4312–4315.
Poldrack, R. A. (2012). The future of fMRI in cognitive neuroscience. NeuroImage, 62, 1216–1220. https://doi.org/10.1016/j.neuroimage.2011.08.007
Portugal, I., Alencar, P., & Cowan, D. (2018). The use of machine learning algorithms in recommender systems: A systematic review. Expert Systems with
Applications, 97, 205–227.
Pruim, R. H., Mennes, M., Buitelaar, J. K., & Beckmann, C. F. (2015). Evaluation of ICA-AROMA and alternative strategies for motion artifact removal in
resting state fMRI. NeuroImage, 112, 278–287.
Pruim, R. H., Mennes, M., van Rooij, D., Llera, A., Buitelaar, J. K., & Beckmann, C. F. (2015). ICA-AROMA: A robust ICA-based strategy for removing motion
artifacts from fMRI data. NeuroImage, 112, 267–277.
Qin, Y., Tong, L., Gao, H., Bu, H., Wang, L., Zeng, Y., & Yan, B. (2017, December). Altered resting-state regional homogeneity after real-time fMRI emotion
self-regulation training. Paper presented at 2017 14th International Computer Conference on Wavelet Active Media Technology and Information
Processing (ICCWAMTIP), IEEE. pp. 285–288.
Qing, Z., Dong, Z., Li, S., Zang, Y., & Liu, D. (2015). Global signal regression has complex effects on regional homogeneity of resting state fMRI signal. Mag-
netic Resonance Imaging, 33(10), 1306–1313.
Qiu, S., Chen, F., Chen, G., Jia, Y., Gong, J., Luo, X., & Huang, L. (2019). Abnormal resting-state regional homogeneity in unmedicated bipolar II disorder.
Journal of Affective Disorders, 256, 604–610.
Qureshi, M. N. I., Oh, J., Cho, D., Jo, H. J., & Lee, B. (2017). Multimodal discrimination of schizophrenia using hybrid weighted feature concatenation of
brain functional connectivity and anatomical features with an extreme learning machine. Frontiers in Neuroinformatics, 11, 59.
Rashid, B., Arbabshirani, M. R., Damaraju, E., Cetin, M. S., Miller, R., Pearlson, G. D., & Calhoun, V. D. (2016). Classification of schizophrenia and bipolar
patients using static and dynamic resting-state fMRI brain connectivity. NeuroImage, 134, 645–657.
Rashid, B., Damaraju, E., Pearlson, G. D., & Calhoun, V. D. (2014). Dynamic connectivity states estimated from resting fMRI Identify differences among
Schizophrenia, bipolar disorder, and healthy control subjects. Frontiers in Human Neuroscience, 8, 897.
Richiardi, J., Eryilmaz, H., Schwartz, S., Vuilleumier, P., & Van De Ville, D. (2010, April). Brain decoding of fMRI connectivity graphs using decision tree
ensembles. Paper presented at 2010 IEEE International Symposium on Biomedical Imaging: From Nano to Macro, IEEE. pp. 1137–1140.
Rish, I. (2001, August). An empirical study of the naive Bayes classifier. IJCAI 2001 Workshop on Empirical Methods in Artificial Intelligence, 3(22), 41–46.
Ronaghan, Stacey. (2018). Deep Learning: Common Architectures. Retrieved from https://medium.com/@srnghn/deep-learning-common-architectures-
6071d47cb383
Rubin-Falcone, H., Zanderigo, F., Thapa-Chhetry, B., Lan, M., Miller, J. M., Sublette, M. E., & Mann, J. J. (2018). Pattern recognition of magnetic resonance
imaging-based gray matter volume measurements classifies bipolar disorder and major depressive disorder. Journal of Affective Disorders, 227,
498–505.
Ryali, S., & Menon, V. (2009). Feature selection and classification of fMRI data using logistic regression with L1 norm regularization. NeuroImage, 47, S57.
Ryali, S., Supekar, K., Abrams, D. A., & Menon, V. (2010). Sparse logistic regression for whole-brain classification of fMRI data. NeuroImage, 51(2), 752–764.
Sacchet, M. D., Prasad, G., Foland-Ross, L. C., Thompson, P. M., & Gotlib, I. H. (2015). Support vector machine classification of major depressive disorder
using diffusion-weighted neuroimaging and graph theory. Frontiers in Psychiatry, 6, 21.
Sankar, A., Zhang, T., Gaonkar, B., Doshi, J., Erus, G., Costafreda, S. G., & Fu, C. H. (2016). Diagnostic potential of structural neuroimaging for depression
from a multi-ethnic community sample. BJPsych Open, 2(4), 247–254.
Sarraf, S., & Tofighi, G. (2016). Classification of Alzheimer's disease using fmri data and deep learning convolutional neural networks. arXiv preprint arXiv:
1603.08631.
Schmidhuber, J. (2015). Deep learning in neural networks: An overview. Neural Networks, 61, 85–117.
Schrouff, J., Kussé, C., Wehenkel, L., Maquet, P., & Phillips, C. (2012). Decoding semi-constrained brain activity from fMRI using support vector machines
and Gaussian processes. PLoS One, 7(4), e35860.
Seghier, M. L., Zeidman, P., Neufeld, N. H., Leff, A. P., & Price, C. (2010). Identifying abnormal connectivity in patients using dynamic causal modelling of
fMRI responses. Frontiers in Systems Neuroscience, 4, 142.
Seo, S., Mohr, J., Beck, A., Wüstenberg, T., Heinz, A., & Obermayer, K. (2015). Predicting the future relapse of alcohol-dependent patients from structural
and functional brain images. Addiction Biology, 20(6), 1042–1055.
Shahin, S., Shayegh, F., Mortaheb, S., & Amirfattahi, R. (2016, November). Improvement of flexible design matrix in sparse Bayesian learning for multi task
fMRI data analysis. Paper presented at 2016 23rd Iranian Conference on Biomedical Engineering and 2016 1st International Iranian Conference on Bio-
medical Engineering (ICBME), IEEE. pp. 47–52.
Shi, B., Chen, Y., Zhang, P., Smith, C. D., Liu, J., & Alzheimer's disease Neuroimaging Initiative. (2017). Nonlinear feature transformation and deep fusion for
Alzheimer's disease staging analysis. Pattern Recognition, 63, 487–498.
Shukla, D. K., Keehn, B., & Müller, R. A. (2010). Regional homogeneity of fMRI time series in autism spectrum disorders. Neuroscience Letters, 476(1),
46–51.
Sladky, R., Höflich, A., Küblböck, M., Kraus, C., Baldinger, P., Moser, E., & Windischberger, C. (2013). Disrupted effective connectivity between the amyg-
dala and orbitofrontal cortex in social anxiety disorder during emotion discrimination revealed by dynamic causal modeling for fMRI. Cerebral Cortex,
25(4), 895–903.
Smith, D. V., Utevsky, A. V., Bland, A. R., Clement, N., Clithero, J. A., Harsch, A. E., … Huettel, S. A. (2014). Characterizing individual differences in functional
connectivity using dual-regression and seed-based approaches. NeuroImage, 95, 1–12.
Soch, J., Haynes, J. D., & Allefeld, C. (2016). How to avoid mismodelling in GLM-based fMRI data analysis: Cross-validated Bayesian model selection.
NeuroImage, 141, 469–489.
RASHID ET AL. 27 of 29

Song, X., Zhang, Y., & Liu, Y. (2014). Frequency specificity of regional homogeneity in the resting-state human brain. PloS One, 9(1), e86818.
Sörös, P., & Witt, K. (2018). Book review: Introduction to neuroimaging analysis. Frontiers in Neuroscience, 12, 674.
Su, M., Wang, S., Fang, W., Zhu, Y., Li, R., Sheng, K., & Cheng, O. (2015). Alterations in the limbic/paralimbic cortices of Parkinson's disease patients with
hyposmia under resting-state functional MRI by regional homogeneity and functional connectivity analysis. Parkinsonism & Related Disorders, 21(7),
698–703.
Sui, J., Castro, E., He, H., Bridwell, D., Du, Y., Pearlson, G. D. & Calhoun, V. D. (2014, August). Combination of FMRI-SMRI-EEG data improves discrimina-
tion of schizophrenia patients by ensemble feature selection. Paper presented at 2014 36th Annual International Conference of the IEEE Engineering
in Medicine and Biology Society, IEEE. pp. 3889–3892.
Suk, H. I., Lee, S. W., Shen, D., & Alzheimer's disease Neuroimaging Initiative. (2015). Latent feature representation with stacked auto-encoder for AD/MCI
diagnosis. Brain Structure and Function, 220(2), 841–859.
Suk, H. I., Lee, S. W., Shen, D., & Alzheimer's disease Neuroimaging Initiative. (2016). Deep sparse multi-task learning for feature selection in Alzheimer's
disease diagnosis. Brain Structure and Function, 221(5), 2569–2587.
Suk, H. I., Lee, S. W., Shen, D., & Alzheimer's disease Neuroimaging Initiative. (2017). Deep ensemble learning of sparse regression models for brain disease
diagnosis. Medical Image Analysis, 37, 101–113.
Suk, H. I., Lee, S. W., Shen, D., & Alzheimer's disease Neuroimaging Initiative. (2014). Hierarchical feature representation and multimodal fusion with deep
learning for AD/MCI diagnosis. NeuroImage, 101, 569–582.
Suk, H. I., & Shen, D. (2013, September). Deep learning-based feature representation for AD/MCI classification. Paper presented at International Confer-
ence on Medical Image Computing and Computer-Assisted Intervention, Springer, Berlin, Heidelberg. pp. 583–590.
Suk, H. I., Wee, C. Y., Lee, S. W., & Shen, D. (2016). State-space model with deep learning for functional dynamics estimation in resting-state fMRI.
NeuroImage, 129, 292–307.
Sundermann, B., Burgmer, M., Pogatzki-Zahn, E., Gaubitz, M., Stüber, C., Wessolleck, E., & Pfleiderer, B. (2014). Diagnostic classification based on func-
tional connectivity in chronic pain: Model optimization in fibromyalgia and rheumatoid arthritis. Academic Radiology, 21(3), 369–377.
Szepesvári, C. (2010). Algorithms for reinforcement learning. Synthesis lectures on artificial intelligence and machine learning, 4(1), 1–103.
Taherkhani, A., Belatreche, A., Li, Y., & Maguire, L. P. (2018). A supervised learning algorithm for learning precise timing of multiple spikes in multilayer spik-
ing neural networks. IEEE Transactions on Neural Networks and Learning Systems, 29(11), 5394–5407.
Tahmassebi, A., Gandomi, A. H., Schulte, M. H., Goudriaan, A. E., Foo, S. Y., & Meyer-Baese, A. (2018). Optimized naive-bayes and decision tree approaches
for fmri smoking cessation classification. Complexity, 2018, 24.
Tailby, C., Kowalczyk, M. A., & Jackson, G. D. (2018). Cognitive impairment in epilepsy: The role of reduced network flexibility. Annals of Clinical and Trans-
lational Neurology, 5(1), 29–40.
Tan, L., Holland, S. K., Deshpande, A. K., Chen, Y., Choo, D. I., & Lu, L. J. (2015). A semi-supervised support vector machine model for predicting the lan-
guage outcomes following cochlear implantation based on pre-implant brain fMRI imaging. Brain and Behavior, 5(12), e00391.
Tang, Y., Jiang, W., Liao, J., Wang, W., & Luo, A. (2013). Identifying individuals with antisocial personality disorder using resting-state FMRI. PloS One, 8(4),
e60652.
Thirion, B., Varoquaux, G., Dohmatob, E., & Poline, J. B. (2014). Which fMRI clustering gives good brain parcellations? Frontiers in Neuroscience, 8, 167.
Tian, L., Jiang, T., Wang, Y., Zang, Y., He, Y., Liang, M., & Zhuo, Y. (2006). Altered resting-state functional connectivity patterns of anterior cingulate cortex
in adolescents with attention deficit hyperactivity disorder. Neuroscience Letters, 400(1-2), 39–43.
Tong, S., & Koller, D. (2001). Support vector machine active learning with applications to text classification. Journal of Machine Learning Research, 2, 45–66.
Turek, J. S., Willke, T. L., Chen, P. H., & Ramadge, P. J. (2017, March). A semi-supervised method for multi-subject fMRI functional alignment. Paper pres-
ented at 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE. pp. 1098–1102.
Turner, J. A., Chen, H., Mathalon, D. H., Allen, E. A., Mayer, A. R., Abbott, C. C., … Bustillo, J. (2012). Reliability of the amplitude of low-frequency fluctua-
tions in resting state fMRI in chronic schizophrenia. Psychiatry Research: Neuroimaging, 201(3), 253–255.
Van Dijk, K. R., Sabuncu, M. R., & Buckner, R. L. (2012). The influence of head motion on intrinsic functional connectivity MRI. NeuroImage, 59(1),
431–438.
Vergun, S., Deshpande, A., Meier, T. B., Song, J., Tudorascu, D. L., Nair, V. A., & Prabhakaran, V. (2013). Characterizing functional connectivity differences
in aging adults using machine learning on resting state fMRI data. Frontiers in Computational Neuroscience, 7, 38.
Vijay, K., & Selvakumar, K. (2015, April). Brain fMRI clustering using interaction K-means algorithm with PCA. Paper presented at 2015 International Con-
ference on Communications and Signal Processing (ICCSP), IEEE. pp. 0909–0913.
Von dem Hagen, E. A., Stoyanova, R. S., Baron-Cohen, S., & Calder, A. J. (2012). Reduced functional connectivity within and between social resting state
networks in autism spectrum conditions. Social Cognitive and Affective Neuroscience, 8(6), 694–701.
Wang, F., Li, Y., & Gu, Z. (2017). An MVPA method based on sparse representation for pattern localization in fMRI data analysis. Neurocomputing, 269,
206–211.
Wang, F., Zhang, P., Qian, B., Wang, X., & Davidson, I. (2014, August). Clinical risk prediction with multilinear sparse logistic regression. Paper presented at
Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining, ACM. pp. 145–154.
Wang, L., Song, M., Jiang, T., Zhang, Y., & Yu, C. (2011). Regional homogeneity of the resting-state brain activity correlates with individual intelligence. Neu-
roscience Letters, 488(3), 275–278.
Wang, M., Li, C., Zhang, W., Wang, Y., Feng, Y., Liang, Y., & Chen, R. (2019). Support vector machine for analyzing contributions of brain regions during
task-state fMRI. Frontiers in Neuroinformatics, 13, 10.
Wang, X., Jiao, Y., & Lu, Z. (2011, July). Discriminative analysis of resting-state brain functional connectivity patterns of attention-deficit hyperactivity dis-
order using kernel principal component analysis. Paper presented at 2011 Eighth International Conference on Fuzzy Systems and Knowledge Discov-
ery (FSKD) (Vol. 3), IEEE. pp. 1938–1941.
Wang, X., Ren, Y., & Zhang, W. (2017). Depression disorder classification of fmri data using sparse low-rank functional brain network and graph-based fea-
tures. Computational and Mathematical Methods in Medicine, 2017, 3609821.
Wang, Z. (2009). A hybrid SVM–GLM approach for fMRI data analysis. Neuroimage, 46(3), 608–615.
Wang, Z., Childress, A. R., Wang, J., & Detre, J. A. (2007). Support vector machine learning-based fMRI data group analysis. NeuroImage, 36(4), 1139–1151.
28 of 29 RASHID ET AL.

Wee, C. Y., Yang, S., Yap, P. T., Shen, D., & Alzheimer's disease Neuroimaging Initiative. (2016). Sparse temporally dynamic resting-state functional connec-
tivity networks for early MCI identification. Brain Imaging and Behavior, 10(2), 342–356.
Wetherill, R. R., Rao, H., Hager, N., Wang, J., Franklin, T. R., & Fan, Y. (2019). Classifying and characterizing nicotine use disorder with high accuracy using
machine learning and resting-state fMRI. Addiction Biology, 24(4), 811–821.
Wig, G. S., Schlaggar, B. L., & Petersen, S. E. (2011). Concepts and principles in the analysis of brain networks. Annals of the New York Academy of Sciences,
1224(1), 126–146.
Woermann, F. G., Jokeit, H., Luerding, R., Freitag, H., Schulz, R., Guertler, S., & Ebner, A. (2003). Language lateralization by Wada test and fMRI in
100 patients with epilepsy. Neurology, 61(5), 699–701.
Wu, J., Qian, Z., Tao, L., Yin, J., Ding, S., Zhang, Y., & Yu, Z. (2015). Resting state fMRI feature-based cerebral glioma grading by support vector machine.
International Journal of Computer Assisted Radiology and Surgery, 10(7), 1167–1174.
Wu, T., Long, X., Zang, Y., Wang, L., Hallett, M., Li, K., & Chan, P. (2009). Regional homogeneity changes in patients with Parkinson's disease. Human Brain
Mapping, 30(5), 1502–1510.
Wu, T., Zang, Y., Wang, L., Long, X., Li, K., & Chan, P. (2007). Normal aging decreases regional homogeneity of the motor areas in the resting state. Neurosci-
ence Letters, 423(3), 189–193.
Xie, X., Cao, Z., & Weng, X. (2008). Spatiotemporal nonlinearity in resting-state fMRI of the human brain. NeuroImage, 40(4), 1672–1685.
Xu, J., Potenza, M. N., & Calhoun, V. D. (2013). Spatial ICA reveals functional activity hidden from traditional fMRI GLM-based analyses. Frontiers in Neuro-
science, 7, 154.
Yan, L., Zhuo, Y., Wang, B., & Wang, D. J. (2011). Suppl 1: Loss of coherence of low frequency fluctuations of bold fmri in visual cortex of healthy aged sub-
jects. The Open Neuroimaging Journal, 5, 105.
Yan, S., Yang, X., Wu, C., & Guo, Y. (2014, August). Integration of sparse Bayesian learning and random subspace for fMRI Multivariate Pattern Analysis. In
2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, IEEE. pp. 1035–1038.
Yang, H., Long, X. Y., Yang, Y., Yan, H., Zhu, C. Z., Zhou, X. P., … Gong, Q. Y. (2007). Amplitude of low frequency fluctuation within visual areas revealed by
resting-state functional MRI. NeuroImage, 36(1), 144–152.
Yang, K., & Shahabi, C. (2007). An efficient k nearest neighbor search for multivariate time series. Information and Computation, 205(1), 65–98.
Yang, Q., Qin, Y., Wang, L., Sun, L., Yan, B., Tong, L. & Li, Z. (2018, October). Resting-state regional homogeneity analysis on real-time fMRI emotion self-
regulation training. Paper presented at 2018 International Conference on Control, Automation and Information Sciences (ICCAIS), IEEE. pp. 146–150.
Yargholi, E., & Hossein-Zadeh, G. A. (2016). Brain decoding-classification of hand written digits from fMRI data employing bayesian networks. Frontiers in
Human Neuroscience, 10, 351.
Yetkin, F. Z., Rosenberg, R. N., Weiner, M. F., Purdy, P. D., & Cullum, C. M. (2006). FMRI of working memory in patients with mild cognitive impairment and
probable Alzheimer's disease. European Radiology, 16(1), 193–206.
Yin, C., Zhu, Y., Fei, J., & He, X. (2017). A deep learning approach for intrusion detection using recurrent neural networks. IEEE Access, 5, 21954–21961.
Yoshida, K., Shimizu, Y., Yoshimoto, J., Takamura, M., Okada, G., Okamoto, Y., & Doya, K. (2017). Prediction of clinical depression scores and detection of
changes in whole-brain using resting-state functional MRI data with partial least squares regression. PloS One, 12(7), e0179638.
Yu, R., Chien, Y. L., Wang, H. L. S., Liu, C. M., Liu, C. C., Hwang, T. J., … Tseng, W. Y. I. (2014). Frequency-specific alternations in the amplitude of low-
frequency fluctuations in schizophrenia. Human Brain Mapping, 35(2), 627–637.
Yuan, K., Jin, C., Cheng, P., Yang, X., Dong, T., Bi, Y., & Liang, J. (2013). Amplitude of low frequency fluctuation abnormalities in adolescents with online
gaming addiction. PLoS One, 8(11), e78708.
Yuan, R., Di, X., Kim, E. H., Barik, S., Rypma, B., & Biswal, B. B. (2013). Regional homogeneity of resting-state fMRI contributes to both neurovascular and
task activation variations. Magnetic Resonance Imaging, 31(9), 1492–1500.
Yuan, X. L., Mei, X., Huang, J. S., & Yang, J. (2014). Feature selection based on random forest algorithm and the applications in fMRI data. Microelectronics &
Computer, 31(8), 132–135.
Yu-Feng, Z., Yong, H., Chao-Zhe, Z., Qing-Jiu, C., Man-Qiu, S., Meng, L., & Yu-Feng, W. (2007). Altered baseline brain activity in children with ADHD rev-
ealed by resting-state functional MRI. Brain and Development, 29(2), 83–91.
Yurgelun-Todd, D. A., Gruber, S. A., Kanayama, G., Killgore, W. D., Baird, A. A., & Young, A. D. (2000). fMRI during affect discrimination in bipolar affective
disorder. Bipolar Disorders, 2, 237–248.
Zafar, R., Malik, A. S., Shuaibu, A. N., ur Rehman, M. J., & Dass, S. C. (2017, September). Classification of fMRI data using support vector machine and con-
volutional neural network. Paper presented at 2017 IEEE International Conference on Signal and Image Processing Applications (ICSIPA), IEEE.
pp. 324–329.
Zang, Y., Jiang, T., Lu, Y., He, Y., & Tian, L. (2004). Regional homogeneity approach to fMRI data analysis. NeuroImage, 22(1), 394–400.
Zeng, H., Pizarro, R., Nair, V. A., La, C., & Prabhakaran, V. (2013). Alterations in regional homogeneity of resting-state brain activity in mesial temporal lobe
epilepsy. Epilepsia, 54(4), 658–666.
Zhang, C., Yao, L., Song, S., Wen, X., Zhao, X., & Long, Z. (2017). Euler elastica regularized logistic regression for whole-brain decoding of fMRI data. IEEE
Transactions on Biomedical Engineering, 65(7), 1639–1653.
Zhang, T., Satterthwaite, T. D., Elliott, M., Gur, R. C., Gur, R. E., & Davatzikos, C. (2012, July). Multivariate fMRI analysis using optimally-discriminative
voxel-based analysis. Paper presented at 2012 Second International Workshop on Pattern Recognition in NeuroImaging, IEEE. pp. 33–36.
Zhang, Y., Hu, X., He, C., Wang, X., Ren, Y., Liu, H. & Liu, T. (2019, April). A two-stage DBN-based method to exploring functional brain networks in natural-
istic paradigm FMRI. Paper presented at 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019), IEEE. pp. 1594-1597.
Zhang, Y., Zhu, C., Chen, H., Duan, X., Lu, F., Li, M., & Zhang, W. (2015). Frequency-dependent alterations in the amplitude of low-frequency fluctuations in
social anxiety disorder. Journal of Affective Disorders, 174, 329–335.
Zhao, Y., Dong, Q., Zhang, S., Zhang, W., Chen, H., Jiang, X., & Liu, T. (2017). Automatic recognition of fmri-derived functional networks using 3-d con-
volutional neural networks. IEEE Transactions on Biomedical Engineering, 65(9), 1975–1984.
Zhong, Y., Wang, H., Lu, G., Zhang, Z., Jiao, Q., & Liu, Y. (2009). Detecting functional connectivity in fMRI using PCA and regression analysis. Brain Topogra-
phy, 22(2), 134–144.
Zhou, Y., Liang, M., Tian, L., Wang, K., Hao, Y., Liu, H., & Jiang, T. (2007). Functional disintegration in paranoid schizophrenia using resting-state fMRI.
Schizophrenia Research, 97(1-3), 194–205.
RASHID ET AL. 29 of 29

Zhu, X. (2017). Semi-supervised learning. In Encyclopedia of machine learning and data mining (pp. 1142–1147). Boston, MA: Springer.
Zhuang, J., Dvornek, N. C., Li, X., Yang, D., Ventola, P., & Duncan, J. S. (2018, April). Prediction of pivotal response treatment outcome with task fMRI using
random forest and variable selection. Paper presented at 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018), IEEE.
pp. 97–100.
Zhuang, J., Peltier, S., He, S., LaConte, S., & Hu, X. (2008). Mapping the connectivity with structural equation modeling in an fMRI study of shape-from-
motion task. NeuroImage, 42(2), 799–806.
Zilverstand, A., Sorger, B., Zimmermann, J., Kaas, A., & Goebel, R. (2014). Windowed correlation: A suitable tool for providing dynamic fMRI-based func-
tional connectivity neurofeedback on task difficulty. PLoS One, 9(1), e85929.
Zou, Q. H., Zhu, C. Z., Yang, Y., Zuo, X. N., Long, X. Y., Cao, Q. J., & Zang, Y. F. (2008). An improved approach to detection of amplitude of low-frequency
fluctuation (ALFF) for resting-state fMRI: Fractional ALFF. Journal of Neuroscience Methods, 172(1), 137–141.

AUTHOR BIOGRAPHI ES

Mamoon Rashid is a Ph.D. Research Scholar in the School of Computer Science & Engineering, University College of Engineering, Punjabi
University, Patiala, India. His main focus of research in his Ph.D. is improving the computational efficiency in 4-Dimensional Functional
Magnetic Resonance Images. His research interests include Big Data Analytics, Machine Learning, and Computational Neuro Imaging. In his
ongoing Ph.D work, he has published papers in indexed journals and presented papers in conferences of International repute.

Dr. Harjeet Singh is presently working as an Associate Professor in the P.G. Department of Computer Science, Mata Gujri College, Fatehgarh
Sahib, Punjab, India. He has recieved his Ph.D. degree from I.K. Gujral Punjab Technical University, India. His areas of interest include Machine
Translation, Neuroimaging, and Natural Language Processing. He has published many papers in indexed journals and is currently guiding
several research scholars at Doctorate level.

Dr. Vishal Goyal is presently working as Professor in Department of Computer Science, Punjabi University, Patiala, India. He is Coordinator,
Research Center for Technology Development for Differently Abled people and co-coordinator of Center for Artificial Intelligence and Data
Science, Punjabi University, Patiala. His main research area is Cognitive Computing, Artificial Intelligence, Natural Language Processing and
Machine Translation Language Technologies. He has been awarded Young Scientist Award in 2005. He has Copyrighted Software Hindi to
Punjabi Machine Translation System and Software Automatic Translation of English to ISL Synthetic Videos. He has worked as Principal
Investigator for Indian Languages Corpora Initiative (ILCI) funded under the Technology Development for Indian Languages (TDIL) program by
Ministry of Communications and Information Technology. He is currently working on research funded project of Plagiarism Detection Tool
Development for Indian Languages with Special Focus on Hindi and Punjabi, funded by R&D in Electronics Group (Innovation and IPR
Division), Ministry of Electronics and Information Technology, Govt. of India. He has published papers in indexed journals and guided number
of research scholars at Doctorate level.

How to cite this article: Rashid M, Singh H, Goyal V. The use of machine learning and deep learning algorithms in functional magnetic
resonance imaging—A systematic review. Expert Systems. 2020;e12644. https://doi.org/10.1111/exsy.12644

View publication stats