MAJOR PROJECT REVIEW

DETECTING AUTISM DISORDER USING COMPUTER

VISION

PRESENTED BY: TEAM 14

B. APOORVA 19071A0506
M. KIRANVINI 19071A0535
B. HARI BHARATH CHANDRA 20075A0501
M. LAVANYA 20075A0503
UNDER THE GUIDANCE OF:
Mrs. S. JAHNAVI
ASSISTANT PROFESSOR
Dept. OF COMPUTER SCIENCE ENGINEERING
AGENDA
● Abstract
● Literature Survey
● Proposed system
● Objectives
● System Requirements
● System Architecture
● Modules
● UML Diagrams
● Implementation
● Test Cases
● Status of Work
● Conclusion
● Future Scope
● References
ABSTRACT

Autism Spectrum Disorder (ASD) is a neurological and developmental disorder that begins in early
childhood and lasts throughout a person’s life. Approximately 1/100 children are diagnosed with
autism spectrum disorder around the world. Specialists believe that people with ASD have distinct
facial features that can be used to help diagnose their ASD and even correlate with ASD severity. This
correlation can be used to train a CV model to detect ASD using those unique facial features. The goal
is to provide a preliminary diagnostic tool that can aid people in their decision to pursue further ASD
testing.
LITERATURE SURVEY
S.No AUTHORS TITLE OF THE PAPER AND YEAR FINDINGS AND DRAWBACKS
OF PUBLICATION

1 Kaushik Vakadkar, Detection of Autism Spectrum Disorder in We begin by preprocessing the dataset to
Diya Purkayastha, Children Using Machine Learning Techniques eliminate missing values and outliers. Once
Deepa Krishnan (2021) the data set has been preprocessed, ML
classification algorithms like Random Forest,
SVM, KNN, Logistic Regression. are used to
predict the output label (ASD or no ASD).
Drawbacks: the scarce availability of large
and open-source ASD datasets.

2 Mindi Ruan Image and Video-based Autism spectrum Model- Based on photographs taken by
disorder detection via deep learning. (2020) participants in a facility. The research is based
on first person perspective from which
behavior can be deduced.
Drawbacks: Expensive and cooperation from
people in the facility.
S.No AUTHORS TITLE OF THE PAPER AND FINDINGS AND DRAWBACKS
YEAR OF PUBLICATION

3 Milan N Parikh, Hailong Enhancing diagnosis of autism with Models used: ML Algorithms: Random Forest,
Li, Lili He optimized Machine Learning models and SVM, KNN, Logistic Regression.
personal characteristic data Developed and compared nine machine
(2019) learning models for ASD classification by
using PCD as input features.
Drawbacks: less accurate, size of dataset,
heterogeneity which can compromise the ML
models for accuracy.

4 Sunghye Cho, Mark Automatic detection of Autism Spectrum Participants: 35 kids(13 female) with ASD and
Liberman, Neville Ryant, Disorder in children using acoustic and text 35 kids with TD(11 female)
Meredith Cola, Robert T. features from brief natural conversations Methods: Biosensor recordings and PCA
Schultz and Julia Parish- (2019) Detection was based on acoustic and text
Morris features extracted from the children’s
conversations.
Drawbacks: Other important features like
topicality, relevance of the topic in
conversation, appropriateness and
conversational flow etc. are not captured by the
automated system.
S.No AUTHORS TITLE OF THE PAPER AND FINDINGS AND DRAWBACKS
YEAR OF PUBLICATION

5 Haishuai Wang, Li Li, Autism Screening using Deep Embedding Participants: 702 persons or children of whom
Lianhua Chi, and Ziping Representation 513 have ASD
Zhao (2019) Method : Algorithms were created via a robust
deep learning classifier and deep embedding
representation for categorical variables to
diagnose ASD.
Drawbacks: Data is taken from screening tests
taken by the guardian or parent- this may
involve a lot of bias.

6 Julia Parish-Morris, Oral-Motor and Lexical Diversity During Statistical Analysis model based on CASS
Evangelos Sariyanidi, (Contextual assessment of Social Skills)
Naturalistic Conversations in Adults with
Casey Zampella, Drawbacks: Affected by top down factors like
G. Keith Bartley, Autism Spectrum Disorder (2018) motivation, interests etc. Also the experiment is
Emily Ferguson, not so reliable as it is conducted in a very
Ashley A. Pallathra, restricted environment.
Leila Bateman,
Samantha Plate,
Meredith Cola,
Juhi Pandey,
Edward S. Brodkin,
Robert T. Schultz,
Birkan Tunç
S.No AUTHORS TITLE OF THE PAPER AND FINDINGS AND DRAWBACKS
YEAR OF PUBLICATION

7 Amir Hossein Memari, Children with Autism Spectrum Disorder and Research was conducted using the data derived
Nekoo Panahi, Elaheh Patterns of Participation in Daily Physical from checklists based on Godin-Shephard
Ranjbar, Pouria and Play Activities (2015) Leisure Time Questionnaire and Autism
Moshayedi, Masih Shafiei, Treatment Evaluation Checklist.
Ramin Kordi, Vahid Ziaee Drawbacks: Disruption from other Non-ASD
related barriers like household structure,
household income etc

8 Quentin Guillon, Visual social attention in autism spectrum Model : Usage of eye tracking systems to
Nouchine Hadjikhani, disorder: Insights from eye tracking studies examine the visual social attention of ASD
Sophie Baduel, Bernadette (2014) patients.
Rogé. Drawbacks: whether the findings of these
studies generalize to real life situations remains
to a large extent unknown
S.No AUTHORS TITLE OF THE PAPER AND FINDINGS AND DRAWBACKS
YEAR OF PUBLICATION

9 María Elena Sampedro-Tobón, Early detection of autism spectrum Model: Statistical Analysis, CDT.
Manuela, disorders: a responsible A specifically designed semi-structured
Susana Vélez-Vieira. decision for a better prognosis interview with the child's(ages between 2.6
(2013) and 10.75 years) parents was conducted.
Drawbacks: the amount of professionals
visited, cost, time of adjustment, acceptance
of this reality and the anguish of the families
during the diagnostic process.

10 Suniti Chakrabarti, MD; A Model of Early Detection and Age group tested: 6 weeks to 3 years
Christine Haubus, RLCST; Diagnosis of Autism Spectrum children.
Sally Dugmore, RGN; Gill Disorder in Young Children Model: Universal screening/surveillance,
Orgill, RGN; Frank Devine, (2005) CDT.
RMN Children identified with a developmental
problem.
Drawbacks: many health care workers,
professional specialists are involved to detect
autism; which can take lots of money and
time.
EXISTING SYSTEM
● Autism Diagnostic Observation Schedule (ADOS): It is a standardized diagnostic assessment tool used to
evaluate and diagnose ASD. It is considered one of the most reliable and valid diagnostic instruments for
ASD. ADOS is a semi-structured assessment that involves a trained examiner observing and interacting with
an individual suspected of having ASD.
● Autism Diagnostic Interview-Revised (ADI-R): It is a semi-structured diagnostic interview used to assess
autism spectrum disorders in individuals of all ages, but particularly in children. It is one of the most widely
used tools for diagnosing autism and is often used in conjunction with other diagnostic tools, such as the
ADOS.
● Modified Checklist for Autism in Toddlers (M-CHAT): It is a screening tool used to identify children who
may be at risk for autism spectrum disorder (ASD). It is a questionnaire that is typically filled out by parents
or caregivers and is designed to be administered to children between the ages of 16 and 30 months.
● Childhood Autism Rating Scale (CARS): It is a tool used by healthcare professionals to help diagnose ASD
in children. It is a 15-item rating scale that assesses behaviors in the areas of social interaction,
communication, and behavior. The items are rated on a 4-point scale, with higher scores indicating greater
severity of autism symptoms.
PROPOSED SYSTEM
In the proposed system, the detection of autism disorder is performed using Computer Vision rather than
Machine Learning Algorithms.

Autism spectrum disorders (ASD) are caused by anomalies in the embryological brain and so it is implied that
children with ASD have facial development that differs significantly from that of typically developing children.
Some of the frequent facial features of autism are a broader upper face, shorter middle face, wider eyes, bigger
mouth, and the philtrum.

This association can be used to create and train a VGG based Facial Recognition model, that assesses if a person
has ASD. VGG models are easy to use for feature representation Transfer learning, where the source and the
target domains look for similar features (Feature extraction) but make different inferences based on those
features. Finally Statistical Analysis is performed to understand the results and predict more events .
OBJECTIVES
The main objective is to provide a preliminary diagnostic tool that can aid people in their decision to pursue
further ASD testing using Computer Vision.

● Early detection: One of the main objectives is to identify children with ASD at an early stage, which
can lead to early intervention and better outcomes.
● Objective assessment: Computer vision can provide objective measures of facial features that are
characteristics of ASD, which can help clinicians make accurate diagnoses and monitor progress over
time.
● Non-invasive assessment: Computer vision techniques can provide a non-invasive method for
assessing children with ASD, which can be less stressful and more comfortable for the child.
● Automation: Computer vision techniques can automate the process of assessing children for ASD,
which can save time and resources for clinicians and researchers.
SYSTEM REQUIREMENTS
Software Requirements: Hardware Requirements:

● Windows 10 & above version ● i5 Processor

● Python ● 8 GB RAM ( 16GB RAM for higher
● Google Colaboratory Performance)
● Gradio ● 64 bit OS
● Python Libraries
● GPU
SYSTEM ARCHITECTURE
MODULES

● Data Collection

● Data Pre-processing

● Generation of training and testing datasets

● Generation of the model

● Evaluating the model

● Statistical Analysis
UML DIAGRAMS

● Use Case Diagram

● Class Diagram

● Sequence Diagram

● Activity Diagram
Use Case Diagram
Class Diagram
Sequence Diagram
Activity Diagram-1

Activity Diagram for Developer

Activity Diagram-2

Activity Diagram for EndUser

Activity Diagram-3

Activity Diagram for Dataset

Activity Diagram-3

Activity Diagram for Model

STATUS OF WORK
Work Time

Literature Survey 2 weeks (Completed)

Dataset Collection 4 weeks (Completed)

Design 4 weeks (Completed)

Implementation 4 weeks (Completed)

Documentation 2 weeks(Completed)
Implementation
Implementation
Implementation
Implementation
Implementation
Implementation
Implementation
Implementation
IMPLEMENTATION
OUTPUT
OUTPUT
TEST CASES
S.No TestCase_ID Input_Img Expected_Result Actual_Result Status

1 tc001 PASS
Autistic Autistic

2 tc002 FAIL
Autistic Non-Autistic

3 tc003 PASS
Autistic Autistic
TEST CASES
S.No TestCase_ID Input_Img Expected_Result Actual_Result Status

4 tc007 FAIL
Non_Autistic Autistic

5 tc008 PASS
Non_Autistic Non_Autistic

6 tc009 PASS
Non_Autistic Non_Autistic
Conclusion
Autism Spectrum Disorder (ASD) is a neurological and developmental disorder that begins in early
childhood and lasts throughout a person’s life. Approximately 1/100 children are diagnosed with
autism spectrum disorder around the world. It goes unnoticed most of the time. So, by diagnosing it
as early as possible is crucial. This project presents VGG model that predicts whether the child has
ASD or not based on facial features like broader upper face, shorter middle face, wider eyes, bigger
mouth, and the philtrum.
Future Scope

● Extending the project for adults

● Increasing the accuracy

● Optimization for low end devices

● Dealing with erroneous inputs

REFERENCES
[1] Vakadkar, Kaushik, Diya Purkayastha, and Deepa Krishnan. "Detection of Autism Spectrum Disorder in
Children Using Machine Learning Techniques." SN Computer Science 2.5 (2021): 1-9.
[2] Ruan, Mindi, "Image and Video-Based Autism Spectrum Disorder Detection via Deep Learning" (2020).
Graduate Theses, Dissertations, and Problem Reports. 7818.
https://researchrepository.wvu.edu/etd/7818
[3] Parikh, Milan N., Hailong Li, and Lili He. "Enhancing diagnosis of autism with optimized machine learning
models and personal characteristic data." Frontiers in computational neuroscience 13 (2019): 9.
[4] Cho, Sunghye, Mark Y. Liberman, Neville Ryant, Meredith Cola, Robert T. Schultz and Julia Parish-Morris.
“Automatic Detection of Autism Spectrum Disorder in Children Using Acoustic and Text Features from Brief Natural
Conversations.” INTERSPEECH (2019)
REFERENCES

[5] H. Wang, L. Li, L. Chi, and Z. Zhao. 2019. “Autism Screening using Deep Embedding Representation.”
International Conference on Computational Science.
[6] Parish-Morris, Julia & Sariyanidi, Evangelos & Zampella, Casey & Bartley, G. & Ferguson, Emily &
Pallathra, Ashley & Bateman, Leila & Plate, Samantha & Cola, Meredith & Pandey, Juhi & Brodkin, Edward & Schultz,
Robert & Tunc, Birkan. (2018). Oral-Motor and Lexical Diversity During Naturalistic Conversations
Adults with Autism Spectrum Disorder. Proceedings of the conference. Association for Computational Linguistics. North
American Chapter. Meeting. 2018. 147-157. 10.18653/v1/W18-0616.
[7] Memari, A. H., Panahi, N., Ranjbar, E., Moshayedi, P., Shafiei, M., Kordi, R., & Ziaee, V. (2015). Children
with autism spectrum disorder and patterns of participation in daily physical and play activities. Neurology research
international, 2015.
REFERENCES
[8] Guillon, Quentin, Nouchine Hadjikhani, Sophie Baduel, and Bernadette Rogé. "Visual social attention in
autism spectrum disorder: Insights from eye tracking studies." Neuroscience & Biobehavioral Reviews 42 (2014): 279-297.
[9] Sampedro-Tobón, María Elena, Manuela González-González, Susana Vélez-Vieira, and Mariantonia Lemos-
Hoyos. "Early detection of autism spectrum disorders: a responsible decision for a better prognosis." Boletín médico del
Hospital Infantil de México 70, no. 6 (2013): 456-466.
[10] Chakrabarti, Suniti, Christine Haubus, Sally Dugmore, Gill Orgill, and Frank Devine. "A model of early
detection and diagnosis of autism spectrum disorder in young children." Infants & Young Children 18, no. 3 (2005): 200-211.
Thank you