Deep Learning for Autism Diagnosis and Facial Analysis in Children

Area of Science:

• Computer Science
• Medical Imaging
• Developmental Pediatrics

Background:

• Autism Spectrum Disorder (ASD) is characterized by challenges in social skills, communication, and repetitive behaviors.
• While genetic factors are implicated, diagnosis often relies on behavioral and facial feature analysis.
• Distinct facial patterns in individuals with autism present an opportunity for image-based diagnostic approaches.

Purpose of the Study:

• To develop and validate a deep learning model for classifying children as healthy or potentially having autism using facial images.
• To investigate the efficacy of using facial feature analysis for cost-effective and efficient autism screening.
• To explore the potential of AI in identifying other diagnosable conditions through similar image analysis techniques.

Main Methods:

• A deep learning model integrating MobileNet and two dense layers was employed for feature extraction and image classification.
• The model was trained and tested on a dataset of 3,014 images, with a 90:10 split for training and testing, respectively.
• The dataset comprised an equal number of images from children with and without autism.

Main Results:

• The deep learning model achieved a classification accuracy of 94.6% in distinguishing between children with and without autism.
• Facial image analysis proved to be a viable method for autism classification.
• The results suggest that AI-driven image analysis can be a powerful tool for medical diagnosis.

Conclusions:

• Diagnosis of autism can be effectively achieved using only facial images with high accuracy.
• This AI-powered approach offers a potentially cheaper and more efficient alternative to traditional diagnostic methods.
• The methodology may be applicable to the diagnosis of other conditions with recognizable facial features.