Abstract:
The process of identifying Autism Spectrum Disorder (ASD) at an early stage is challenging mainly because the condition exhibits diverse symptoms that current diagnostic techniques focus on subjective behavioral assessments, lacks objectivity. Recently, eye-tracking technology has emerged as an effective method to measure initial visual attention patterns and social interaction patterns that are associated with ASD. This work brings a comprehensive literature review on eye-tracking–based ASD detection, followed by the proposal of early screening in infants, specifically children below three years with the help of an artificial intelligence-based system. The existing literatures are studied with possible identification of gaze-based biomarkers, feature representations, experimental paradigms, and machine learning approaches to highlight the strengths and weaknesses, bring about the formation of the research gap in this area. Additionally, a conceptual framework is proposed utilizing the features of eye-tracking for detection of ASD at an early stage that incorporates modality-specific preprocessing and unimodal learning components such as fixation times, saccades and gaze patterns. Thus this study provides a systematic approach with futuristic implementation that supports clinical validation, scalability, and early diagnosis.
