Model Development for Early Detection of Autism
Spectrum Disorder in Children Based on Machine learning
from Image
1.Introduction
Autism is a neurological disease that influences how people perceive and interact with their
surroundings. It is not an illness that can be cured, but rather a distinct way of being. Individuals
with autism frequently have difficulties with speech, social engagement, and sensory processing.
Some people thrive at one thing, such as problem solving or creative thinking, while struggling
with another. Early diagnosis of autism in the initial years of life greatly enhances the prospects
for affected children by enabling intervention during rapid brain development.1,2 In the United
States, however, autism is typically not diagnosed until around age four, leaving about 27% of
children untreated by age eight.3 This delay is largely due to the absence of adequate screening
tools and a shortage of specialists capable of assessing at-risk children. Thus, employing more
precise screening methods to prioritize children for specialist evaluation is crucial. According to
Bangabandhu Sheikh Mujib Medical University (BSMMU), around 2 out of every 1000
youngsters in Bangladesh have ASD . According to a 2016 research by the Bangladesh Ministry
of Social Welfare, autism accounted for 19% of all neurological diseases recorded.
2.Background of the study
A neurodevelopmental illness known as autism spectrum disorder impacts a person's ability to
interact, communicate, and learn. Autism can be diagnosed at any age, but symptoms usually
start to show in the first two years of life and get worse over time [1]. Individuals with autism
spectrum disorders encounter various obstacles, including difficulties focusing, learning
disabilities, mental health issues like anxiety and depression, motor difficulties, sensory issues,
and numerous others.The prevalence of autism is currently skyrocketing globally and growing at
an extremely rapid pace. According toWHO [2], about 1 out of every 160 children has ASD.
Some people with this disorder can live independently, while others require life-long care and
support.
The majority of autism screeners in use today are built around questions for the parent or
healthcare provider, and the findings are obtained by comparing the sum of the answers to
predefined cutoff points. Two notable examples are the Child Behavior Checklist (CBCL) and
the Modified Checklist for Autism in Toddlers, Revised (M-CHAT), a checklist-based screening
tool for autism that is meant to be used during developmental screenings for children between
the ages of 16 and 30 months. These are screening tools that parents fill out. When the total for
both instruments exceeds a predefined level, the kid is deemed to be at high risk of autism. The
answers to each question are added together, with equal weighting assigned to each.
The diagnosis of autism takes a long time and money. Early diagnosis of autism can be very
beneficial in terms of providing patients with the right medicine at the right time.It can stop the
patient's illness from getting worse and assist cut down on the long-term expenses related to a
delayed diagnosis. Therefore, it is imperative to have a quick, accurate, and simple screening test
instrument that can anticipate an individual's features associated with autism and determine
whether a thorough autism assessment is necessary.The goal of this effort is to create a mobile
application that can accurately identify an individual's autism features at any age and to propose
an autism prediction model using machine learning techniques.
3.Literature Review
The works pertaining to the ASD prediction methods are included in this section. The ability of
machine learning to forecast various illness kinds based on symptoms is pretty impressive.For
instance, Cruz et al. attempted to use ML to detect cancer in [3], whereas Khan et al. utilized ML
to predict whether or not a person has diabetes in [4]. Alternating Decision Trees (ADTrees)
were utilized by Wall et al. [5] to shorten screening times and speed up the identification of ASD
features. Using data from 891 people, they employed the Autism Diagnostic Interview, Revised
(ADI-R) approach and attained a high degree of accuracy. However, the test's range was only 5
to 17 years old, and it proved unable to predict ASD in individuals of diverse ages (children,
adolescents, and adults).
Using support vector machines (SVM) and machine learning (ML), Bone et al. [6] achieved
89.2% sensitivity and 59% specificity. 1264 people with ASD and 462 people with features that
are NOT associated with ASD were included in their study. However, because of the large age
range (4–55 years), their research was not accepted as a screening method for individuals in any
age group.Allison et al [7] used ‘Red Flags’ tool for screening ASDwith Autism Spectrum
Quotient for children and adult, the shortlisted them to AQ-10 with more than 90%
accuracy.Thabtah [8] compared the previous works on ML algorithms for prediction of autism
traits, while Hauck and Kliewer [9]tried to identify relatively more important screening ques-
tions for ADOS (Autism Diagnostic Observation Schedule)and ADI-R (Autism Diagnostic
Interview Revised) screening methods and found that ADI-R and ADOS are performs better.
Bekerom [10] identified ASD features in children, such as developmental delay, obesity, and
decreased physical activity, by using a variety of machine learning algorithms, such as naive
bayes, SVM, and random forest algorithm, and compared the outcomes. In their work on using a
short screening test and validation to define autism, Wall et al. [11] discovered that the ADTree
and the functional tree functioned well, exhibiting high sensitivity, specificity, and accuracy.
Using a sizable brain imaging dataset from the Autism Imaging Data Exchange (ABIDE I),
Heinsfeld[12] used a deep learning algorithm and neural network to identify individuals with
ASD. The results showed a mean classification accuracy of 70% with a range of 66% to 71%.
The Random Forest classifier obtained mean accuracy of 63%, whilst the SVM classifier attained
mean accuracy of 65%.
Liu [13] used a machine learning system to analyze an eye movement dataset in order to
investigate if face scanning patterns could be helpful in identifying children with ASD. The
results of this investigation indicated an 88.51% accuracy, 86.21% specificity, 93.10%
sensitivity, and 89.63% AUC.Bone et al. [14] used their machine learning approach to replicate
the results after analyzing the earlier works of Wall et al. [11] and Kosmicki et al. [15] to
pinpoint the problems with conceptual problem formulation, methodological implementation,
and interpretation.
The literature study makes it clear that, despite the fact that many studies in this area have been
conducted, the researchers were unable to reach a consensus on the use of machine learning
(ML) to generalize the autism screening test instrument across age groups. Prior to now, various
methods and tools for autism screening tests have been developed; however, no app-based
solution tailored to specific age groups has been developed.
4. Rationale of the study
Researching the detection of autism using machine learning (ML) is crucial for several reasons.
Here’s a structured outline of the logic:
1. Prevalence and Impact of Autism
● High Prevalence: Autism Spectrum Disorder (ASD) affects approximately 1 in 54
children in the United States, according to the CDC.
● Lifelong Impact: Early diagnosis and intervention can significantly improve the quality
of life and developmental outcomes for individuals with ASD.
2. Challenges in Traditional Diagnosis
● Subjective Assessments: Traditional diagnosis relies on behavioral assessments and
developmental history, which can be subjective and vary between practitioners.
● Late Diagnosis: Many children are not diagnosed until age 4 or later, missing critical
early intervention periods.
3. Benefits of Machine Learning in Detection
● Accuracy and Objectivity: ML algorithms can analyze large datasets, identifying
patterns and correlations that may not be apparent to human clinicians, leading to more
accurate and objective diagnoses.
● Early Detection: By analyzing early-life data, such as genetic information, behavioral
patterns, and medical history, ML can potentially identify signs of ASD earlier than
traditional methods.
● Scalability: ML systems can process large amounts of data quickly, making them
scalable solutions for widespread screening.
4. Potential ML Approaches
● Behavioral Analysis: Using video and audio data to analyze social interactions, eye
movement, and speech patterns.
● Genetic Data: Analyzing genetic markers that are associated with ASD.
● Health Records: Mining electronic health records for patterns that correlate with ASD
diagnoses.
5. Ethical and Practical Considerations
● Data Privacy: Ensuring the privacy and security of sensitive health data.
● Bias and Fairness: Addressing biases in training data to avoid disparities in diagnosis
across different populations.
● Clinical Integration: Ensuring ML tools are integrated into clinical workflows in a way
that supports, rather than replaces, human clinicians.
Researching the detection of autism using machine learning offers the potential to transform the
diagnostic landscape. It promises earlier, more accurate diagnoses, improved patient outcomes,
and more efficient use of healthcare resources. Given the significant impact of ASD on
individuals and society, advancing this research is both a scientific and a moral imperative.
5. Research Gap
In the existing works no works taking questions answer and take a snap of the individuals and
also don't give suggestions. But in our by using machine learning models like Random Forest,
Artificial Neural network we will not only detect autism but also for give valid suggestions,
6.Research Questions
In order to investigate the opportunities and difficulties of employing a machine learning
algorithm to identify ASD in children, this study tackles a few research problems. The following
are the study questions:
a) Which models are currently in use for identifying ASD?
b) How can CSV and IMAS data be analyzed from Bangladesh's perspective to identify ASD
early on?
c) What is the suggested approach that makes use of a machine-learning algorithm to categorize
patients with ASD?
d) How can the suggested model use physiological cues in children to accurately identify ASD?
7. Research Objectives
The thesis will be accomplished by meeting the following research objectives. The objectives
are:
● Improved Diagnostic Tools: Developing tools that can assist clinicians in diagnosing
ASD more accurately and earlier.
● Personalized Interventions: Using ML to tailor interventions to the specific needs of
each individual with ASD.
● Cost Reduction: Reducing the overall cost of diagnosis and treatment by streamlining
the process and enabling earlier, less intensive interventions.
8.Method
a.Data Collection
The AQ-10 dataset, which is made up of three distinct datasets based on questions from the AQ-
10 screening tool, was utilized to create an efficient predictive model [16]. These three datasets
provide information for the age ranges of children (4–11 years), adolescents (12–16 years), and
adults (18 years or more). The Autism Spectrum or AQ-10 To determine if a person has to be
referred for a thorough autism assessment, a quotient tool is utilized.A variety of domains,
including communication, imagination, social interaction, attention to detail, and attention
switching, are the emphasis of the AQ-10 screening questions. The questions have a single point
value assigned to each of the ten possible answers.User may score 0 or 1 point on each question
based on their answer [17]. Datasets of child, adolescent and adult contain 292, 104 and 704
instances respectively. Each of the three datasets contains twenty-one attributes which are a mix
of numerical and categorical data, that includes: Age, Gender, Ethnicity, If born with Jundice,
Family member with PDD, Who is completing the test, Country of Residence, Used the
screening app before, Screening method type, Question 1-10, Result and Class.
b. Preprocessing of data
The collected data were synthesized to remove irrelevant features. For example, the ID column
was irreverent to develop a prediction model, thus it was removed. To handle null values, a
listwise deletion technique was applied where a particular observation was deleted if it had one
or more missing values. Then to extract unnecessary features from the dataset, a decision tree
algorithm was used. Results showed dropping ‘relation’, ‘age desc’, ‘used app before’ and ‘age’
columns would result in more accurate classification and so those columns were dropped.
c. Building Machine learning model
After doing feature extractions from the dataset we will build several machine learning models to
detect ASD. We will use Random Forest, Artificial Neural Network. We will use Pytorch for
developing these models.
d. Features and Framework
In this work we will build mobile apps data for detecting ASD. The key features of this models
are given below:
Login and Logout: There will be a login and logout page for users.
User interface: There will be a user interface for users.
Symptoms and Detection: In our project this feature by using machine learning approach we will
set a question set as well as a video upload session so that we can detect ASD.
Suggestion: In this feature our apps will give suggestions to user and also suggests to consult
nearby doctors.
Tools: Python, db sqlite 3.0, pytorch
Figure: Detecting ASD using machine learning
9.Thesis/Project Timeline
10. Conclusion
Machine learning can play a very important role in improving the effectiveness of behavioral
health screeners. We have achieved a significant improvement over established screening tools
for autism in children as demonstrated in a multi-center clinical trial. We have also shown some
important pitfalls when applying machine learning in this domain, and quantified the benefit of
applying proper solutions to address them.
11. Reference
[1] U. Frith and F. Happé, “Autism spectrum disorder,” Current biology,
vol. 15, no. 19, pp. R786–R790, 2005.
[2] WHO, Autism spectrum disorders, 2017 [Accessed August 22, 2018].
[Online]. Available: http://www.who.int/news-room/fact-sheets/detail/
autism-spectrum- disorders
[3] J. A. Cruz and D. S. Wishart, “Applications of machine learning in
cancer prediction and prognosis,” Cancer informatics, vol. 2, 2006.
[4] N. S. Khan, M. H. Muaz, A. Kabir, and M. N. Islam, “Diabetes
predicting mhealth application using machine learning,” in 2017 IEEE
International WIE Conference on Electrical and Computer Engineering
(WIECON-ECE). IEEE, 2017, pp. 237–240.
[5] D. P. Wall, R. Dally, R. Luyster, J.-Y. Jung, and T. F. DeLuca, “Use
of artificial intelligence to shorten the behavioral diagnosis of autism,”
PloS one, vol. 7, no. 8, p. e43855, 2012.
[6] D. Bone, S. L. Bishop, M. P. Black, M. S. Goodwin, C. Lord, and S. S.
Narayanan, “Use of machine learning to improve autism screening and
diagnostic instruments: effectiveness, efficiency, and multi-instrument
fusion,” Journal of Child Psychology and Psychiatry, vol. 57, 2016.
[7] C. Allison, B. Auyeung, and S. Baron-Cohen, “Toward brief “red flags”
for autism screening: the short autism spectrum quotient and the short
quantitative checklist in 1,000 cases and 3,000 controls,” Journal of the
American Academy of Child & Adolescent Psychiatry, vol. 51, 2012.
[8] F. Thabtah, “Autism spectrum disorder screening: machine learning
adaptation and dsm-5 fulfillment,” in Proceedings of the 1st International
Conference on Medical and Health Informatics 2017. ACM, 2017.
[9] F. Hauck and N. Kliewer, “Machine learning for autism diagnostics:
Applying support vector classification.”
[10] B. van den Bekerom, “Using machine learning for detection of autism
spectrum disorder,” 2017.
[11] D. Wall, J. Kosmicki, T. Deluca, E. Harstad, and V. Fusaro, “Use of
machine learning to shorten observation-based screening and diagnosis
of autism,” Translational psychiatry, vol. 2, no. 4, p. e100, 2012.
[12] A. S. Heinsfeld, A. R. Franco, R. C. Craddock, A. Buchweitz, and
F. Meneguzzi, “Identification of autism spectrum disorder using deep
learning and the abide dataset,” NeuroImage: Clinical, vol. 17, 2018.
[13] W. Liu, M. Li, and L. Yi, “Identifying children with autism spectrum
disorder based on their face processing abnormality: A machine learning
framework,” Autism Research, vol. 9, no. 8, pp. 888–898, 2016.
[14] D. Bone, M. S. Goodwin, M. P. Black, C.-C. Lee, K. Audhkhasi,
and S. Narayanan, “Applying machine learning to facilitate autism
diagnostics: pitfalls and promises,” Journal of autism and developmental
disorders, vol. 45, no. 5, pp. 1121–1136, 2015.
[15] J. Kosmicki, V. Sochat, M. Duda, and D. Wall, “Searching for a minimal
set of behaviors for autism detection through feature selection-based
machine learning,” Translational psychiatry, vol. 5, no. 2, p. e514, 2015.
[16] F. Thabtah, “UCI machine learning repository,” 2017. [Online].
Available: https://archive.ics.uci.edu/ml
[17] T. Booth, A. L. Murray, K. McKenzie, R. Kuenssberg, M. O’Donnell,
and H. Burnett, “An evaluation of the aq-10 as a brief screening
instrument for asd in adults.”
[18] A. Liaw, M. Wiener et al., “Classification and regression by randomfor-
est,” R news, vol. 2, no. 3, pp. 18–22, 2002.