Mobile app for ASD detection in children from static facial features using hybrid Pakistani corpus

Background Autism Spectrum Disorder (ASD) is a neurodevelopmental condition affecting communication, social interaction, and behavior. ASD often causes delays in reaching developmental milestones. Purpose Early therapeutic intervention can aid in progress. Detection methods typically involve psychological assessments and activity analysis. Studies indicate that craniofacial abnormalities are associated with ASD, proposing facial features as a potential biomarker for early diagnosis. Research Design This study presents a deep learning-based mobile application for ASD detection, utilizing pre-trained models—InceptionV3, Xception, MobileNet, and ResNet 50—andhybrid dataset containing local (Pakistani) and available labeled images. Data Anaysis Experimental results show InceptionV3 achieving the highest accuracy of 88%, highlighting its potential for reliable ASD detection. Conclusions Our work contributes by introducing a mobile-based ASD detection system that leverages static facial features and a hybrid Pakistani corpus, offering a culturally adapted and accessible tool for early screening.