A Lightweight Convolutional Neural Network for High-Accuracy Autism Spectrum Disorder Detection from Facial Images

Abstract

Autism spectrum disorder (ASD) is a neurodevelopmental disorder with diverse and complex manifestations. This condition requires early detection which is crucial in finding a cure. Traditional diagnostic procedures are biased and time-consuming. These create a pressing need for objective and accessible screening tools. The rapid development of artificial intelligence, particularly deep learning (DL) methods, offers promising resolutions for automating ASD detection. Such detection is done through analyzing behavioral and physical biomarkers. This study proposes a convenient but complex neural network to support early ASD screening through facial images. The model classifies facial images of children as autistic or non-autistic by using the publicly available Autistic Children Facial Dataset. Various experiments assess the model’s performance. The design is developed to be statistically efficient, which makes it suitable for mobile and embedded systems for real-time screening. Results show that our proposed model shows better performance than the existing approaches, especially in the levels of accuracy and area under curve (AUC). This further confirms its efficiency and prospects for autism screening applications. Achieving accuracy of 98.97%, the model delivers better performance than the evaluated benchmark models. These findings affirm that lightweight image-focused DL approaches are well suited to provide clinicians with a trustworthy and scalable method for early ASD screening. Thus, it is evident that the proposed model facilitates timely support for individuals and families.