Facial Features Detection System To Identify Children With Autism Spectrum Disorder: Deep Learning Models

Area of Science:

• Neurology
• Computer Science
• Medical Imaging

Background:

• Autism spectrum disorder (ASD) is a neurodevelopmental condition impacting brain development and facial features.
• Distinct facial feature patterns differentiate children with ASD from typically developed (TD) children.

Purpose of the Study:

• To create an accessible deep learning-based web application for diagnosing ASD using facial analysis.
• To assist families and psychiatrists in early autism detection.

Main Methods:

• Utilized a convolutional neural network with transfer learning and a Flask framework.
• Employed pretrained models: MobileNet, Xception, and InceptionV3 for image classification.
• Trained models on a public dataset of 3,014 facial images (1,507 ASD, 1,507 TD).

Main Results:

• MobileNet achieved 95% classification accuracy on validation data.
• Xception attained 94% accuracy, while InceptionV3 reached 89% accuracy.
• The deep learning approach demonstrated high efficacy in distinguishing facial features associated with ASD.

Conclusions:

• The developed web application provides a promising, non-invasive tool for autism detection.
• Deep learning models, particularly MobileNet, show significant potential for identifying ASD through facial feature analysis.
• This technology can aid in early diagnosis and intervention for children with autism spectrum disorder.