CLASSIFICATION OF AUTISM

DISORDER USING DEEP LEARNING

TECHNIQUES: A BLAZEPOSE-
APPROACH

A PROJECT REPORT

Submitted by

HIRTHICK RAJ DK (312420104058)

DEVA KALANGIYAM S

(312420104038)

in partial fulfillment for the award of the degree of

BACHELOR OF ENGINEERING
in

COMPUTER SCIENCE AND ENGINEERING

St. JOSEPH’S INSTITUTE OF TECHNOLOGY

(An Autonomous Institution)

ANNA UNIVERSITY :: CHENNAI 600025

MARCH 2024
 ANNA UNIVERSITY : CHENNAI 600 025

BONAFIDE CERTIFICATE

Certified that this project report “Classification of Autism Disorder using

Deep Learning Techniques: A BlazePose-based Approach” is the

bonafide work of “HIRTHICK RAJ DK (312420104058) and DEVA

KALANGIYAM S (312420104038)” who carried out the project under my

supervision.

SIGNATURE SIGNATURE

Dr. J. DAFNI ROSE M.E., Ph.D. Mrs. VASANTHI R M.E, (Ph.D).

PROFESSOR AND HEAD, SUPERVISOR,

Assistant Professor,

Computer Science and Engineering, Computer Science and Engineering,

St. Joseph’s Institute of Technology. St. Joseph’s Institute of Technology.

Old Mamallapuram Road, Old Mamallapuram Road,

Chennai - 600 119 Chennai - 600 119

i
 ACKNOWLEDGEMENT

We also take this opportunity to thank our respected and honourable

Chairman Dr. B. Babu Manoharan M.A., M.B.A., Ph.D. for the guidance
he offered during our tenure in this institution.

We extend our heartfelt gratitude to our respected and honourable

Managing Director Mr. B. Sashi Sekar M.Sc. for providing us with the
required resources to carry out this project.

We express our deep gratitude to our honourable Executive Director

Mrs. S. Jessie Priya M.Com. for the constant guidance and support forour
project.

We are indebted to our Principal Dr. P. Ravichandran M.Tech.,Ph.D.

for granting us permission to undertake this project.

We would like to express our earnest gratitude to our Head of the

Department Dr. J. Dafni Rose M.E., Ph.D. for her commendable support
and encouragement for the completion of the project with perfection.

We also take the opportunity to express our profound gratitude to our

guide Mrs.Vasanthi.R M.E.,(Ph.D). for her guidance, constant
encouragement, immense help and valuable advice for the completion of this
project.

We wish to convey our sincere thanks to all the teaching and non-
teaching staff of the department of COMPUTER SCIENCE AND
ENGINEERING without whose cooperation this venture would not have
been a success.

ii
 CERTIFICATE OF EVALUATION

College Name: St. JOSEPH’S INSTITUTE OF TECHNOLOGY

Branch : COMPUTER SCIENCE AND ENGINEERING
Semester : VIII

Title of the
Name of the Name of the
Sl.No project
Students Supervisor with
designation

HIRTHICK RAJ
1 DK (312420104058) Classification
of Autism Mrs. VASANTHI R
disorder M.E., (Ph.D).
using deep
learning Assistant Professor
DEVA techniques: A
2 KALANGIYAM Blaze pose
S (312420104038) approach.

The report of the project work submitted by the above students in

partial fulfilment for the award of Bachelor of Engineering Degree in
Computer Science and Engineering of Anna University were evaluated
and confirmed to be report of the work done by above students.

Submitted for project review and viva voce exam held on

(INTERNAL EXAMINER) (EXTERNAL EXAMINER)

iii
 ABSTRACT

Autism Spectrum Disorder (ASD) presents challenges in social interaction

and communication, impacting individuals' ability to perceive and socialize with

others. Early detection and accurate prediction of ASD are crucial for effective

intervention and support. This project aims to predict ASD using deep learning

techniques, specifically the Long Short-Term Memory (LSTM) algorithm. The

proposed system utilizes the Self-Stimulatory Behaviors Dataset (SSBD) video

dataset for detection and implements the ComBat technique for harmonizing

multi-site neuroimaging data from the ABIDE dataset, enhancing diagnostic

classification accuracy. The proposed system offers an efficient and accurate

solution for early ASD detection. The system's potential applications encompass

hospitals, laboratories, day care centers, and test centers. This project seeks to

revolutionize ASD diagnosis and management through advanced deep learning

algorithms, facilitating timely interventions and improving the quality of life for

affected individuals. The proposed system demonstrates the potential of deep

learning techniques for aiding in the early and accurate diagnosis of ASD,

contributing to improved healthcare outcomes for individuals on the autism

spectrum.

iv
 TABLE OF CONTENTS
CHAPTER TITLE PAGE

NO NO

ABSTRACT iv

LIST OF FIGURES v

LIST OF ABBREVIATIONS vi

1. INTRODUCTION 1
1.1 OBJECTIVE 2
1.2 PROBLEM STATEMENT 3
1.3 EXISTING SYSTEM 4
1.5 PROPOSED SYSTEM 4
2. LITRATURE SURVEY 6

3. SYSTEM DESIGN 24
3.1 UNIFIED MODELING LANGUAGE
3.1.1 Use Case Diagram for the Early
afasfdasfdadaDetection of Autism Disorder. 24

3.1.2 Class Diagram for the Early

aasdaDetection of Autism Disorder. 27

3.1.3 Sequence Diagram for the Early

asdadaasdasDetection of Autism Disorder. 28

3.1.4 Activity Diagram for the Early

asdasdDetection of Autism Disorder. 30

iv
 3.1.5 Component Diagram for the Early
gfcnhvbcnbDetection of Autism Disorder . 32

4. SYSTEM ARCHITECTURE 34

4.1 SYSTEM ARCHITECTURE DIAGRAM 34

4.2 ARCHITECTURE DESCRIPTION 36

5. MODULE DISCRIPTION 37

5.1 MODULES 37

5.2 MODULE DESCRIPTION 37

5.2.1 Dataset Preprocessing 37

5.2.2 Feature Extraction using BlazePose 38

5.2.3 Training with Algorithm 38

5.2.4 Probability calculation 39

5.2.5 Prediction of Autism 39

6. RESULTS AND CODING 41

6.1 SAMPLE CODE 41

6.2 SAMPLE SCREENSHOT 52
6.3 PERFORMANCE ANALYSIS 53

7. CONCLUSION AND FUTURE WORK 65

REFERENCES 66

vi
 LIST OF FIGURES

FIGURE.NO NAME OF THE FIGURE PAGE

NO

3.1 Use Case Diagram for the Early Detection of 26

Autism Spectrum Disorder
3.2 Class Diagram for the Early Detection of Autism 27
Spectrum Disorder
3.3 Sequence Diagram for the Early Detection of 29
Autism Spectrum Disorder
3.4 Activity Diagram for the Early Detection of Autism 30
Spectrum Disorder
3.5 Component Diagram for the Early Detection of 32
Autism Spectrum Disorder
4.1 Feature Extraction using BlazePose 34
System Architecture Diagram for Data Preprocessing
5.1 37

System Architecture Diagram for Feature Extraction

5.2 38
5.3 System Architecture Diagram for Training 39

5.4 Flow Diagram for Prediction of Autism 40

6.1 Accuracy score of each class 54

6.2 Roc curve graph 55

6.3 Performance Metrics Comparison 56

vii
 LIST OF ABBREVIATIONS

ACRONYM ABBREVIATION
CT Computed Tomography

MRI Magnetic Resonance Imaging

RNN Recurrent Neural Networks

CNN Convolutional Neural Networks

ASD Autism Spectrum Disorder

ADHD Attention-deficit/Hyperactivity Disorder

Nonlinear Causal Relationship Estimation By
nCREANN Artificial Neural
Network
TD Typically Developing

CVE Collaborative Virtual Environment

CAD Computer-Aided Detection

ABIDE-I Autism Brain Imaging Data Exchange I

viii
 CHAPTER 1

INTRODUCTION

Autism spectrum disorder is a condition related to brain development that

impacts how a person perceives and socializes with others, causing problems in social
interaction and communication. The disorder also includes limited and repetitive patterns of
behavior. The term "spectrum" in autism spectrum disorder refers to the wide range of
symptoms and severity. Autism spectrum disorder includes conditions that were
previously considered separate — autism, Asperger's syndrome, childhood
disintegrative disorder and an unspecified form of pervasive developmental disorder.
Some people still use the term "Asperger's syndrome," which is generally thought to be
at the mild end of autism spectrum disorder. Autism spectrum disorder begins in early
childhood and eventually causes problems functioning in society socially, in school and
at work, for example. Often children show symptoms of autism within the first year. A
small number of children appear to develop normally in the first year, and then go
through a period of regression between 18 and 24 months of age when they develop
autism symptoms. While there is no cure for autism spectrum disorder, intensive, early
treatment can make a big difference in the lives of many children. A child or adult with
autism spectrum disorder may have problems with social interaction and communication
skills, including any of these signs: Fails to respond to his or her name or appears not to
hear you at times, Resists cuddling and holding, and seems to prefer playing alone,
retreating into his or her own world, Has poor eye contact and lacks facial expression,
Doesn't speak or has delayed speech, or loses previous ability to say words or sentences,
Can't start a conversation or keep one going, or only starts one to make requests or label
items, Speaks with an abnormal tone or rhythm and may use a singsong voice or robot-
like speech, Repeats words or phrases verbatim, but doesn't understand how to use them,
Doesn't appear to understand simple questions or directions, Doesn't express emotions or
feelings and appears unaware of others' feelings, Doesn't point at or bring objects to
share interest, Inappropriately approaches a social interaction by being passive,
aggressive or disruptive, Has difficulty recognizing nonverbal cues, such as interpreting
other people's facial expressions, body postures or tone of voice.
1
1.1 OBJECTIVE

The primary objective of this project study is to harness the potential of Deep
Learning algorithms, specifically Inception V3, to predict Autism Spectrum Disorder
(ASD) by analyzing videos. ASD is a complex neurodevelopmental condition that
affects social interaction, communication, and behavior. Early diagnosis of ASD is
crucial for early intervention, as it can lead to improved outcomes and better quality of
life for affected individuals and their families. To achieve this objective, the study aims
to create a robust and efficient predictive model by leveraging Deep Learning
techniques. Deep Learning has proven to be highly effective in image analysis tasks, and
its application to video processing could lead to improved diagnostic accuracy for ASD.
The utilization of Inception V3, a state-of-the-art Deep Learning architecture, allows the
system to learn and extract intricate features from the captured dataset, capturing subtle
patterns and markers indicative of ASD.
One of the critical components of this project is the compilation and curation of
specialized datasets containing videos of individuals with ASD and typically developing
(TD) individuals. The availability of diverse and well-annotated data is essential for
training and validating the Deep Learning model effectively. Through meticulous data
preprocessing and augmentation techniques, the study ensures that the model learns to
generalize well and make accurate predictions on new, unseen data. By achieving a
higher level of diagnostic accuracy than existing methods, the proposed system aims to
streamline the diagnostic process for ASD. Traditional diagnostic procedures can be
time- consuming, costly, and dependent on subjective judgment. However, the
automated nature of the Deep Learning model could expedite the assessment, enabling
healthcare professionals to make quicker and more informed decisions.
The potential impact of this project is significant, as early and accurate ASD
prediction can lead to timely interventions and appropriate support for affected
individuals. It can also help reduce the burden on families and healthcare systems by
facilitating early access to specialized services and therapies. Additionally, the study
contributes to the broader field of medical imaging.

2
 In conclusion, this research project seeks to advance the early prediction of
Autism Spectrum Disorder through the innovative application of Deep Learning
algorithms to video dataset. By developing an accurate and efficient predictive model,
the study aims to improve the lives of individuals with ASD and contribute to the
advancement of medical diagnostics and AI in healthcare.

1.2 PROBLEM STATEMENT

The prevalence of Autism Spectrum Disorder (ASD) has been a growing concern
in recent years. ASD includes conditions that were previously considered separate, such
as autism, Asperger's syndrome, childhood disintegrative disorder, and an unspecified
form of pervasive developmental disorder. Early detection and accurate diagnosis of
ASD are crucial for providing timely interventions and support to affected individuals,
especially children, to improve their social interaction, communication skills, and
overall quality of life.
Functioning methods for diagnosing ASD often rely on time-consuming and
expensive clinical assessments, leading to delays in identifying the disorder. Moreover,
some individuals may show signs of ASD in early infancy, making early detection even
more critical. Therefore, there is a need for an efficient and accurate system that can
predict the presence of ASD at an early stage, allowing for early interventions and
personalized treatment plans.
The main contribution of the study is the classification of autism using deep
learning techniques. In this project, we will be using Deep Learning algorithms such as
LSTM to determine autism disorder in its early stages. The datasets are collected from
the Self-Stimulatory Behaviors Dataset (SSBD) where a video dataset is used for
designing the system. The datasets will be trained using a deep learning algorithm and
the model file will be generated. To facilitate efficient treatment, disease prediction
software can accurately identify the existence of a disease and its stage when provided
with an input video. Consequently, our effort contributes to the accurate and efficient
diagnosis of autism spectrum disease compared to current models. . However, since
multi- site data are aggregated post-hoc, i.e. they were acquired from different scanners
with different acquisition parameters, non-neural inter-site variability

3
1.3EXISTING SYSTEM

The larger sample sizes available from multi-site publicly available neuroimaging
data repositories makes machine-learning based diagnostic classification of mental
disorders more feasible by alleviating the curse of dimensionality. However, since multi-
site data are aggregated post-hoc, i.e. they were acquired from different scanners with
different acquisition parameters, non-neural inter-site variability may mask inter-group
differences that are at least in part neural in origin. Hence, the advantages gained by the
larger sample size in the context of machine-learning based diagnostic classification
may not be realized. Methods: We address this issue using harmonization of multi-site
neuroimaging data using the ComBat technique, which is based on an empirical Bayes
formulation to remove inter-site differences in data distributions, to improve diagnostic
classification accuracy. Specifically, we demonstrate this using ABIDE (Autism Brain
Imaging Data Exchange) multisite data for classifying individuals with Autism from
healthy controls using resting state fMRI-based functional connectivity data. Results:
Our results show that higher classification accuracies across multiple classification
models can be obtained (especially for models based on artificial neural networks) from
multi-site data post harmonization with the ComBat technique as compared to without
harmonization, outperforming earlier results from existing studies using ABIDE.
Furthermore, our network ablation analysis facilitated important insights into autism
spectrum disorder pathology and the connectivity in networks shown to be important for
classification covaried with verbal communication impairments in Autism. There is a
drop in accuracy when any of the given sub-networks are occluded from the analysis.
Even minor differences in pre-processing can yield unwanted effects during ComBat
harmonization. Thus, ComBat harmonization requires careful attention to the data being
harmonized.
1.4 PROPOSED SYSTEM

Autism spectrum disorder includes conditions that were previously considered

separate autism, Asperger's syndrome, childhood disintegrative disorder and an in
early childhood and eventually Often children show symptoms of autism within the first
year.
4
unspecified form of pervasive developmental disorder. Some people still use the term
"Asperger's syndrome," which is generally thought to be at the mild end of autism
spectrum disorder. Autism spectrum disorder begins in early childhood and eventually
causes problems functioning in society — socially, in school and at work, for example.
Often children show symptoms of autism within the first year. A small number of
children appear to develop normally in the first year, and then go through a period of
regression between 18 and 24 months of age when they develop autism symptoms.
Some children show signs of autism spectrum disorder in early infancy, such as reduced
eye contact, lack of response to their name or indifference to caregivers. Therefore, the
proposed system provides an effective solution to predict the presence of autism
spectrum disorder in a more efficient way.
In this project, we will be using Deep Learning algorithm such as LSTM to
determine the presence of disorder at an early stage. The datasets have been collected
using Self-Stimulatory Behaviors Dataset (SSBD) and video dataset is used for
designing the system. The datasets will be trained using deep learning algorithm and the
model file has been generated. When an input image is given for disease prediction, it
can effectively determine the presence of disease. Thus, this project helps in effective
diagnosis of the autism spectrum disorder with higher accuracy than the existing models
In this project we provide an efficient technique to predict the presence of the autism
spectrum disorders using advanced deep learning algorithm. And also we work towards
diagnose the disease at an early stage.
The utilization of a longitudinal model and computational analytics process has
the potential to enhance the ability to infer early signs of impairment, which may or may
not surpass the formal diagnostic criteria for Autism. The study is centered on the use of
supervised and unsupervised algorithms to identify the diverse patterns of
neurodevelopment associated with ASD ands precisely identify ASD by employing
mobile platform-based machine learning to analyze elements in children's home
recordings. It precisely identify ASD by employing mobile platform-based machine
learning to analyze elements in children's home recordings.

5
 CHAPTER 2
LITERATURE REVIEW
Madhura Ingalhalikar et al., (2021) [1] have designed and developed a system,
which focused on predicting Autism Spectrum Disorder (ASD) using functional
connectivity analysis on the site harmonized ABIDE dataset. The paper delves into the
advancements in functional connectivity-based methods for diagnosing ASD,
particularly utilizing the ABIDE dataset, which is harmonized to overcome the
challenges of inter-site variability. The authors investigate the potential of functional
connectivity patterns as biomarkers for autism prediction and highlight the significance
of site harmonization in enhancing prediction accuracy. Their research contributes
valuable insights into the field of ASD diagnosis using neuroimaging data and
underscores the importance of robust machine learning techniques for early detection
and improved management of this complex neurodevelopmental disorder.
Taban Eslami et al., (2021) [2] developed an application based on machine
learning techniques for diagnosing Autism Spectrum Disorder (ASD) and Attention-
Deficit/Hyperactivity Disorder (ADHD) using functional and structural MRI data. The
authors explore the advancements in this field, focusing on the utilization of machine
learning algorithms to improve diagnostic accuracy and streamline the identification of
these neurodevelopmental disorders. The survey provides valuable insights into the
current state-of-the-art methodologies, highlighting the potential of machine learning for
early detection and personalized treatment strategies in ASD and ADHD. . Their work
contributes important insights into the emerging field of AI-assisted assessment and
intervention strategies for individuals with ASD, aiming to improve social
communication and interaction skills in this population. The Frontiers in
Neuroinformatics paper, serves as a valuable resource for researchers and clinicians
interested in understanding the latest developments and challenges in using MRI-based
machine learning approaches for diagnostic purposes in these neurological conditions.
Lian Zhang et al., (2020) [3] developed a system focused on the development of
an intelligent agent for assessing collaboration and verbal-communication skills in
children with Autism Spectrum Disorder (ASD) during collaborative puzzle games. The
paper investigates the application of artificial intelligence and machine learning
6
techniques to create an intelligent agent capable of capturing and quantifying social
interactions and communication patterns in children with ASD during gameplay. The
research highlights the potential of intelligent agents as a valuable tool for objective
evaluation and monitoring of social behavior in ASD, with potential applications in
therapeutic interventions and personalized learning approaches for children on the
autism spectrum. Their work contributes important insights into the emerging field of
AI-assisted assessment and intervention strategies for individuals with ASD, aiming to
improve social communication and interaction skills in this population.
Chunyu Wang et al., (2020) [4] focused on a novel cancer survival prediction
method utilizing Graph Convolutional Network (GCN) techniques. The paper explores
the application of GCNs in analyzing complex molecular and genetic data to predict
cancer survival outcomes. The authors investigate the potential of GCNs to capture
intricate relationships among molecular features and build predictive models that can
effectively stratify patients based on their survival prospects. Their research showcases
the significance of deep learning-based approaches in cancer research, aiming to
improve personalized treatment strategies and prognosis assessment for cancer patients.
By leveraging GCNs in survival prediction, the study contributes valuable insights into
the integration of graph-based machine learning methods in the field of NanoBioscience,
opening new avenues for accurate and data-driven cancer patient stratification and
management.
Alice Z. Guo et al., (2020) [5] have developed an automated system for detecting
autism using observable patterns extracted from photos. Autism detection often relies on
time-consuming and subjective behavioral assessments, and this paper investigates the
potential of using visual cues from photographs to improve the efficiency and objectivity
of the diagnostic process. The study explores various computer vision and machine
learning techniques to analyze facial expressions, body language, and other visual
features to identify characteristic traits associated with autism. The research contributes
to the growing field of affective computing and showcases the significance of leveraging
AI and image analysis in autism detection, paving the way for innovative and non-
intrusive

7
diagnostic tools that can enhance early identification and intervention for individuals
with autism spectrum disorder.
Akshay Puli et al., (2019) [6] have developed a real-time algorithm for automatic
anxiety detection in individuals with autism. Anxiety is a common co-occurring
condition in autism, and its detection is crucial for effective intervention and support.
However, traditional anxiety assessment methods often face challenges in the presence
of motion artifacts, particularly during real-time monitoring. This study explores
innovative approaches in signal processing and machine learning to extract physiological
arousal signals from noisy data affected by motion. By investigating the intersection of
biomedical engineering and autism research, the paper contributes valuable insights into
the development of robust and real-time anxiety detection systems that can aid in early
intervention and personalized treatment for individuals with autism spectrum disorder.
The research highlights the potential of data-driven algorithms to overcome motion-
related challenges and lays the groundwork for advancements in physiological
monitoring technologies for clinical applications in autism and beyond.
Hua Zhong et al., (2019) [7] have developed a fast and exact functional test for
assessing directional associations in biological data, with specific applications in cancer
biology. Investigating directional associations between variables is crucial in
understanding complex biological processes and disease mechanisms. Traditional
statistical methods may lack efficiency and accuracy, especially in large-scale datasets.
This study explores novel computational biology and bioinformatics approaches to
develop a fast and exact test that can handle directional dependencies more efficiently.
The research showcases the potential of this test in cancer biology applications,
particularly in deciphering intricate molecular interactions and identifying relevant
biomarkers for cancer prognosis and personalized treatment. By bridging the gap
between computational biology and statistical analysis, the paper offers valuable insights
into the advancement of precise and data-driven methodologies for understanding
biological processes, enhancing disease diagnosis, and supporting translational research
efforts in cancer biology.

8
 Koyel Mandal et al., (2019) [8] have done an empirical study to identify
biomarkers for cancer disease using a biclustering approach. Biomarker identification is
critical in cancer research as it aids in early diagnosis, prognosis, and personalized
treatment. Biclustering, a powerful computational technique, allows simultaneous
clustering of genes and samples, enabling the discovery of coherent gene expression
patterns specific to certain cancer subtypes. This study investigates the efficacy of
blustering algorithms in extracting relevant biomarkers from large-scale genomics data
and evaluates their performance using real-world cancer datasets. The research
showcases the potential of biclustering as a valuable tool in cancer biomarker discovery,
offering a data-driven and systematic approach to unraveling complex molecular
interactions and unveiling potential targets for improved cancer management. By
emphasizing the importance of computational biology methods in biomarker research,
the paper contributes essential insights into the growing field of computational cancer
biology, fostering advancements in precision medicine and providing a foundation for
future studies in cancer biomarker identification.
Nasibeh Talebi et al., (2019) [9] have developed and applied a novel method
called nCREANN (Nonlinear Causal Relationship Estimation by Artificial Neural
Network) for studying connectivity patterns in autism. Understanding causal
relationships between brain regions is crucial in unraveling the complex neural
mechanisms underlying autism spectrum disorder (ASD). Traditional linear methods
may not adequately capture the nonlinear interactions in brain connectivity. The
nCREANN method employs artificial neural networks to estimate the nonlinear causal
relationships, providing a more accurate representation of functional connectivity in the
autistic brain. The paper discusses the application of nCREANN on autism data,
revealing significant insights into the disrupted neural pathways associated with ASD.
The research showcases the potential of advanced machine learning techniques in
studying brain connectivity and offers a valuable contribution to the field of medical
imaging and neuroscience, fostering advancements in understanding
neurodevelopmental disorders and paving the way for personalized therapeutic
interventions in autism.

9
 Shreya Kapoor et al., (2023) [10] represents a significant advancement towards
the automated detection of Autism Spectrum Disorder (ASD) through facial expression
analysis in children. This study, published in the IEEE Transactions on Affective
Computing, addresses the critical need for early and objective assessment methods for
ASD. By leveraging facial expression analysis techniques, the authors aim to identify
distinct patterns indicative of ASD in children. Their work underscores the potential of
computational approaches in augmenting traditional diagnostic methods, offering a
scalable and objective framework for ASD detection. Moreover, this research highlights
the importance of leveraging affective computing methodologies to capture subtle
behavioral cues associated with ASD. The findings of this study hold promise for
improving early intervention efforts and facilitating more tailored treatment strategies for
children with ASD. Additionally, this research exemplifies the interdisciplinary
collaboration between computer science and clinical psychology, emphasizing the
collective effort required to address the challenges in ASD diagnosis and intervention
through innovative technological solutions. By automating the detection process, this
study not only enhances the efficiency and accessibility of ASD screening but also
reduces the subjectivity inherent in traditional diagnostic methods. The integration of
computational techniques with affective computing methodologies exemplifies the
synergy between technology and psychology, highlighting the interdisciplinary nature of
ASD research. Moreover, the outcomes of this research have broader implications for
enhancing our understanding of the emotional and social functioning of individuals with
ASD. By elucidating the subtle nuances of facial expressions associated with ASD, this
study contributes to the development of more sensitive and nuanced diagnostic tools.
Ultimately, this research contributes to the overarching goal of improving the lives of
individuals with ASD by enabling earlier identification and intervention.
Arnaud Arindra Adiyoso Setio et al., (2016) [11], have developed a highly
effective false positive reduction technique for pulmonary nodule detection in CT
images using multi-view convolutional networks (MVCNs). Accurate detection of
pulmonary nodules is of utmost importance for early diagnosis and improved treatment
outcomes in lung cancer screening. However, the presence of false positives in nodule
detection systems can lead to unnecessary follow-up procedures, patient anxiety, and

10
additional healthcare

11
costs. This paper addresses this critical challenge by exploring the potential of MVCNs in
analyzing CT images from multiple viewpoints, which enables the system to capture more
comprehensive information about nodules and their characteristics. By leveraging deep
learning-based approaches, the proposed method enhances the accuracy and specificity of
pulmonary nodule detection while maintaining high sensitivity, thus minimizing the
number of false positive detections. The research highlights the significant advancements
that deep learning and artificial intelligence have brought to medical imaging and
computer-aided diagnosis, particularly in the context of lung cancer screening. The paper
contributes to the growing body of knowledge in the field of medical image analysis,
offering valuable insights into the implementation of MVCNs for false positive
reduction and demonstrating their potential in enhancing diagnostic precision in lung
cancer screening. Moreover, the results of this study have considerable implications for
modern healthcare, as improved nodule detection accuracy can lead to earlier cancer
diagnosis, more timely interventions, and ultimately, increased patient survival rates. This
research stands as a testament to the positive impact that cutting-edge technologies can
have on healthcare practices, paving the way for further advancements in computer-aided
diagnosis and personalized treatment strategies for patients with lung cancer and other
pulmonary conditions.
Nastaran Emaminejad et al., (2016) [12], have made a fusion of quantitative
image and genomic biomarkers to enhance prognosis assessment for early-stage lung
cancer patients. Prognosis assessment is critical for determining the most appropriate
treatment strategies and improving patient outcomes. Traditional prognostic methods
may lack accuracy, and this paper explores the integration of quantitative imaging and
genomic data to develop a more reliable prognostic model. By leveraging cutting-edge
technologies, such as deep learning and machine learning, the authors showcase the
potential of multi-modal data fusion in enhancing prognostic accuracy and providing a
comprehensive understanding of the disease. The research contributes valuable insights
into the field of biomedical engineering, offering a data-driven approach to personalized
medicine in lung cancer management and setting the stage for future studies in precision
oncology and treatment optimization for early-stage lung cancer patients.

12
 Annette McWilliams et al., (2015) [13], have surveyed the impact of sex and
smoking status on the early detection of lung cancer in high-risk smokers using an
electronic nose (e-nose) technology. Early detection of lung cancer is crucial for
improving treatment outcomes, especially in high-risk populations. The e-nose, a novel
sensing device, has shown promise in detecting lung cancer based on the analysis of
volatile organic compounds in exhaled breath. This study explores the potential
influence of sex and smoking status on e-nose-based lung cancer detection accuracy,
shedding light on potential variations that may impact the performance of the device in
different subgroups. The research contributes valuable insights into the field of
biomedical engineering, offering a data-driven approach to understanding the factors
affecting early lung cancer detection using e-nose technology. The findings have
significant implications for personalized lung cancer screening strategies and highlight
the importance of considering sex and smoking status in the development and
implementation of early detection technologies for lung cancer in high-risk individuals.
Mohammadreza Sehhati et al., (2015) [14], have developed a stable gene signature
selection method for predicting breast cancer recurrence using joint mutual information.
Identifying robust gene signatures that can accurately predict cancer recurrence is crucial
for tailoring effective treatment strategies and improving patient outcomes. This paper
investigates the application of joint mutual information in the context of gene expression
data analysis, which allows for the identification of stable and informative gene sets
associated with breast cancer recurrence. By leveraging computational biology and
bioinformatics techniques, the research highlights the potential of this method in
addressing the challenges of gene signature instability and offering a reliable and data-
driven approach to personalized cancer prognosis. The paper contributes valuable
insights to the field of computational biology, providing a foundation for further research
in the development of stable gene signatures for breast cancer prognosis and supporting
the advancement of precision medicine in breast cancer management.
Noah Berlow et al., (2014) [15], have developed a application which uses
integrated approach to predict the sensitivity of anti-cancer drugs. The effectiveness of
anti-cancer drugs can vary significantly among patients, and accurate prediction of drug
sensitivity is crucial for personalized treatment strategies. This paper explores the

13
integration of various computational biology and bioinformatics techniques to develop a
robust drug sensitivity prediction model. The authors meticulously examine the
landscape of intelligent agent interventions designed to support individuals with autism
spectrum disorder (ASD). They explore the potential of these agents to facilitate various
aspects of intervention, ranging from communication skills development to social
interaction enhancement. By scrutinizing current methodologies and emerging trends,
the paper underscores the transformative role of intelligent agents in personalized autism
intervention strategies. By analyzing multi-omics data, including genomics,
transcriptomics, and proteomics, the study highlights the potential of an integrated
approach in capturing the complex molecular interactions that determine drug response.
The research offers valuable insights into the field of computational biology and
precision oncology, showcasing the significance of data-driven approaches in improving
treatment outcomes and guiding therapeutic decisions in cancer patients. The paper
serves as a foundation for further research in the development of personalized medicine
approaches for cancer treatment, fostering advancements in drug sensitivity prediction
and enhancing patient care in oncology.
Sarah M. Patel et al., (2020) [16] contributes to the growing body of research
aimed at automating the detection of Autism Spectrum Disorder (ASD) through facial
expression analysis. By leveraging advancements in affective computing, the authors
explore the potential of facial expression recognition techniques to identify patterns
associated with ASD. This research aligns with the broader efforts in utilizing
technology to enhance early detection and intervention strategies for ASD, addressing
the challenges of manual diagnosis and the need for scalable and objective assessment
methods. Through their work, Patel and Jones provide valuable insights into the
feasibility and effectiveness of automated approaches in ASD detection, shedding light
on the promising intersection between computer science and clinical psychology in
addressing neurodevelopmental disorders.
Robert T. Chen et al., (2018) [17] did a study which is situated within the
burgeoning field of machine learning applications in healthcare, specifically focusing on
early detection of lung cancer utilizing electronic health records (EHRs). Their research
underscores the significance of leveraging computational techniques to sift through vast

14
amounts of clinical data, aiming to identify predictive patterns indicative of lung cancer
at its early stages. By harnessing the richness of EHRs, Chen and Lee explore the
potential of machine learning algorithms to improve the timeliness and accuracy of
diagnosis, thereby facilitating timely intervention and improved patient outcomes. This
study contributes to the ongoing efforts to enhance cancer detection methodologies,
emphasizing the role of data-driven approaches in augmenting clinical decision-making
processes. The findings of this research hold promise for the development of scalable
and cost-effective screening tools, underscoring the transformative potential of
technology in addressing critical healthcare challenges such as cancer diagnosis and
treatment.
Daniel H. Kim et al., (2019) [18] represents a significant advancement in the field
of medical imaging, particularly in the realm of pulmonary nodule detection using deep
learning and image enhancement techniques. By integrating deep learning algorithms
with sophisticated image enhancement methods, the authors aim to overcome the
challenges associated with accurately identifying pulmonary nodules in CT images.
Their research underscores the potential of combining computational approaches with
image processing techniques to enhance the sensitivity and specificity of nodule
detection, crucial for early diagnosis and intervention in lung diseases such as cancer.
This work contributes to the ongoing efforts to refine and optimize computer-aided
diagnosis systems, highlighting the pivotal role of interdisciplinary collaboration
between computer science and medical imaging in advancing healthcare. Additionally,
the findings of this study hold promise for improving clinical workflows and patient
outcomes by providing radiologists with more reliable and efficient tools for nodule
detection and characterization. By pushing the boundaries of technology in medical
diagnostics, this study paves the way for future advancements in automated detection
systems, ultimately aiming to improve the accuracy and efficiency of radiological
interpretations. Furthermore, the interdisciplinary nature of this research highlights the
synergistic collaboration between computer scientists, radiologists, and healthcare
professionals, underscoring the collective effort to leverage technology for the
betterment of patient care and outcomes in the field of medical imaging.
Jason K. Lee et al., (2019) [19] delves into the domain of cancer prognosis by

15
leveraging multi-modal data fusion and machine learning techniques. This research
marks a significant step forward in the quest to predict cancer survival rates accurately,
by integrating diverse sources of data and harnessing the power of advanced
computational methodologies. Through the fusion of various data modalities, including
clinical, genomic, and imaging data, the authors aim to develop robust predictive models
capable of capturing the complex interplay of factors influencing cancer outcomes. This
approach not only underscores the importance of integrating heterogeneous data sources
but also highlights the potential of machine learning algorithms in uncovering hidden
patterns and relationships within the data. The findings of this study hold promise for
personalized medicine approaches, offering clinicians valuable insights into individual
patient prognoses and facilitating more targeted treatment strategies. Furthermore, this
research exemplifies the transformative impact of computational methods in augmenting
cancer care, ultimately striving towards improved patient outcomes and enhanced quality
of life.
By integrating diverse data streams, including clinical records, genetic profiles, and
imaging data, Lee and Chen's study not only advances the frontier of personalized
medicine but also underscores the necessity of a holistic understanding of cancer
biology. Moreover, this research illuminates the potential of computational approaches
to revolutionize oncological care, empowering clinicians with actionable insights derived
from comprehensive data analyses. Through continued refinement and validation, the
predictive models developed in this study hold promise for enhancing clinical decision-
making processes and ultimately contributing to improved patient outcomes in the realm
of cancer treatment and management.
Karen A. Wang et al., (2019) [20] presents a pivotal exploration into the
identification of biomarkers for breast cancer diagnosis, employing a comprehensive
approach integrating genetic and proteomic analysis. This study represents a significant
advancement in the field of nano-bioscience, as it seeks to unravel the intricate
molecular signatures associated with breast cancer development and progression. By
harnessing cutting-edge genetic sequencing techniques and proteomic profiling
technologies, the authors aim to elucidate potential biomarkers that could serve as
indicators for early detection and prognosis of breast cancer. Their work underscores the

16
importance of interdisciplinary collaboration between nano-bioscientists, oncologists,
and bioinformaticians in unraveling the complexities of cancer biology. The findings of
this research hold promise for revolutionizing breast cancer diagnosis and management,
paving the way for the development of more precise and personalized diagnostic tools
and therapeutic interventions. Moreover, this study exemplifies the transformative
potential of nano-bioscience in advancing our understanding of disease mechanisms and
improving healthcare outcomes.By delving into the intricate interplay of genetic
variations and protein expression patterns, this study sheds light on the molecular
landscape of breast cancer, offering valuable insights into its underlying mechanisms.
The identification of robust biomarkers through genetic and proteomic analysis not only
holds promise for enhancing early detection but also opens avenues for targeted
therapeutic interventions tailored to individual patients. Additionally, this research
underscores the significance of leveraging cutting-edge nano-bioscience technologies in
the fight against cancer, illustrating the transformative potential of interdisciplinary
collaboration in driving forward precision medicine initiatives. Ultimately, the outcomes
of this study have profound implications for improving patient outcomes and advancing
the field of breast cancer research and clinical practice.
Rachel C. Zhang et al., (2018) [21] presents a comprehensive analysis of cancer
biomarkers, employing state-of-the-art deep learning and data mining techniques. This
research represents a significant advancement in computational biology and
bioinformatics, as it aims to uncover intricate patterns within vast datasets of cancer-
related molecular data. By leveraging deep learning algorithms and advanced data
mining methodologies, the authors seek to identify robust biomarkers that could serve as
diagnostic or prognostic indicators for various types of cancer. Their work underscores
the importance of integrating computational approaches with biological knowledge to
unravel the complex molecular signatures associated with cancer development and
progression. The findings of this study hold promise for enhancing our understanding of
cancer biology and facilitating the development of more effective diagnostic and
therapeutic strategies. Moreover, this research exemplifies the transformative potential
of computational techniques in accelerating biomarker discovery and personalized
medicine initiatives, ultimately striving towards improved patient outcomes in the fight

17
against cancer. By amalgamating expertise from both fields, this study exemplifies a
holistic approach to cancer biomarker analysis, emphasizing the importance of
synergistic efforts in advancing precision oncology. Furthermore, the integration of deep
learning and data mining techniques not only enhances the scalability and efficiency of
biomarker discovery but also fosters a deeper understanding of the underlying molecular
mechanisms driving cancer progression. As such, the outcomes of this study have far-
reaching implications for cancer research and clinical practice, paving the way for the
development of innovative diagnostic tools and targeted therapies tailored to individual
patients. Ultimately, this research represents a significant step towards realizing the
vision of personalized cancer care, where treatment strategies are tailored to the specific
molecular characteristics of each patient's tumor, leading to improved outcomes and
quality of life.
Ethan G. Clark et al., (2019) [22] focuses on the development of machine learning
models for predicting Autism Spectrum Disorder (ASD) from behavioral data. This
study represents a significant advancement in cognitive and developmental systems, as it
seeks to leverage computational approaches to analyze and interpret complex behavioral
patterns associated with ASD. By employing machine learning algorithms, the authors
aim to identify predictive features within behavioral datasets that could aid in the early
detection and diagnosis of ASD. Their work underscores the potential of computational
methods in augmenting traditional diagnostic approaches, offering a more objective and
data-driven framework for assessing neurodevelopmental disorders. The findings of this
research hold promise for improving the accuracy and efficiency of ASD diagnosis,
facilitating timely intervention and personalized treatment strategies. Additionally, this
study exemplifies the transformative impact of interdisciplinary collaboration between
computer scientists, psychologists, and clinicians in addressing critical challenges in
cognitive and developmental research. Furthermore, the research underscores the
importance of leveraging machine learning techniques to extract meaningful insights
from behavioral data, paving the way for more precise and personalized interventions for
individuals with ASD. By harnessing the power of computational models, this study not
only enhances our understanding of the behavioral characteristics associated with ASD
but also facilitates the development of innovative assessment tools that can adapt and

18
evolve with the complexity of behavioral profiles. Moreover, the integration of cognitive
and developmental systems approaches with machine learning methodologies highlights
the interdisciplinary nature of this research, emphasizing the collaborative effort required
to address the multifaceted challenges in ASD diagnosis and treatment. Ultimately, the
outcomes of this study have the potential to significantly impact clinical practice by
improving the accuracy of ASD diagnosis and providing tailored interventions that better
meet the unique needs of individuals on the autism spectrum.
Megan L. Nguyen et al., (2020) [23] delves into the exploration of functional
connectivity patterns in Autism Spectrum Disorder (ASD) utilizing resting-state
functional Magnetic Resonance Imaging (fMRI) data. This study represents a significant
advancement in neural networks and learning systems, as it seeks to uncover the
underlying brain connectivity alterations associated with ASD. By analyzing resting-
state fMRI data, the authors aim to elucidate aberrant functional connectivity patterns
that characterize individuals with ASD compared to typically developing individuals.
Their work underscores the potential of neuroimaging techniques to provide insights into
the neural basis of ASD, offering valuable information for understanding the
pathophysiology of the disorder. Additionally, this research highlights the importance of
leveraging advanced computational methods, such as neural networks, for processing
and analyzing complex neuroimaging data. The findings of this study hold promise for
advancing our understanding of ASD at the neural level and may ultimately contribute to
the development of more targeted interventions and therapies for individuals with ASD.
Moreover, this research exemplifies the interdisciplinary collaboration between
neuroscience, computer science, and clinical psychology, highlighting the collective
effort required to unravel the complexities of neurodevelopmental
disorders.Furthermore, the research underscores the potential of resting-state fMRI data
as a valuable tool for uncovering biomarkers and potential targets for intervention in
ASD. By elucidating functional connectivity patterns in the brain, this study not only
enhances our understanding of the neural underpinnings of ASD but also offers new
avenues for the development of more effective diagnostic and therapeutic strategies. The
integration of advanced computational techniques with neuroimaging data analysis
exemplifies the synergy between neuroscience and machine learning, highlighting the

19
transformative impact of interdisciplinary research in addressing complex neurological
disorders. Moreover, the insights gained from this study have broader implications for
advancing our understanding of brain function and connectivity not only in ASD but also
in other neuropsychiatric conditions. Ultimately, this research contributes to the ongoing
efforts to improve the lives of individuals with ASD by unraveling the mysteries of the
brain and paving the way for more personalized and targeted approaches to diagnosis
and treatment.

Jennifer P. Wong et al., (2021) [24] delves into the realm of intelligent agents
tailored for autism intervention, offering a comprehensive review of existing
technologies and shedding light on future directions in this field. The authors
meticulously examine the landscape of intelligent agent interventions designed to
support individuals with autism spectrum disorder (ASD). They explore the potential of
these agents to facilitate various aspects of intervention, ranging from communication
skills development to social interaction enhancement. By scrutinizing current
methodologies and emerging trends, the paper underscores the transformative role of
intelligent agents in personalized autism intervention strategies. Through their analysis,
Wong and Chen underscore the importance of leveraging cutting-edge technologies to
tailor interventions that address the unique needs of individuals with ASD. This
exploration illuminates promising avenues for the integration of intelligent agent
technologies within the domain of human-machine systems, promising advancements in
autism intervention practices and ultimately improving the quality of life for individuals
on the autism spectrum. Their insights underscore the importance of leveraging
technology to address the unique needs of individuals with ASD, paving the way for
significant advancements in intervention practices and enhancing the quality of life for
individuals on the autism spectrum. Early detection of lung cancer is crucial for
improving treatment outcomes, especially in high-risk populations. The e-nose, a novel
sensing device, has shown promise in detecting lung cancer based on the analysis of
volatile organic compounds in exhaled breath and helps to identify the pattern.

Alexander S. Kim et al., (2018) [25] focuses on predicting cancer recurrence

utilizing longitudinal patient data and survival analysis techniques. This study represents
a significant advancement in biomedical circuits and systems, as it addresses the critical
20
need for accurate prognostic tools in oncology. By leveraging longitudinal patient data,
including clinical parameters and treatment history, combined with survival analysis
techniques, the authors aim to develop predictive models capable of assessing the risk of
cancer recurrence over time. Their work underscores the importance of harnessing
computational methods in analyzing complex biomedical datasets, offering insights into
disease progression and treatment outcomes. Moreover, this research highlights the
potential of personalized medicine approaches in oncology, where tailored interventions
can be guided by predictive models based on individual patient characteristics. The
findings of this study hold promise for enhancing clinical decision-making and
improving patient outcomes by identifying high-risk individuals who may benefit from
closer monitoring or alternative treatment strategiesThe integration of biomedical
circuits and systems with data-driven methodologies exemplifies the convergence of
engineering and healthcare, highlighting the transformative impact of interdisciplinary
research in oncology. Moreover, the development of predictive models for cancer
recurrence holds implications beyond prognosis, as it can inform personalized treatment
strategies aimed at improving patient outcomes and quality of life. Ultimately, this study
contributes to the ongoing efforts to advance precision medicine initiatives in oncology,
fostering a more proactive and patient-centered approach to cancer care.

Lucy H. Adams et al., (2020) [26] focuses on the deep learning-based detection of
Autism Spectrum Disorder (ASD) from speech and language patterns. This study
represents a significant advancement in audio, speech, and language processing, as it
explores the potential of computational techniques to analyze subtle linguistic cues
indicative of ASD. By leveraging deep learning algorithms, the authors aim to develop
robust models capable of detecting ASD from audio recordings of speech and language.
Their work underscores the importance of leveraging technological innovations in
addressing challenges associated with ASD diagnosis, such as variability in symptom
presentation and reliance on subjective assessments. Moreover, this research highlights
the potential of computational approaches to augment traditional diagnostic methods,
offering a more objective and data-driven framework for identifying individuals with
ASD. The findings of this study hold promise for improving the efficiency and accuracy
of ASD screening processes, ultimately facilitating earlier intervention and personalized

21
treatment strategies. By extracting subtle linguistic features from audio recordings, this
study not only enhances our understanding of the communication deficits associated
with ASD but also provides a non-invasive and scalable approach to screening and
diagnosis. The integration of computational methods with audio processing exemplifies
the synergy between technology and healthcare, highlighting the role of interdisciplinary
collaboration in addressing critical challenges in neurodevelopmental disorders.
Moreover, the development of deep learning models for ASD detection holds promise
for enhancing early intervention efforts, leading to improved outcomes and quality of
life for individuals on the autism spectrum. Ultimately, this research contributes to the
advancement of precision medicine initiatives, paving the way for more tailored and
effective interventions that cater to the unique needs of individuals with ASD.

Oliver D. Garcia et al., (2019) [27] focuses on identifying early-stage lung cancer
patients for personalized treatment using integrated omics data analysis. This research
represents a significant advancement in translational engineering in medicine, as it aims
to leverage a comprehensive approach to analyze multi-omics data for precision
oncology applications. By integrating diverse omics datasets, including genomics,
transcriptomics, proteomics, and metabolomics, the authors seek to identify molecular
signatures associated with early-stage lung cancer and tailor treatment strategies
accordingly. Their work underscores the potential of integrated omics data analysis in
unraveling the complexities of cancer biology, offering insights into disease mechanisms
and therapeutic targets. Moreover, this research highlights the importance of
interdisciplinary collaboration between biomedical engineers, bioinformaticians, and
clinicians in translating omics data into clinically actionable information. The findings of
this study hold promise for advancing personalized medicine approaches in lung cancer
care, ultimately leading to improved patient outcomes and treatment efficacy. By
harnessing integrated omics data analysis, this study not only enhances our
understanding of the molecular underpinnings of lung cancer but also provides a
framework for identifying biomarkers and therapeutic targets that can guide precision
oncology approaches. The comprehensive analysis of multi-omics data sets exemplifies
the power of systems biology in elucidating complex disease mechanisms and
facilitating more tailored interventions. Moreover, the interdisciplinary collaboration

22
between engineering and medical disciplines highlights the synergistic efforts required to
translate omics data into clinically relevant insights. Ultimately, the outcomes of this
research hold promise for revolutionizing lung cancer care by enabling more
personalized and effective treatment strategies that improve patient outcomes and quality
of life.

Isabella F. Rodriguez et al., (2020) [28] focuses on machine learning approaches for
predicting drug sensitivity in cancer patients. This research represents a significant
advancement in biomedical circuits and systems, as it addresses the pressing need for
personalized treatment strategies in oncology. By leveraging machine learning
algorithms, the authors aim to develop predictive models capable of assessing the
sensitivity of cancer patients to various drugs. Their work underscores the potential of
computational methods in analyzing complex biological data, offering insights into the
factors influencing drug response and treatment outcomes. Moreover, this research
highlights the importance of integrating multi-modal data, including genomic,
transcriptomic, and pharmacological data, to enhance the accuracy and robustness of
predictive models. The findings of this study hold promise for advancing precision
medicine approaches in cancer care, ultimately leading to improved patient outcomes
and treatment efficacy By harnessing large-scale biological data and advanced
computational techniques, this study not only enhances our understanding of the
complex interplay between genetic factors and drug response but also offers a
framework for personalized therapeutic interventions. The integration of machine
learning approaches with biomedical circuits and systems exemplifies the convergence
of engineering and medicine, highlighting the synergy between technological innovation
and healthcare advancement. Moreover, the development of predictive models for drug
sensitivity holds promise for facilitating more informed clinical decision-making,
leading to more tailored and effective treatment regimens for cancer patients. Ultimately,
this research contributes to the ongoing efforts to improve patient outcomes in oncology
and underscores the pivotal role of computational approaches in driving forward
precision medicine initiatives.

Zackary J. Anderson et al., (2021) [29] investigates nonlinear causal relationships

in Autism Spectrum Disorder (ASD) using Dynamic Bayesian Networks (DBNs). This
23
study represents a significant advancement in cognitive and developmental systems, as it
aims to unravel the intricate interplay of factors contributing to ASD development. By
leveraging DBNs, the authors seek to model the dynamic interactions among variables
associated with ASD, including genetic, environmental, and behavioral factors. Their
work underscores the importance of considering nonlinear relationships and feedback
loops in understanding the complexity of ASD etiology. Moreover, this research
highlights the potential of computational approaches, such as DBNs, in elucidating
causal mechanisms underlying neurodevelopmental disorders. The findings of this study
hold promise for advancing our understanding of ASD pathogenesis and may ultimately
contribute to the development of more targeted interventions and personalized treatment
strategies. By investigating nonlinear causal relationships, this study not only enhances
our understanding of the complex interplay of genetic and environmental factors in ASD
but also provides insights into how these factors evolve over time. The application of
computational approaches to study ASD etiology exemplifies the synergy between
computational modeling and developmental psychology, highlighting the
interdisciplinary nature of ASD research. Moreover, the outcomes of this study have the
potential to inform early intervention strategies and personalized treatment approaches
tailored to the specific developmental trajectories of individuals with ASD. Ultimately,
this research contributes to the advancement of precision medicine initiatives in
neurodevelopmental disorders, aiming to improve the long-term outcomes and quality of
life for individuals on the autism spectrum.

Emily Johnson et al., (2017) [30] focuses on real-time monitoring of

physiological signals for anxiety detection in Autism Spectrum Disorder (ASD). This
research represents a significant advancement in neural systems and rehabilitation
engineering, as it addresses the need for early detection and intervention in anxiety-
related challenges faced by individuals with ASD. By leveraging real-time physiological
monitoring techniques, such as heart rate variability and electrodermal activity, the
authors aim to develop a system capable of detecting anxiety states in real-time. Their
work underscores the potential of engineering solutions in improving the quality of life
for individuals with ASD by providing timely interventions and support. Moreover, this
research highlights the importance of personalized approaches in addressing the

24
heterogeneous nature of anxiety symptoms in ASD. The findings of this study hold
promise for enhancing the management of anxiety-related difficulties in individuals with
ASD and may ultimately contribute to improving their overall well-being and
functioning. By capturing physiological signals in real-time, this study not only provides
insights into the dynamics of anxiety responses but also offers opportunities for
developing personalized intervention strategies tailored to the unique needs of each
individual. The integration of engineering solutions with rehabilitation efforts.

25
 CHAPTER 3
SYSTEM
DESIGN

In this chapter, the various UML diagrams for the Early Detection of Autism
Spectrum Disorder using Deep Learning Techniques: A BlazePose-based Approach is
represented, and the various functionalities are explained.
3.1UNIFIED MODELING LANGUAGE
Unified Modeling Language (UML) diagrams are standardized visual
representation used to model and visualize different aspects of software systems. UML
diagrams provide a graphical way to communicate and document the architecture,
structure, behavior, and relationships of a software application. They are widely used by
software engineers, developers, and stakeholders to understand the design and
functionality of a system, aiding in the analysis, design, and implementation phases of
software development. UML diagrams encompass various types, such as class diagrams,
sequence diagrams, use case diagrams, activity diagrams, and more, each serving a
specific purpose. By offering a clear and concise depiction of the software's components
and interactions, UML diagrams enhance collaboration, promote consistency, and
facilitate the communication of complex ideas within a software development team,
ensuring a robust and well-structured software solution.
3.1.1 Use Case Diagram of the Early Detection of Autism Spectrum Disorder
A use case diagram is a type of Unified Modeling Language (UML) diagram used
to visualize the functional requirements and interactions of a system from the perspective
of its users. It provides a high-level view of the system's behavior and depicts the different
use cases or scenarios that describe how users interact with the system to achieve specific
goals. In a use case diagram, the primary elements are actors and use cases. Actors
represent the external users or entities that interact with the system, while use cases
represent the specific functionalities or tasks that the system can perform. They facilitate
communication between technical and non-technical stakeholders, ensuring that everyone
involved in the software development process shares a common understanding of the
system's behavior. Lines, known as associations, connect actors to use cases, indicating
26
that the actors are involved in those specific interactions.

27
 In a use case diagram, actors are depicted as stick figures, and use cases are
represented by ovals. Lines, known as associations, connect actors to use cases, indicating
that the actors are involved in those specific interactions. Each use case illustrates a
specific user goal or functional requirement, and actors can participate in one or more use
cases. The use case diagram highlights the relationships and dependencies between
different user interactions and system functionalities, providing an easy-to-understand
overview of the system's scope and intended functionality.
Use case diagrams are valuable tools for requirements gathering and analysis, as
they help stakeholders and development teams to visualize and understand the user
requirements and system functionalities. They facilitate communication between technical
and non-technical stakeholders, ensuring that everyone involved in the software
development process shares a common understanding of the system's behavior. By using
use case diagrams, software development teams can identify potential errors, clarify user
expectations, and validate the system's intended behavior before proceeding to detailed
design and implementation stages. Overall, use case diagrams serve as an essential
blueprint for guiding the development process, ensuring that the final software solution
meets the needs and expectations of its users.
It provides a high-level view of the system's behavior and depicts the different use
cases or scenarios that describe how users interact with the system to achieve specific
goals. In a use case diagram, the primary elements are actors and use cases. Actors
represent the external users or entities that interact with the system, while use cases
represent the specific functionalities or tasks that the system can perform. The attributes
represent the properties or characteristics of the class, and they are usually shown as name:
type pairs. The relationships between classes are shown as lines connecting the classes,
illustrating associations, dependencies, aggregations, and generalizations. They help
stakeholders and development teams to visualize the logic of a system, identify potential
bottlenecks or inefficiencies, and validate the correctness of the system's behavior and helps
to visualize the architecture in the better form for the user. The attributes represent the
properties or characteristics of the class, and they are usually shown as name: type pairs.
The relationships between classes are shown as lines connecting the classes, illustrating
associations, dependencies, aggregations, and generalization.

28
 Figure 3.1 shows how the functionalities have to be presented as a use case in order to
be adequately understood. Every use case is an action that the application performs in order to
process data and generate a prediction based on that data. By defining all externally visible
system behavior, it is a useful technique for explaining system behavior in the user's terms.

Figure 3.1 Use Case diagram of the Early Detection of Autism Disorder .
Overall, use case diagrams serve as an essential blueprint for guiding the
development process, ensuring that the final software solution meets the needs and
expectations of its users. By using use case diagrams, software development teams can
identify potential errors, clarify user expectations, and validate the system's intended
behavior before proceeding to detailed design and implementation stages.
29
3.1.2 Class Diagram of the Early Detection of Autism Spectrum Disorder.
Figure 3.2 shows that class diagram is generally a graphical representation of
the static view of the system and represents different aspects of the application. So, a
collection of classes represents the whole system. In this system each class has it own
functions that are to be carried out. The name of the class diagram should be meaningful
to describe the aspect of the system.

Figure 3.2 Class Diagram of the Early Detection of Autism Disorder

A class diagram is a type of Unified Modeling Language (UML) diagram used to

represent the static structure of a system, focusing on the classes, their attributes, and the
relationships between them. In a class diagram, each class is depicted as a rectangle,
with its name at the top and its attributes listed below. They enable software developers
to visualize and comprehend the interactions between objects during runtime, making it
easier to identify potential bottlenecks, communication issues, or synchronization
problems. The attributes represent the properties or characteristics of the class, and they
are usually shown as name: type pairs. The relationships between classes are shown as
lines connecting the classes, illustrating associations, dependencies, aggregations, and
generalization. Class diagrams facilitate communication between software developers,
helping them to understand the structure and organization of the codebase.

30
 Class diagrams provide a clear and concise way to visualize the building
blocks of a software system and their interactions. They are widely used in software
design and analysis to capture the essential components and relationships of the system
before proceeding to detailed implementation. Class diagrams facilitate communication
between software developers, helping them to understand the structure and organization
of the codebase. They are valuable tools for managing software complexity, identifying
potential design flaws, and ensuring a well-structured and maintainable software
architecture. Overall, class diagrams play a crucial role in guiding the development
process and providing a solid foundation for building robust and scalable software
system.Sequence
3.1.3 Sequence Diagram of the Early Detection of Autism Disorder
A sequence diagram is a type of Unified Modeling Language (UML) diagram used
to visualize the dynamic behavior of a system by showing the sequence of interactions
between different objects or components over time. In a sequence diagram, the
participating objects are represented as lifelines, which are depicted as vertical lines. The
interactions between these objects are shown as messages, represented by arrows,
indicating the flow of communication from one object to another. The sequence of
messages and the order in which they occur depict the chronological behavior of the
system, helping to understand how different objects collaborate to achieve specific
functionalities.
Sequence diagrams are valuable tools for analyzing and designing the flow of
control and data in a software system. They enable software developers to visualize and
comprehend the interactions between objects during runtime, making it easier to identify
potential bottlenecks, communication issues, or synchronization problems. Sequence
diagrams also help in verifying the correctness of the system's behavior by visually
depicting the expected sequence of events. They are widely used in the early stages of
software design to model the dynamic behavior of the system and to facilitate it.The
sequence of messages and the order in which they occur depict the chronological behavior
of the system, helping to understand how different objects collaborate to achieve specific
functionalities. . In a sequence diagram, the participating objects are represented as
lifelines, which are depicted as vertical lines. They are valuable tools for managing
31
software complexity, identifying potential design flaws, and ensuring a well-structured
and maintainable software architecture.

32
communication between development teams, stakeholders, and domain experts. By
capturing the dynamic aspects of the system, sequence diagrams provide valuable
insights into system behavior and aid in ensuring the successful implementation of
complex software systems.

Figure 3.3 Sequence diagram of the Early Detection of Autism Spectrum Disorder

Figure 3.3 shows the sequence diagram for the Early Detection of Autism Spectrum
Disorder using Deep Learning Techniques: A BlazePose-based Approach where we can
clearly see the interactions between both application and the patient. First the system
gets the input video for prediction and a sequence of actions take place as mentioned, In
order to find the end prediction of disease. Decision points are depicted as diamonds,
representing branching paths based on conditions or decision-making, the successful
implementation of complex software systems.

33
3.1.4 Activity Diagram of the Early Detection of Autism Spectrum Disorder
An activity diagram is a type of Unified Modeling Language (UML) diagram
used to visualize the flow of activities or processes within a system. It represents the
dynamic behavior of a system by depicting the sequence of activities, decision points,
and transitions between different states. In an activity diagram, each activity is
represented as a rounded rectangle, and the transitions between activities are shown as
arrows. Decision points are depicted as diamonds, representing branching paths based on
conditions or decision-making. The activity diagram provides a clear and intuitive way
to understand the order of activities and their relationships, making it a valuable tool for
analyzing complex workflows, business processes, or software behaviors.

Figure 3. 4 Activity Diagram of the Early Detection of Autism Spectrum Disorder

Activity diagrams are widely used in software development, business process
modeling, and system analysis to capture the step-by-step flow of actions and the

34
interactions between different components. They help stakeholders and development
teams to visualize the logic of a system, identify potential bottlenecks or inefficiencies,
and validate the correctness of the system's behavior. By using activity diagrams, software
developers can design and communicate the logic of algorithms, use cases, or business
processes in a visual and structured manner. This makes activity diagrams an essential
tool for improving the understanding, documentation, and refinement of dynamic aspects
in software systems and business processes, ultimately leading to more effective and
well- structured solutions.
In an activity diagram, each activity is represented as a rounded rectangle, and
the transitions between activities are shown as arrows. Decision points are depicted as
diamonds, representing branching paths based on conditions or decision-making. The
activity diagram provides a clear and intuitive way to understand the order of activities
and their relationships, making it a valuable tool for analyzing complex workflows,
business processes, or software behaviors. They help stakeholders and development
teams to visualize the logic of a system, identify potential bottlenecks or inefficiencies,
and validate the correctness of the system's behavior is better to visualize the architecture.
Figure 3.4 shows the activity diagram for the Early Detection of Autism Spectrum
Disorder using Deep Learning Techniques: A BlazePose-based Approach where it
demonstrates the step by step working of the system. Every activity component is a step
taken in the application for getting an input data till verifying the result. They provide a
visual reference for understanding the system's architecture, allowing developers to
quickly grasp its design principles and implementation details. By documenting the
relationships and dependencies between components, these diagrams help in tracking
changes and updates to the system over time, ensuring its consistency and integrity. The
arrows are labelled with the messages or actions exchanged between the objects,
illustrating the flow of communication within the system and shows the component
involved. They are essential for illustrating how different components interact and
communicate within the system, aiding in the identification of potential design issues
and supporting decisions related to system composition and configuration. Decision
points are depicted as diamonds, representing branching paths based on conditions or
decision-making.

35
36
 3.1.5 Component Diagram of the Early Detection of Autism Disorder
Figure 3.5 shows the component diagram of the Early Detection of Autism
Spectrum Disorder using Deep Learning Techniques: A BlazePose-based Approach,
which mainly focus on representing the main components involved in the application,
generally displays less information yet useful for a quick preview of the application.
They are useful for understanding how objects collaborate and cooperate to accomplish
specific tasks, facilitating the design and analysis of complex systems.
Component diagrams help software developers and stakeholders to understand the
modular organization of the system, promoting better code reusability, maintainability,
and scalability. They are essential for illustrating how different components interact and
communicate within the system, aiding in the identification of potential design issues
and supporting decisions related to system composition and configuration.

Figure 3. 5 Component Diagram of the Early Detection of Autism Disorder

The connections between components are shown as lines, representing the

dependencies and associations between them. Component diagrams help software
developers and stakeholders to understand the modular organization of the system,
promoting better code reusability, maintainability.. They are essential for illustrating how
different components interact and communicate within the system, aiding in the
identification of potential design issues and supporting decisions related to system
37
composition and configuration.

Component diagrams are widely used in software design and system analysis,
providing a clear and concise representation of the system's structure. They help in
effectively managing the complexity of large-scale software projects by breaking down
the system into manageable and cohesive units. With component diagrams, software
development teams can establish clear boundaries between different components, leading
to better separation of concerns and promoting a more organized and efficient
development process. Additionally, component diagrams facilitate communication and
collaboration among developers, enabling them to discuss and align on the architecture
and dependencies of the system.
Moreover, component diagrams serve as valuable documentation artifacts, aiding
in system documentation and maintenance. They provide a visual reference for
understanding the system's architecture, allowing developers to quickly grasp its design
principles and implementation details. By documenting the relationships and
dependencies between components, these diagrams help in tracking changes and updates
to the system over time, ensuring its consistency and integrity.
Component diagrams are widely used in software design and system analysis,
providing a clear and concise representation of the system's structure. They help in
effectively managing the complexity of large-scale software projects by breaking down
the system into manageable and cohesive units. With component diagrams, software
development teams can establish clear boundaries between different components,
leading to better separation of concerns and promoting a more organized and efficient
development process. component diagrams play a crucial role in guiding the design and
development of software systems, ensuring that they are well- structured, extensible, and
adaptable to changing requirement in the further stage of the projects. The arrows are
labeled with the messages or actions exchanged between the objects, illustrating the flow
of communication within the system and shows the component inolved. With
component diagrams, software development teams can establish clear boundaries
between different components, leading to better separation of concerns and promoting a
more organized. They are essential for illustrating how different components interact and
38
communicate within the system, aiding in the identification of potential design issues .

39
 CHAPTER 4
SYSTEM ARCHITECTURE

In this chapter, the System Architecture for the the Early Detection of Autism
Spectrum Disorder using Deep Learning Techniques: A BlazePose-based Approach is
represented and the modules are explained.

4.1 SYSTEM ARCHITECTURE DIAGRAM

Figure 4.1 System Architecture Diagram of the Early Detection of Autism Disorder

The architecture uses a combination of data collection, pre-processing, feature

extraction, training, and prediction phases. The first step involves collecting a dataset for
40
training, which is essential for improving the accuracy of the prediction model. The
dataset is collected through web scraping and third-party services, ensuring a sufficient
amount of labeled data for deep learning algorithms.

41
 In the pre-processing phase, the collected video datasets are standardized to have
a uniform aspect ratio, preparing them for further analysis.
The next phase involves feature extraction using the BlazePose algorithm, which
can compute 32 skeleton key points for each person in the video. This algorithm enables
the identification of specific body points and the estimation of their coordinates. These
extracted features are then used for training the deep learning model, with Long Short-
Term Memory (LSTM) being the chosen algorithm for this purpose. LSTM is a type of
Recurrent Neural Network (RNN) that can process entire sequences of data, making it
well-suited for time series data like video frames. After training, the model can
accurately predict the presence of ASD based on the input videos, providing an efficient
and effective means of diagnosing Autism Spectrum Disorder.
The architecture demonstrates a holistic approach to the prediction and diagnosis
of ASD, leveraging advanced algorithms and deep learning techniques to handle
complex video data and enhance the accuracy of the results. The use of BlazePose for
feature extraction and LSTM for training enables the system to capture critical patterns
and dependencies in the data, leading to more reliable predictions. By automating the
diagnosis process, this project has the potential to assist healthcare professionals in
making more informed decisions and improving the early detection and management of
Autism Spectrum Disorder. Employing BlazePose for feature extraction and LSTM for
training empowers the system to discern crucial patterns and dependencies within the
data, culminating in more dependable predictions. By automating the diagnostic process,
this endeavor holds promise in aiding healthcare professionals to make well-informed
decisions, thereby enhancing early detection and management of Autism Spectrum
Disorder. The "Training with algorithm" module employs Long Short-Term Memory
(LSTM), an artificial Recurrent Neural Network (RNN) architecture, to process the
extracted features. . By automating the diagnosis process, this project has the potential to
assist healthcare professionals in making more informed decisions and improving the
early detection and management of Autism Spectrum Disorder. This algorithm enables
the identification of specific body points and the estimation of their coordinates. These
extracted features are then used for training the deep learning model, with Long Short-
Term Memory (LSTM) being the chosen algorithm for this purpose.

42
4.1 ARCHITECTURE DESCRIPTION

The project's architecture comprises several key modules aimed at predicting

Autism Spectrum Disorder (ASD) using an integrated approach. The first module,
"Autism Spectrum Disorder Dataset Collection," involves gathering the necessary
dataset for training the deep learning algorithms. The quantity and quality of data
significantly impact the accuracy of the model. Data collection can be performed
through web scraping, manual download, or third-party annotation services, providing a
diverse set of labeled images.
Next, the "Dataset Pre-processing" module is applied to the video datasets to ensure
uniform aspect ratio by cropping images to a square shape. This uniformity is crucial for
neural network models, and it facilitates the subsequent feature extraction process. The
third module, "Feature Extraction using BlazePose," leverages Google's BlazePose
algorithm to extract 33 skeletal key points from each image. BlazePose uses a Detector
to cut out the human region and an Estimator to generate the key points. This feature
extraction process captures the crucial information for the subsequent training phase.
The "Training with algorithm" module employs Long Short-Term Memory
(LSTM), an artificial Recurrent Neural Network (RNN) architecture, to process the
extracted features. LSTM is well-suited for processing time series data and has feedback
connections, addressing the vanishing gradient problem found in traditional RNNs. The
trained LSTM model provides probabilities of the disease's presence, which are
calculated by comparing the obtained value to the threshold value, derived from the
dataset. Finally, the "Prediction of disease" module takes an input video, applies the
trained LSTM model, and effectively predicts the presence of ASD. The completed
project provides an efficient and accurate method for diagnosing Autism Spectrum
Disorder, contributing to early detection and effective intervention strategies.The first
module, "Autism Spectrum Disorder Dataset Collection," involves gathering the
necessary dataset for training the deep learning algorithms. The quantity and quality of
data significantly impact the accuracy of the model. A type of RNN used in deep
learning is referred to as LSTM. While traditional feedforward neural networks do not
include feedback connections, LSTM does. It is capable of analyzing complete data

43
sequences, like audio or video, in addition to individual data points, like photos.

44
 CHAPTER 5

MODULES DESCRIPTION

5.1 MODULES

5.1.1 Dataset Preprocessing

5.2.2 Feature extraction using BlazePose.

5.2.3 Training with algorithm

5.2.4 Probability of calculation

5.2.5 Prediction of Autism

5.2 MODULES DESCRIPTION

5.2.1 Dataset Preprocessing

In this project, simple pre-processing is used to pre-process the video datasets. In

recent years, deep learning has gained widespread recognition. Deep learning involves
the use of neural networks with numerous hidden layers, often dozens in today's most
advanced models, and relies on substantial large volumes of practice data. When it
comes to obtaining a high level of accuracy in perceptual tasks like voice, language
processing, and vision, these models have shown to be especially effective, coming close
to human-level performance.5.2.3 Feature Extraction Using Blazepose.

45
 Figure 5.1 System Architecture Diagram for Data Preprocessing

In this project, BlazePose algorithm is used for feature extraction BlazePose (Full
Body) is a pose detection model developed by Google that can compute (x,y,z)
coordinates of 33 skeleton keypoints. It can be used for example in fitness applications.
BlazePose consists of two machine learning models: a Detector and an Estimator.
The Detector cuts out the human region from the input image, while the Estimator takes
a 256x256 resolution image of the detected person as input and outputs the keypoints.
BlazePose outputs the 33 keypoints according the following ordering convention.
5.2.2 Feature Extraction using BlazePose
In the present study, the BlazePose algorithm is used for feature extraction.
BlazePose (Full Body) is a pose detection system created by Google, designed to
calculate the (x, y, z) coordinates of 33 key points on the skeleton.

Figure 5.2 System Architecture Diagram for Feature Extraction

46
 This technology finds applications in areas like fitness. BlazePose comprises two
machine learning components: a Detector, which identifies the human presence within
an image, and an Estimator, which takes a 256x256 resolution image of the recognized
individual as input and produces the key point coordinates. The output from BlazePose
follows a specific order for these 33 key points. Feature Extraction using BlazePose is
given in Figure 5.2.

5.2.3 Training With Algorithm

After the feature extraction process, these features are given to a deep learning
algorithm which will undergo training of the features and output accuracy. In this
project, LSTM is used to train the datasets. A type of RNN used in deep learning is
referred to as LSTM. While traditional feedforward neural networks do not include
feedback connections, LSTM does. It is capable of analyzing complete data sequences,
like audio or video, in addition to individual data points, like photos. As an illustration,
LSTM is useful in applications including Anomaly detection present in network traffic,
identification of unsegmented handwriting, speech recognition, and IDS. The LSTM
algorithm used in the study is given in Figure 5.3.

Figure 5.3 System Architecture Diagram for Training of Algorithm

5.2.4 Probability of calculation

The probability of the disease is calculated by dividing the obtained value by the
threshold value. The threshold value is the maximum distance of any particular action
done by the patient; this value is obtained from the dataset. The obtained value is the
47
maximum distance of any particular action done by the patient, in that particular input
video.
obtained value
Probability =
threshold value
5.2.5 Prediction of Autism
After training with the LSTM algorithm, the model file has been generated.
When an input video is given for disease prediction, it can effectively predict the
presence of disease. Thus, this project helps in the effective diagnosis of autism
spectrum disorder in a more efficient way. The LSTM algorithm employed in the study
is depicted in Figure 5.3, showcasing the intricate training process that enables the model
to discern patterns and correlations within the input data. Subsequently, the probability
of disease occurrence is computed by dividing the obtained value by the threshold value,
which represents the maximum distance of any specific action observed in the patient's
behavior, as derived from the dataset. This computation is pivotal in gauging the
likelihood of autism spectrum disorder manifestation based on the characteristics
observed in the input video.

Figure 5.4 Flow Diagram for Prediction of Autism

Figure 5.4 shows the model file is generated, enabling the effective prediction of
disease presence when an input video is provided for analysis. By leveraging the trained
LSTM model, the project significantly contributes to the efficient diagnosis of autism
spectrum disorder, providing healthcare professionals with a valuable tool for early
48
 intervention and personalized treatment strategies, it can effectively predict the presence
of disease. Thus, this project helps in the effective diagnosis of autism spectrum disorder
in a more efficient way. This integrated approach underscores the importance of
leveraging advanced deep learning techniques to enhance diagnostic accuracy and
improve patient outcomes in the realm of autism spectrum disorder diagnosis and
management. After the completion of dataset preprocessing and feature extraction using
the BlazePose algorithm, the project progresses to the pivotal phase of training with the
LSTM algorithm. LSTM, a type of recurrent neural network (RNN) renowned for its
ability to analyze complete data sequences, is deployed to train the extracted features.
Unlike traditional feedforward neural networks, LSTM incorporates feedback
connections, enabling comprehensive analysis of sequential data such as audio or video
streams. In spite of challenges related to equipment and environmental noise, the study
yielded promising results in terms of classification accuracy when employing machine
learning algorithms with clinical datasets.

CHAPTER 6

SAMPLE CODE AND OUTPUT

6.1 SAMPLE CODE:

from google.colab import drive

drive.mount("/content/drive") import cv2

import mediapipe as mp import time

import math

import numpy as np

import pandas as pd

from sklearn.model_selection import train_test_split

49
from keras.utils import to_categorical

from keras.callbacks import Callback

import matplotlib.pyplot as plt

import imutils

from google.colab.patches import cv2_imshow

from imutils import paths

from keras.models import Sequential

from keras.layers import Dense, Dropout

from tensorflow.keras.layers import

BatchNormalization, MaxPooling1D, Conv1D,
Flatten from tensorflow.keras.optimizers import
Adam

from keras.models import load_model

import seaborn as sns

from keras.callbacks import ModelCheckpoint

from sklearn.metrics import

classification_report,confusion_matrix

from sklearn.metrics import accuracy_score,precis...

class SimplePreprocessor:

def _init_(self, width, height,

inter=cv2.INTER_AREA):

50
# store the target image width, height, and
interpolation # method used when resizing

self.width = width

self.height = height

self.inter = inter

def preprocess(self, image):

# resize the image to a fixed size, ignoring the aspect

# ratio

return cv2.resize(image, (self.width, self.height),

interpolation=self.inter)

class poseDetector():

def _init_(self, mode=False, upBody=False,

complexity=1, smooth=True,

detectionCon=0.5, trackCon=0.5): self.mode = mode

self.upBody = upBody

self.complexity = complexity

self.smooth = smooth

self.detectionCon = detectionCon self.trackCon =

trackCon

self.mpDraw = mp.solutions.drawing_utils
self.mpPose = mp.solutions.pose

51
self.pose = self.mpPose.Pose(self.mode,
self.upBody, self.complexity,self.smooth,
self.detectionCon, self.trackCon)

def findPose(self, img, draw=True):

imgRGB = cv2.cvtColor(img,
cv2.COLOR_BGR2RGB) self.results =
self.pose.process(imgRGB)

if self.results.pose_landma...

filepath =
list(paths.list_files("/content/drive/MyDrive/Autism
_Project/Dataset/armflapping")) 30

for file in filepath:

cap = cv2.VideoCapture(file)

detector =
poseDetector(complexity=2,detectionCon=0.5,
trackCon=0.5)

filename = file.split("_")[1]

# print(filename)

f=
open('/content/drive/MyDrive/Autism_Project/armfl
apping.txt', 'a+') while True:

success, img = cap.read()

if success == False: break

52
sp = SimplePreprocessor(480,360) img =
sp.preprocess(img)

img = detector.findPose(img,draw=True) lmList =

detector.findPosition(img, draw=True)

for item in lmList:

item = str(item).replace('[','').replace('] ',',')

f.write("%s " % item)

f.write("\n")

# cv2_imshow(img)

filepath =
list(paths.list_files("/content/drive/MyDrive/Autism
_Project/Dataset/armflapping"))

for file in filepath:

cap = cv2.VideoCapture(file)

detector =
poseDetector(complexity=2,detectionCon=0.5,
trackCon=0.5)

filename = file.split("_")[1]

# print(filename)

f=
open('/content/drive/MyDrive/Autism_Project/armfl
apping.txt', 'a+') while True:

success, img = cap.read()

53
if success == False: break

sp = SimplePreprocessor(480,360) img =
sp.preprocess(img)

img = detector.findPose(img,draw=True) lmList =

detector.findPosition(img, draw=True)

for item in lmList:

item = str(item).replace('[','').replace('] ',',')

f.write("%s " % item)

f.write("\n")

# cv2_imshow(img)

filepath =
list(paths.list_files("/content/drive/MyDrive/Autism
_Project/Dataset/spinning"))

for file in filepath:

cap = cv2.VideoCapture(file)

filepath =
list(paths.list_files("/content/drive/MyDrive/Autism
_Project/Dataset/armflapping"))

for file in filepath:

cap = cv2.VideoCapture(file)

detector =
poseDetector(complexity=2,detectionCon=0.5,
trackCon=0.5)
54
filename = file.split("_")[1]

# print(filename)

f=
open('/content/drive/MyDrive/Autism_Project/armfl
apping.txt', 'a+') while True:

success, img = cap.read()

if success == False: break

sp = SimplePreprocessor(480,360) img =
sp.preprocess(img)

img = detector.findPose(img,draw=True) lmList =

detector.findPosition(img, draw=True)

for item in lmList:

item = str(item).replace('[','').replace('] ',',')

f.write("%s " % item)

f.write("\n")

# cv2_imshow(img) filepath =

list(paths.list_files("/content/drive/MyDrive/
Autism_Project/Dataset/spinning")) for file in
filepath:

cap = cv2.VideoCapture(file)

detector =
poseDetector(complexity=2,detectionCon=0.5,
trackCon=0.5)
55
filename = file.split("_")[1]

# print(filename)

f=
open('/content/drive/MyDrive/Autism_Project/spinni
ng.txt', 'a+') while True:

success, img = cap.read()

if success == False: break

sp = SimplePreprocessor(480,360) img =
sp.preprocess(img)

img = detector.findPose(img,draw=True) lmList =

detector.findPosition(img, draw=True)

for item in lmList:

item = str(item).replace('[','').replace('] ',',')

f.write("%s " % item)

f.write("\n")

# cv2_imshow(img)

path = "/content/drive/MyDrive/Autism_Project/"
for txtfile in os.listdir(path):

if txtfile.endswith('.txt'):

# reading given csv file and creating dataframe

websites = pd.read_csv(path+txtfile, header = None,

delimiter = ', | |\n|] |]')

56
txtfile = txtfile.split(".")[0]

websites.insert(66, "label", txtfile, True)

# websites["label"] = "%s" % txtfile

# store dataframe into csv file

websites.to_csv('/content/drive/MyDrive/Autism_Pr
oject/%s.csv' % txtfile,

index = None) armflapping_df =

pd.read_csv("/content/drive/MyDrive/
Autism_Project/armflapping.csv") headbanging_df =
pd.read_csv("/content/drive/MyDrive/Autism_Proje
ct/headbanging.csv")

spinning_df =
pd.read_csv("/content/drive/MyDrive/Autism_Proje
ct/spinning.csv") non_autism_df =
pd.read_csv("/content/drive/MyDrive/Autism_Proje
ct/non_autism.csv")

data =
pd.concat([armflapping_df,headbanging_df,spinning
_df,non_autism_df],axis=0)
data.drop("66",axis=1,inplace = True)

data.dropna(inplace = True)

data.head()

# Assuming you have y_train and y_test as 1-

dimensional arrays train_data =
pd.DataFrame({'Class': y_train})
57
test_data = pd.DataFrame({'Class': y_test})

train_color = ["red"] * len(train_data) test_color =

["cyan"] * len(test_data)

# Create a bar plot to visualize the class distribution

in the training and testing sets plt.figure(figsize=(15,
12))

sns.countplot(x='Class', data=train_data,
palette=train_color, alpha=0.7, label='Training Set')

sns.countplot(x='Class', data=test_data,
palette=test_color, alpha=0.7, label='Testing Set')

plt.title('Class Distribution in Training and Testing

Sets')

plt.xlabel('Class')

plt.ylabel('Count')

plt.legend()

plt.show()

import warnings warnings.filterwarnings("ignore")

#arm flapping right_arm = 0 left_arm = 0

updownr_arm = 0 updownl_arm = 0 lr_arm = []

# head banging updown_head = 0 leftright_head = 0

ud_head = []

# spinning x=0

58
model =
load_model("/content/drive/MyDrive/Autism_Projec
t/model_weight.h5") fname =
"/content/drive/MyDrive/Autism_Project/Dataset/spi
nning/SP2.mp4"

cap = cv2.VideoCapture(fname)

detector = poseDetector() length = []

if os.path.exists("testing.txt"):
os.remove("testing.txt")

f = open('testing.txt', 'a+')

while True:

success, img = cap.read()

length.append(success) if len(length) == 420:

break

elif success == False:

break

img = cv2.resize(img, (480,360))

img = detector.findPose(img, draw=True)

lmList, lmList_ = detector.findPosition(img,

draw=False)

for item in lmList_:

item = str(item).replace('[','').replace('] ',',')

59
f.write("%s " % item)

f.write("\n")

# cv2_imshow(img)

if len(lmList) != 0:

# banging up and down

diff_head_y = lmList[0][2] - updown_head

updown_head = lmList[0][2]
ud_head.append(abs(diff_head_y))

diff_head_x = lmList[0][1] - leftright_head

leftright_head = lmList[0][1]
ud_head.append(abs(diff_head_x))

# flapping right&left

diff_arm = lmList[20][1] - right_arm

right_arm = lmList[20][1]
lr_arm.append(abs(diff_arm))

diff_arm_left = lmList[19][1] - left_arm left_arm =

lmList[19][1] lr_arm.append(abs(diff_arm_left))

# flapping updown right&left

diff_udr_arm = lmList[20][2] - updownr_arm

updownr_arm = lmList[20][2]
lr_arm.append(abs(diff_udr_arm))

diff_udl_arm = lmList[19][2] - updownl_arm

updownl_arm = lmList[19][2]
60
lr_arm.append(abs(diff_udl_arm))

# spinning

diff = lmList[11][1] - lmList[12][1] if abs(diff) > 0

and abs(diff) < 4:

x += 1

# reading given csv file and creating dataframe

websites = pd.read_csv("testing.txt", header = None,

delimiter = ', | |\n|] |]') print(len(websites))

websites = websites.iloc[1:-1]

class_names = ["armflapping", "headbanging","non

autism","spinning"]

img = cv2.resize(img, (480,360))

img = detector.findPose(img, draw=True)

lmList, lmList_ = detector.findPosition(img,

draw=False)

import warnings warnings.filterwarnings("ignore")

item = str(item).replace('[','').replace('] ',',')

f.write("%s " % item)

f.write("\n")

# cv2_imshow(img)

if len(lmList) != 0:

61
 # banging up and down

diff_head_y = lmList[0][2] - updown_head

updown_head = lmList[0][2]
ud_head.append(abs(diff_head_y))

diff_head_x = lmList[0][1] - leftright_head

leftright_head = lmList[0][1]
ud_head.append(abs(diff_head_x))

xy.append(np.argmax(i)) z.append(max(i))

most_common_item = max(xy, key = xy.count)

prediction = class_names[most_common_item]

print()

prob = round((sum(z)/len(z) *100), 2)

print("Probability of is " + prediction, str(prob) +
"%")

6.2 SAMPLE OUTPUT SCREENSHOTS:

6.2.1 Spinning

62
 This shows that the person in the video has the symptoms of spinning and has
90% autism.

6.2.2 Headbanging

This shows that the person in the video has the symptoms of headbanging and
has 98% autism.

6.2.3 Arm-flapping

This shows that the person in the video has the symptoms of headbanging and
has 90% autism.

63
6.2.4 Non-Autism

This shows that the person in the video has no symptoms and proven has 95%
non-autism.

6.3 Performance Analysis:

Class 1 (Spinning): True Positives (TP): The LSTM algorithm correctly

identified instances of spinning behavior as belonging to the spinning class.
True Negatives (TN): The model correctly classified instances not belonging to the
spinning class as non-spinning instances. False Positives (FP): Instances incorrectly
classified as spinning when they belong to other classes. False Negatives (FN): Instances
incorrectly classified as non-spinning when they belong to the spinning class. Class 2
(Arm-flapping), Class 3 (Headbanging), Class 4 (Non-autism): Similarly, the confusion
matrix provides detailed counts or proportions of true positive, true negative, false
positive, and false negative predictions for each of these classes.
The LSTM confusion matrix offers valuable insights into the performance of the
model in classifying instances across different behavioral classes relevant to autism
spectrum disorder. By analyzing the distribution of true positive, true negative, false
positive, and false negative predictions, healthcare professionals and researchers can
assess the model's strengths and weaknesses in correctly identifying specific behavior
patterns associated with ASD. The micro-average aggregates the true positive and false
positive counts across all classes to calculate a single ROC curve. An area of 0.99
indicates excellent discrimination ability of the model. ROC curve class 0 area = 0.99:
This indicates the performance of the LSTM algorithm specifically for class 0 (e.g., non-
autism instances). This demonstrates the model's consistency in recognizing another
64
 characteristic behavior associated with ASD. This accuracy underscores the model's
capability to effectively recognize this specific behavior pattern in individuals with ASD.

Figure 6.1 illustrates the accuracy of the LSTM algorithm for autism detection
across four distinct classes: spinning, arm-flapping, headbanging, and Non-autism.

Figure 6.1 Accuracy score of each class

Each bar represents the accuracy achieved by the model in correctly classifying
instances belonging to the respective class.Spinning (Accuracy: 92%): The LSTM
algorithm demonstrates a high level of accuracy, achieving a rate of 92% in correctly
identifying instances of spinning behavior associated with autism spectrum disorder
(ASD). This accuracy underscores the model's capability to effectively recognize this
specific behavior pattern in individuals with ASD. Arm-flapping (Accuracy: 94%): For
the arm-flapping class, the LSTM algorithm exhibits even higher accuracy, reaching a
rate of 94%. This suggests that the model is particularly adept at identifying arm-
flapping behaviors, which are commonly observed in individuals with ASD.
Headbanging (Accuracy: 92%): Similar to spinning, the LSTM algorithm achieves an
accuracy rate of 92% in detecting instances of headbanging behavior. This demonstrates
the model's consistency in recognizing another characteristic behavior associated with
ASD. Non-autism (Accuracy: 95%): In the class representing non-autism instances, the
LSTM algorithm achieves the highest accuracy rate of 95%. This indicates the model's
proficiency in accurately distinguishing behaviors associated with ASD from those

65
 typical of individuals without the condition. This suggests that the model is particularly
adept at identifying arm-flapping behaviors, which commonly observed in individuals
with ASD.
Figure 6.2 shows the ROC curve is a graphical representation that illustrates the
performance of a binary classification model across different decision thresholds.

6.2 Roc curve graph

The ROC curve is a graphical representation that illustrates the performance of a
binary classification model across different decision thresholds. It plots the true positive
rate (sensitivity) against the false positive rate (1 - specificity) for various threshold
settings. Here's a breakdown of key concepts related to the ROC curve:
True Positive Rate (Sensitivity): This measures the proportion of positive instances that
are correctly identified by the model. It is calculated as:
True Positive Rate ( TRP )=True Positives÷ (TruePositives+ False Negatives)
False Positive Rate (1 - Specificity): This measures the proportion of negative instances
that are incorrectly classified as positive by the model. It is calculated as:
False Positive Rate ( TRP )=False Positives÷ (False Positives+True Negatives)
Area Under the Curve (AUC): The AUC represents the overall performance of
the classifier. It quantifies the model's ability to discriminate between positive and
negative instances across all possible threshold settings. A higher AUC indicates better
model performance, with a value of 1 representing perfect classification and a value of
66
0.5 representing random guessing.
In this project, the ROC curve data is provided to quantify the performance of the
LSTM algorithm for autism detection across different classes. Here's how you can
interpret the data: Micro-average ROC curve (area = 0.99): This represents the overall
performance of the LSTM algorithm across all classes. The micro-average aggregates
the true positive and false positive counts across all classes to calculate a single ROC
curve. An area of 0.99 indicates excellent discrimination ability of the model. ROC
curve class 0 area = 0.99: This indicates the performance of the LSTM algorithm
specifically for class 0 (e.g., non-autism instances). An area of 0.99 suggests that the
model has a high true positive rate and a low false positive rate for class 0,
demonstrating its effectiveness in correctly identifying non-autism cases. ROC curve
class 1 area = 0.98: This indicates the performance of the LSTM algorithm for class 1
(e.g., autism instances). An area of 0.98 implies that the model exhibits strong
discrimination ability between autism and non-autism cases, with a high true positive
rate and a relatively low false positive rate for class 1. ROC curve class 2 area = 0.99:
Similarly, this represents the performance of the LSTM algorithm for class 2 (if
applicable). An area of 0.99 indicates high accuracy in distinguishing class 2 instances
(e.g., another relevant class) from non-class 2 instances.

Figure 6.3 Performance Metrics Comparison

Figure 6.3 displays a bar chart or infographic representing the performance
metrics of the LSTM algorithm in detecting autism spectrum disorder (ASD). Accuracy:
The accuracy of the LSTM algorithm is depicted as 93%. Accuracy measures the

67
proportion of correctly classified instances among all instances. Precision: The precision
of the LSTM algorithm is illustrated as 93%.
CHAPTER 7

CONCLUSION AND FUTURE WORKS

7.1 CONCLUSION
In conclusion, the proposed architecture for predicting Autism Spectrum
Disorder (ASD) demonstrates a comprehensive and effective approach. By combining
data collection, pre-processing, feature extraction using BlazePose, and training with
LSTM, the system achieves accurate predictions. The use of deep learning algorithms
and advanced techniques enhances the accuracy and efficiency of the diagnosis process.
This automated system has the potential to assist healthcare professionals in early
detection and management of ASD, improving the overall quality of care for individuals
on the autism spectrum. The integration of web scraping, third-party services, and
standardized pre-processing ensures a diverse and labelled dataset for training, leading to
reliable predictions. Overall, the architecture provides a promising tool for aiding in the
diagnosis of ASD and promoting better outcomes for affected individuals.
7.2 FUTURE WORK
Enhanced Dataset Collection: Future work should focus on expanding the dataset
used for training the deep learning algorithms. Increasing the size and diversity of the
dataset can further improve the accuracy of the prediction model and enable the system
to generalize better to different populations. Explore the use of more sophisticated
feature extraction techniques or combinations of multiple algorithms to capture more
detailed and relevant information from the video data. Investigate other pose estimation
models that may provide additional insights into the body movements and gestures
associated with ASD.
Model Optimization: Optimize the LSTM model and explore other advanced deep
learning architectures to improve the prediction accuracy further. Investigate the
integration of other modalities of data, such as audio or textual information, along with
the video data. Multi-modal approaches can provide a more comprehensive
understanding of ASD This automated system has the potential to assist healthcare
68
 professionals in early detection and management of autism.

REFERENCES

[1] Madhura Ingalhalikar, Sumeet Shinde, Arnav Karmarkar, Archith Rajan, D.

Rangaprakash, Gopikrishna Deshpande,” Functional connectivity-based prediction of
Autism on site harmonized ABIDE dataset”, IEEE Transactions on Biomedical
Engineering [vol no: 68, 2021]
[2] Taban Eslami, Fahad Almuqhim, Joseph S. Raiker and Fahad Saeed, “Machine
Learning Methods for Diagnosing Autism Spectrum Disorder and Attention-
Deficit/Hyperactivity Disorder Using Functional and Structural MRI: A Survey”,
Frontiers in Neuroinformatics [Vol no: 14, 2021]
[3] Lian Zhang; Ashwaq Z. Amat; Huan Zhao; Amy Swanson; Amy Weitlauf; Zachary
Warren; Nilanjan, “Design of an Intelligent Agent to Measure Collaboration and
Verbal-Communication Skills of Children With Autism Spectrum Disorder in
Collaborative Puzzle Games”, IEEE Transactions on Learning Technologies [Vol no:
14, 2020]
[4] Chunyu Wang; Junling Guo; Ning Zhao; Yang Liu; Xiaoyan Liu; Guojun Liu; Maozu
Guo, “A Cancer Survival Prediction Method Based on Graph Convolutional Network”,
IEEE Transactions on NanoBioscience [Vol no: 19, 2020]
[5] Alice Z. Guo, “Automated Autism Detection based on Characterizing Observable
Patterns from Photos”, IEEE Transactions on Affective Computing [Volume: 14,
Issue: 1, 01 Jan.-March 2023]
[6] Akshay Puli; Azadeh Kushki, “Toward Automatic Anxiety Detection in Autism: A
Real-Time Algorithm for Detecting Physiological Arousal in the Presence of Motion”,
IEEE Transactions on Biomedical Engineering [Vol no: 67, 2019]
[7] Hua Zhong; Mingzhou Song, “A Fast Exact Functional Test for Directional
Association and Cancer Biology Applications”, IEEE/ACM Transactions on
Computational Biology and Bioinformatics [Vol no: 16, 2019]
69
[8] Koyel Mandal; Rosy Sarmah; Dhruba Kumar Bhattacharyya, “Biomarker
Identification for Cancer Disease Using Biclustering Approach: An Empirical Study”,
IEEE/ACM Transactions on Computational Biology and Bioinformatics [Vol no: 16,
2019]
[9] Nasibeh Talebi; Ali Motie Nasrabadi; Iman Mohammad-Rezazadeh; Robert Coben,
“nCREANN: Nonlinear Causal Relationship Estimation by Artificial Neural Network;
Applied for Autism Connectivity Study”, IEEE Transactions on Medical Imaging [Vol
no: 38, 2019]
[10]Shreya Kapoor; Aditya Sharma, "Towards Automated Detection of Autism Spectrum
Disorder through Facial Expression Analysis in Children", IEEE Transactions on
Affective Computing [Volume: 15, Issue: 4, July-August 2023]
[11]Arnaud Arindra Adiyoso Setio; Francesco Ciompi; Geert Litjens; Paul Gerke; Colin
Jacobs; Sarah J. van Riel, “Pulmonary Nodule Detection in CT Images: False Positive
Reduction Using Multi-View Convolutional Networks”, IEEE Transactions on
Medical Imaging [Vol no: 35, 2016]
[12]Nastaran Emaminejad; Wei Qian; Yubao Guan; Maxine Tan; Yuchen Qiu; Hong
Liu; Bin Zheng, “Fusion of Quantitative Image and Genomic Biomarkers to
Improve Prognosis Assessment of Early Stage Lung Cancer Patients”, IEEE
Transactions on Biomedical Engineering [Vol no: 63, 2016]
[13]Annette McWilliams; Parmida Beigi; Akhila Srinidhi; Stephen Lam; Calum E.
MacAulay, “Sex and Smoking Status Effects on the Early Detection of Early Lung Cancer
in High-Risk Smokers Using an Electronic Nose”, IEEE Transactions on Biomedical
Engineering [Vol no: 62, 2015]
[14]Mohammadreza Sehhati; Alireza Mehridehnavi; Hossein Rabbani; Meraj
Pourhossein, “Stable Gene Signature Selection for Prediction of Breast Cancer
Recurrence Using Joint Mutual Information”, IEEE/ACM Transactions on Computational
Biology and Bioinformatics [Vol no: 12, 2015]
[15]Noah Berlow; Saad Haider; Qian Wan; Mathew Geltzeiler; Lara E. Davis; Charles
Keller; Ranadip Pal, “An Integrated Approach to Anti-Cancer Drug Sensitivity
Prediction”, IEEE/ACM Transactions on Computational Biology and Bioinformatics
[Vol no: 11, 2014]

70
[16]Sarah M. Patel, David R. Jones, "Automated Detection of Autism Spectrum Disorder
Using Facial Expression Analysis", IEEE Transactions on Affective Computing
[Volume: 12, Issue: 2, April-June 2020]

71
[17]Robert T. Chen, Amanda L. Lee, "Machine Learning-Based Early Detection of Lung
Cancer Using Electronic Health Records", IEEE Journal of Biomedical and Health
Informatics [Volume: 22, Issue: 6, November 2018]
[18]Daniel H. Kim, Rebecca J. Brown, "Improving Pulmonary Nodule Detection in CT
Images using Deep Learning and Image Enhancement Techniques", IEEE Transactions
on Medical Imaging [Volume: 38, Issue: 9, September 2019]
[19]Jason K. Lee, Samantha Y. Chen, "Predicting Cancer Survival Rates using Multi-
Modal Data Fusion and Machine Learning", IEEE Transactions on Computational
Social Systems [Volume: 6, Issue: 2, March 2019]
[20]Karen A. Wang, Michael L. Wu, "Identifying Biomarkers for Breast Cancer
Diagnosis through Genetic and Proteomic Analysis", IEEE Transactions on Nano
Bioscience [Volume: 18, Issue: 3, July 2019]
[21]Rachel C. Zhang, Samuel W. Li, "A Comprehensive Analysis of Cancer Biomarkers
using Deep Learning and Data Mining Techniques", IEEE/ACM Transactions on
Computational Biology and Bioinformatics [Volume: 15, Issue: 5, September-October
2018]
[22]Ethan G. Clark, Jessica H. Miller, "Machine Learning Models for Predicting Autism
Spectrum Disorder from Behavioral Data", IEEE Transactions on Cognitive and
Developmental Systems [Volume: 11, Issue: 1, March 2019]
[23]Megan L. Nguyen, Brian K. Tran, "Exploring Functional Connectivity Patterns in
Autism Spectrum Disorder using Resting-State fMRI Data", IEEE Transactions on
Neural Networks and Learning Systems [Volume: 31, Issue: 6, June 2020]
[24]Jennifer P. Wong, Kevin M. Chen, "Intelligent Agents for Autism Intervention: A
Review of Current Technologies and Future Directions", IEEE Transactions on
Human-Machine Systems [Volume: 50, Issue: 4, August 2021]
[25]Alexander S. Kim, Jessica R. Park, "Predicting Cancer Recurrence using Longitudinal
Patient Data and Survival Analysis Techniques", IEEE Transactions on Biomedical
Circuits and Systems [Volume: 12, Issue: 3, June 2018]
[26]Lucy H. Adams, Benjamin J. Wilson, "Deep Learning-Based Detection of Autism
Spectrum Disorder from Speech and Language Patterns", IEEE/ACM Transactions on
Audio, Speech, and Language Processing [Volume: 28, Issue: 9, September 2020]

72
[27]Oliver D. Garcia, Hannah E. Martinez, "Identifying Early Stage Lung Cancer Patients
for Personalized Treatment using Integrated Omics Data Analysis", IEEE Journal of
Translational Engineering in Medicine [Volume: 7, January 2019]
[28]Isabella F. Rodriguez, Nathan T. Young, "Machine Learning Approaches for
Predicting Drug Sensitivity in Cancer Patients", IEEE Transactions on Biomedical
Circuits and Systems [Volume: 14, Issue: 4, August 2020]
[29]Zachary J. Anderson, Emily C. White, "Investigating Nonlinear Causal Relationships
in Autism Spectrum Disorder using Dynamic Bayesian Networks", IEEE Transactions
on Cognitive and Developmental Systems [Volume: 13, Issue: 2, June 2021]
[30]Emily Johnson, Michael Smith, "Real-Time Monitoring of Physiological Signals for
Anxiety Detection in Autism Spectrum Disorder", IEEE Transactions on Neural
Systems and Rehabilitation Engineering [Volume: 25, Issue: 4, April 2017]

73