Identification of Autism Spectrum Disorder using Residual Attention Network for Facial Image Analysis.

Abstract

C. Gnanaprakasam, Manoj Kumar Rajagopal

The goal of the present paper is to apply deep learning algorithms to identify autism spectrum disorder (ASD) patients from a face image dataset based solely on the patient's face activation patterns. We investigated ASD Patients' face imag- ing data from a worldwide multispecialty database known as Autism Face Imaging Data Exchange. ASD is a brain-based disorder characterized by social deficiency and the symptoms are different scenarios. According to recent Centers for Disease Control data, ASD affects About 1 in 54 children who have been identified with autism spectrum disorder (ASD) according to estimates from CDC’s Autism and Developmental Disabilities Monitoring (ADDM) Network. We investigat- ed patterns of functional connectivity that objectively identify ASD participants from functional face imaging data and attempted to unveil the facial patterns that emerged from the classification. With the proposed module, standard CNNs are made, like ResNet-50 have more discriminative power for deep face recognition, and results improved the state-of-the-art by achieving 99% accuracy in the identification of ASD versus control patients in the dataset. We present the results and identify the areas of facial expressions that contributed most to differentiating ASD from typically developing controls as per our deep learning method. For verification purposes, the videos collected in real-time manually from different chil- dren we retested and an accuracy score of 99.90% and an F1 score of 99.67% were achieved.

PDF