Researchers employ computer vision for early autism detection

A groundbreaking study leveraging computer vision aims to facilitate early detection of Autism Spectrum Disorder (ASD) in children aged 18 to 42 months. Conducted by researchers at the International Institute of Information Technology Bangalore (IIITB) in collaboration with St. John’s Hospital, the project seeks to analyse behavioural patterns for automated diagnosis. The research was showcased at the Bengaluru Tech Summit held from November 19–21.

Also read: Karnataka: Glucose supplied found having black fungus; hospitals warned

The team, led by Prof. Dinesh Babu Jayagopi, Head of Data Science and Artificial Intelligence at IIITB, developed a novel play protocol tailored to the Indian context. "Early detection is challenging, particularly in India. This protocol includes structured and unstructured activities designed to capture relevant behaviours," Prof. Babu explained.

In structured tasks, children follow explicit instructions, while unstructured activities allow free play, all recorded by cameras in a specially designed room. Using deep learning, the videos are analysed for 26 behavioural markers across facial expressions, social communication, and play, surpassing previous studies that focused on single domains.

The system, currently trained on data from 100 ASD-diagnosed children and 40 neurotypical children, achieves an average prediction accuracy of 82%. "We aim to make this system accessible even in rural areas, enabling ASHA workers to conduct tests," Prof. Babu added.

The project, funded by IIITB's MINRO Center, is poised for larger-scale validation to ensure its effectiveness in diverse settings. Researchers hope this tool will enhance diagnostic accuracy and expand access to early intervention.