Researchers use computer vision for early detection of autism  

The research is currently using records of 100 children diagnosed with ASD and 40 children without ASD as a comparative group.
| Photo Credit: Getty Images/iStockphoto

The researchers from the International Institute of Information Technology Bangalore (IIITB), along with the Centre for Advanced Research and Excellence in Autism and Developmental Disorders, St. John’s Hospital, are currently running a study that aims to help in the early detection of Autism Spectrum Disorder (ASD) via computer vision. The study was showcased at the Bengaluru Tech Summit held between November 19–21. 

Through the project, the researchers aim to detect ASD automatically in children in the age group of 18 to 42 months by analysing their behaviour. “Early detection of ASD is more challenging. Some key aspects are different in the Indian context when compared to similar work done in the Western world. That is why we designed a new play protocol which is specifically developed for the Indian context,” said Dinesh Babu Jayagopi, HoD, Data Science and Artificial Intelligence, IIITB. 

The protocol has both structured and unstructured activities. While the children are provided with explicit instruction about the tasks they should perform in structured activities, they are free to engage in play or explore independently without any specific directions in unstructured activities. The activities are performed in a room fitted with cameras, and the interactions between the child, parent and examiner (doctor) are recorded.  

The videos are then analysed to identify and recognise 26 behaviours of interest across all three behavioural domains (unlike previous research studies which had considered single domain) – facial expression–based, social communication-based – and play-based – with the help of deep learning techniques.  

“The computer will analyse if there are any repeated activities if the children make eye contact and such other patterns. Instead of any manual intervention, computer vision will be used. We are building models for it, and we are also ensuring that the privacy of the children is protected,” said Prof. Babu.  

These behavioural markers are used to enhance diagnostic accuracy and create a dataset that will help to infer if a child is autistic or not systematically. The research is currently using records of 100 children diagnosed with ASD and 40 children without ASD as a comparative group. Around 50 – 100 children have been subjected to this study, which has been in the works for the last couple of years.  

The system has achieved ASD prediction accuracy averaging 82%, according to the researchers. However, as the studies have just been conducted in limited settings so far, it has to be conducted outside on a larger scale to be validated. “We want to make the system as simple as possible that even in smaller towns, the tests can be run by ASHA workers,” Prof. Babu said.  

Prof. Babu has been working with a post doctoral researcher, a Ph.D researcher and Shyam Rajagopalan from Institute of Bioinformatics and Applied Biotechnology (IBAB) on this study which is funded by Machine INtelligence and RObotics (MINRO) Center at IIITB.