Attention-Driven Deep Learning Approach for Autism Spectrum Disorder Detection using Eye-Tracking Scanpath Representation
Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that manifests itself through unusual social interactions and visual attention. Thus, it is of prime importance to diagnose the condition correctly. In this context, this paper proposes a deep learning-based framework for the detection of Autism Spectrum Disorder by utilizing eye-tracking images. The structured framework of the paper is as follows: the images are first preprocessed through a structured framework involving resizing, min-max normalization, and Otsu threshold-based segmentation. Then, rotation-based data augmentation is performed on the images. Finally, the EfficientNet-B4 network is used in combination with Squeeze-and-Excitation (SE) and Convolutional Block Attention Module (CBAM) for the detection of ASD. The experimental results of the framework have shown that the framework has achieved a classification accuracy of 98.1%, thereby outperforming than other state-of-the-art techniques.