Autism Spectrum Disorder Diagnosis using Deep Learning Techniques

Autism spectrum disorder (ASD) represents a neurological condition rather than a mental illness, characterized by atypical brain development that can subsequently manifest in distinct facial features. Children with ASD often exhibit unique facial landmarks that differentiate them from Typically Developed individuals. This pioneering research introduces a novel approach, focusing on the detection of ASD through the analysis of both social media data and biomedical images, specifically by harnessing the power of face recognition technology and Convolutional Neural Networks (CNN). Deep learning techniques are precise identification of these facial features. The research aims to benefit both communities and mental health professionals by providing an accessible web application based on a CNN with transfer learning and Flask integration. In this refined approach, two robust models are exclusively retained, DenseNet121 with an accuracy of 54% and EfficientNetB0 with an impressive 90% accuracy rate. These models are intended for improving the accuracy and convenience for determining ASD using face traits, which will simplify the process of early diagnosis and intervention for those who need it. The dataset contains 2,940 images of faces that were retrieved from the Kaggle platform. Standard evaluation metrics are used to evaluate the effectiveness of the two CNN models, including sensitivity, specificity and accuracy. This research hopes to help people with ASD by finding a way to identify it early. Early help is very important for making things better for people with autism. So, the study aims to create a useful tool that can tell if someone might have ASD when the patients are very young.