ICVGIP 2025 Paper ID 176 - Severity Grading of Autism Spectrum Disorder from Eye Gaze скачать в хорошем качестве

ICVGIP 2025 Paper ID 176 - Severity Grading of Autism Spectrum Disorder from Eye Gaze

Severity Grading of Autism Spectrum Disorder from Eye Gaze Scanpath Trajectory using Deep Learning "With the increasing prevalence of Autism Spectrum Disorder, determining the severity of ASD has become a pivotal research focus. It plays a key role in designing personalized therapies and tracking treatment progress alongside early detection. In this study, a deep learning-based framework is introduced for classification of ASD severity using eye gaze data. A novel scanpath generation method has been developed first by detecting the facial landmarks of the eye region from the eye tracking videos of patients frame-by-frame. The mean coordinates of eye landmarks per frame are computed next and sequentially connected to form a temporally ordered gaze trajectory. A deep neural network, with backbone of VGG16 architecture, has been trained and validated thereafter using the extracted scanpath images for final severity classification. This proposed model has achieved a test accuracy of $0.9677$. Additional metrics such as precision ($0.9722$), recall ($0.9688$), F1-score ($0.9686$), and AUC ($1.000$) have further confirmed the effectiveness of model. This work introduces a clinically interpretable and non-invasive method feasible of low-cost webcam-based scanpath imaging for ASD severity grading in an Indian pediatric cohort on real-time data. The proposed method presents a promising approach as a supportive adjunct to ISAA-based severity grading by offering an additional data-driven perspective from eye gaze scanpath trajectories, facilitating the monitoring of therapeutic progress, particularly in diverse and low-resource settings." Authors: Debmani Saha