Harnessing AI to ease the stress of managing autism5 min read . Updated: 18 Oct 2020, 10:54 PM IST
The protocols for the clinical trials had to be altered after covid, as hospital committees had to approve digital collection of video data remotely
Dr Swati Kohli was an educator for special needs children. So, when her own son, Ekagra, first showed signs of being different, she consulted the head of a department at a renowned institution. He dismissed her concerns. It was only after her son turned four that he was diagnosed with autism.
It’s the first six years of an autistic person’s life when intervention has the biggest benefit. The brain’s neuroplasticity, or ability to change, is the most during that period. Vital time had gone by without therapy for Ekagra.
“After the diagnosis, we were fortunate to be able to take good care of him as we moved to Europe and the US," says Manu Kohli, Ekagra’s father, who was a global manager for multinational firms.
That was over a decade ago, but the stressful experience of delayed diagnosis and then struggling to find suitable therapy and schooling remained with the couple.
Three years ago, when Ekagra was into his teens, they moved back to India and founded Cogniable to use AI for early detection and affordable management of autism spectrum disorder (ASD). Manu Kohli also took up a PhD programme in application of AI at IIT Delhi which helped with academic collaboration. Two other co-founders of the Gurugram-based startup are Joshua Pritchard, a certified behaviour analyst in Florida who runs clinics for children with ASD, and Prathosh A.P., an assistant professor at IIT-D, who holds patents in AI and computer vision.
Cogniable is building two products. The first one uses AI analysis of video recorded during a guided play session with a child. It maps behavioural cues to criteria defined and classified in the Diagnostic and Statistical Manual of Mental Disorders (DSM), which helps a clinician detect ASD.
The Cogniable team first collected data from 37 children in India and the US for six months, then built an AI model, patented it and published their results in a paper at the Association for the Advancement of Artificial Intelligence conference (AAAI-20) in New York in February this year. The product is now undergoing clinical trials at Maulana Azad Medical College hospital and Safdarjung Hospital in Delhi.
The traditional approach of a clinician is to repetitively note the responses of a child to a variety of stimuli. Cogniable automates this by matching a child’s actions, such as eye contact, with similar actions in publicly available video datasets.
“We could teach our algorithm ‘this is poor eye contact’ and so on. Our AI model could clearly identify behavioural markers of autism in children aged between 18 months and five years," says Kohli.
This can improve the speed, accuracy and availability of autism detection. Videos of kids from remote locations can be recorded on a smartphone app and analysed. “Parents simply play with their child at home, based on a script we provide, and record that session in our app," he says.
The protocols for the clinical trials had to be altered after covid, as hospital committees had to approve digital collection of video data remotely. “About two-thirds of the data collection is done and we will have firm results on the trials in six to eight months," he says.
Cogniable’s second product is an aid for therapy that several clinics and paediatricians in India are already using. One is Dr Anjali Bangalore, developmental neurologist and director of the ICON Centre for Assisted Learning in Aurangabad.
“So many people are involved in treating autism—paediatricians, psychologists, speech and occupational therapists. Four hours of daily therapy is recommended. All this can become overwhelming and unaffordable for a parent, and even more so in small towns and rural areas where qualified personnel are very few," says Dr Bangalore.
That’s where Cogniable’s app, which Dr Bangalore has been using with 25 patients since July, has made a difference. It enables digital assessment of a child with which a clinician can quickly build a customized therapy programme. This includes long-term and short-term goals, the skills to be taught and the protocols to teach each skill along with video tutorials.
The child’s parent can log into the app to provide therapy at home, get reminders on activities to be done and measure the child’s progress. The app helps caregivers understand and cope with everyday situations, such as tantrums or communication difficulties. Most of all, it gives parents a roadmap and awareness, which reduces chances of getting ripped off by clinics or schools that charge a bomb for dubious interventions with tall claims.
“It’s a platform where objectives and outcomes can be set and a physician like me can be in touch with the parent who is also trained to be a caregiver at home. This is different from the usual experience of a parent with an autistic child running from one therapist to another. And it saves a lot on the cost of taking the child to a clinic frequently," says Dr Bangalore.
The challenge is in delivering it to every nook and corner of India, but a start has been made with interest from clinics as well as schools. “We recently signed a contract with Fortis who want to take our product to multiple clinics in India as well as Southeast Asia and Africa where early intervention services are lacking," says Kohli. “A chain of clinics in Bangladesh has signed up with us, and next month we will be launching the product in the US."
Entry into the US will be a big milestone because of the potential for much higher revenue in a market where autism intervention is worth $250 billion. It’s a market where healthcare is tied closely with insurance. This creates an additional value proposition for the quantified measurement of progress on the app, which makes insurance claims for therapy easier.
The next big milestone will be the deployment of the AI-based screening product. Last year, a study by researchers at the Children’s Hospital of Philadelphia raised questions about the accuracy of the most widely used screening tool currently—Modified Checklist for Autism in Toddlers (M-CHAT). The study, which followed 26,999 children screened between the ages of 16 and 26 months, found paediatricians using M-CHAT failed to detect autism in 60% of those later diagnosed with ASD. This makes the case for AI tools stronger. Cogniable’s main rival here is Palo Alto-based Cognoa, whose AI diagnostic tool for ASD recently completed a trial that awaits the FDA’s nod.
Cogniable’s differentiator is its deep neural AI model, which uses computer vision to analyse the actions of children in videos. Cognoa’s AI, on the other hand, uses text inputs from questionnaires filled by parents, the paediatrician and an autism specialist who watches videos of the child.
“We work with video data for AI that’s extremely challenging," says Prathosh.
The challenge also lies in training the AI model because video data of autistic children is scarce and difficult to record.