Manu, with more than 15 years of experience in technical management roles, and Swati, who works with children with special needs – believed there lay a greater purpose for them following their son’s diagnosis. The couple returned to India from the USA, and set up CogniABle – machine learning based assistive technology for early detection of ASD. They were joined by Dr Prathosh AP and Dr. Joshua Pritchard. Dr. Pratosh is an assistant professor at IIT Delhi, holds multiple patents in CV and AI and leads the ML initiatives for CogniABle. Dr. Pritchard serves as the President of the Florida Association of Behaviour Analysis and also runs several schools and clinics in the USA for children with ASD. He is responsible for developing and validating evidence-based curriculum to manage autism especially in low resource setting nations like India.

Like almost every health condition, be it physical or mental, early detection can radically change outcomes and course of treatment. Statistics indicate that there are 5,000 children born every day in India. That’s 18 million a year. Out of this 18 million, nearly 1% have autism. Unlike physical disorders, a condition like ASD is a developmental disorder that affects speech and communication – and there aren’t exactly a battery of tests available to screen for milestones, although the Indian Academy of Paediatrics, all children should be screened using standard autism detection tools before 24 months of age.

“A child’s milestones have to be observed. Has he turned on his side at the right age? Is he making eye contact adequately and tracking stimuli like light and sound? There are so many of these milestones that indicate if a child’s brain is growing properly. This is why early detection in ASD is so critical because the brain’s neuroplasticity is at its highest until the age of 6. If symptoms are detected, then early intervention can go a long way in determining the quality of outcomes for the child,” explains Kohli.

This is where technology comes in. Being an avid believer in the power of technologies to solve problems at scale, and combining his wife’s extensive knowledge in the area of developmental delays in children, Kohli and his team set out to develop a comprehensive screening platform. Parents can upload videos of their children using the CogniABle mobile application, and deep learning models work to identify gross motor skills, fine motor skills and other actions. Kohli did a trial on 37 children while at IIT Delhi. “We saw a lot of latency in responses. The video datasets were annotated by humans. We built a model to gauge responses in children, and it was around 87% accurate. Since then, we have been using publicly available data to feed the AI engine and extrapolate useful insights from the same,” adds Kohli. The global screening standards for autism recommends a 115-item checklist, but just 12-13 from this list is sufficient to gauge a child’s developmental deficits, if any. Using computer vision, team CogniABle has developed transfer learning techniques, which allows the team to make ML models with about 25% of video data. The team has filed two patents for this, adds Kohli.

Autism screening is a one-time feature, but the real journey starts after. The team, with inputs from seven behavioural psychologists and four technical experts, provides personalized intervention plans for children spanning areas like language, social skills and behaviour. “Behavioural therapies are key to helping children overcome barriers in social settings and environments. Moreover, our platform is also addressing the high costs involved in procuring treatment for those with ASD, and assisting parents and educators with state of the art curriculum at an affordable price,” says Kohli.