Cogniable: Revolutionizing Autism Treatment

Cogniable was founded out of the need to provide healthcare for autistic patients. Recently awarded a special mention in the 2022 global UNICEF EdTech Awards, we talked with Professor Prathosh AP, Chief Engineer, Cogniable, to understand how they are revolutionising Autism treatment using AI.

Genesis

In 2017, Professor Prathosh joined IIT Delhi as a new faculty. This was when he met Mr. Manu Kohli (Cogniable's current CTO), who had returned to India from the west to pursue a part-time Ph.D. at the Department of Management Studies (DMS IIT Delhi). Professor Prathosh was giving a talk about his previous projects at Xerox, which consisted of estimating human body vitals, using video cameras, and applying them to paediatric use cases.

Mr. Kohli and his wife Swati were parents of a child with Autism, so they were well versed with issues that arise during the treatment. They had already established a physical therapy centre in Gurgaon for children diagnosed with Autism to help them acquire some social and motor skills. Fascinated by Professor Prathosh's talk, Mr. Manu approached him with the possibility of working together, inviting him to take a look at the physical therapy centre he had built. On visiting, Professor Prathosh realised that many things being done manually at the centre could be automated using state-of-the-art computer vision and deep learning techniques. This was the birth of the company Cogniable, and a journey had begun.

How does it work?

Autism is a spectrum disorder, Which means that even though we club various symptoms under the umbrella of Autism, patients differ vastly from each other. Therefore, this disease requires personalised treatment, and there is no one-size-fits-all solution. This is also why we cannot scale up Autism treatment facilities rapidly, which proves to be very costly.

One of the ways to treat Autism is via Applied Behavior Analysis (ABA) therapy. This therapy requires a therapist to engage with the child. They set a goal for the child, then try to achieve it with the child using positive reinforcements. For example, touching and identifying the nose or maintaining eye contact can be a goal, while toys, sweets, or encouragement can be a positive reinforcement.

Professor Prathosh realised that this therapy could be conducted using computer vision, where a couple of cameras can monitor the child's movements, and an untrained parent/caregiver can provide positive reinforcement when the app says so.

In his paper ¹Guided Weak Supervision for Action Recognition with Scarce Data to Assess Skills of Children with Autism, Professor Prathosh and his team propose to automate the response measurement through video recording of the scene following the use of Deep Neural models for human action recognition from videos. However, supervised learning of neural networks demands large amounts of annotated data that is hard to come by. This issue is addressed by leveraging the 'similarities' between the action categories in publicly available large-scale video action datasets and the dataset of interest. A technique called 'Guided Weak Supervision' is proposed, which is used to create the target data from the source data.

Through this paper, Professor Prathosh and his team show how this technique can build a scalable environment where therapies can be conducted at lower costs and without manual intervention.

Data-Driven Approach

The decision of how the therapy has to progress for a given child is a subjective thing. There are no concrete guidelines of what is to be done. The path to learning something under a curriculum is different for each child. A child may learn a task slowly while other children do it faster, while the same child may finish other tasks quicker than other children. Even the pace of the same child can vary with time, as there may be a learning curve at play here.

Therefore, the learning pace of each child is different, and they have different needs. Cogniable takes this into account, taking a data-driven approach, where the child's performance is taken into account while deciding the subsequent therapy sessions. This is beneficial for the child mainly because there are goals that are easy to achieve with some children while difficult with others.

What's Next

Currently, Cogniable focuses exclusively on the autism treatment facilities, improving upon the computer vision applications. Right now, the app requires a controlled environment where the cameras must be put in certain positions and orientations. The goal is to relax these constraints and make the application more robust. The treatment for Autism is currently only available in major cities such as Delhi and Mumbai. The treatment also proves costly because the therapy sessions require the therapist to always be available. Cogniable seeks to spread its facilities across India to make autism treatment easier to access.

Conclusion

Autism Treatment has always been a problem due to its unique nature from patient to patient. Cogniable's success can be attributed to the multidisciplinary research conducted by psychologists, behavioural therapists, and engineers. Cogniable's approach to this problem with deep learning is an excellent way to handle this uniqueness and serve patients on a large scale.