Cognoa’s device is the first FDA-authorized diagnosis aid designed to help primary care physicians diagnose autism in young children with the goal of enabling earlier interventions


In June, the FDA granted a De Novo classification request to Canvas Dx, a machine learning-driven autism diagnostic solution, developed by pediatric behavioral health company Cognoa. The AI-based device is the first FDA-authorized diagnosis aid designed to help physicians to diagnose autism in the primary care setting.

Launched in 2014 and gaining regulatory recognition as a Class II diagnostic medical device for autism in 2018, Canvas Dx was created based on the research work of Dennis P. Wall.

The Stanford Medical School Associate Professor of Pediatrics, Psychiatry and Biomedical Data Sciences found preliminary success leveraging machine learning to diagnose autism using behavioral phenotype, captured by inputs from three modalities. A questionnaire to be answered by the child’s caregiver; a questionnaire for an analyst who watches short videos of the child in his/her natural environment recorded by the caregiver, and a questionnaire for the primary care physician.

“The Canvas Dx algorithm analyzes all three inputs in a complex and nonlinear manner and looks for subtle patterns of behaviors that are indicative of autism spectrum,” says Halim Abbas, Cognoa’s Chief AI Officer. “It would then give a reliable and deterministic answer for the healthcare provider to decide if that’s a suggestion of positive, negative or inconclusive of autism.”

The challenges of diagnosing autism

Autism diagnosis has always been done using empirically validated scoring systems or instruments that require exmaining a child for a lengthy period and in a very controlled setting. “With things like wallpapers and toys, all must be displayed and arranged in certain ways,” Abbas continues. “As such, assessments can only be done in certain clinics by certain people. All these make things cumbersome and costly.”

“Autism is a heterogenous disorder with children presenting with a range of symptoms,” adds Dr. Shrief Taraman, Cognoa’s Chief Medical Officer. “The physician is a stranger, especially when the children are at the clinic for the very first time. Their behaviors may not coincide with what the parents observed at home. Also, autism spectrum disorder overlaps with other developmental conditions. That’s why we need long observations to avoid having a skewed understanding of what happens. Having a clear guidance helps in the understanding of comorbidities and how autism is manifested differently across gender, ethnicities, and age.”

Personal biases may permeate as physicians translate what has been observed by the child’s caregiver into these scoring sheets. “I remember a parent at one of the forums. He was telling the physician X, Y, Z about his child’s development,” recalls Dr. Taraman. “The physician not only rejected those claims but also indicated things that the child hadn’t done to adhere to the format and items on the scoring sheet.

“The beauty of having a machine learning based solution is to quantify the problem, pull all subtle features together, make sense of ambiguities, minimize variabilities, generate well-defined boundaries, and force a level of objectivity that was missing from traditional diagnostic tools.”

When parents, worried about their child’s development, finally decide to visit a specialist for an evaluation, depending on where they live, it may take another 12 to 15 months before the first visit takes place. “All these are unnecessary delays,” says Dr. Taraman. “The critical neurodevelopment window where one can benefit the most through interventions closes at around five or six years old. If we don’t identify the children early, we will miss that window of opportunity for them to grow up feeling less overwhelmed by their condition.”

From diagnostic to therapeutic

The lack of historical understanding of how autism manifests in girls and minority populations means some autistic individuals will not receive a formal diagnosis until they reach adulthood. Past research has shown that on average girls are diagnosed 1.5 years later than boys; while one in four children under the age of eight, the majority of whom are Black or Hispanic, are not diagnosed or misdiagnosed into other conditions.

Canvas Dx has undergone rigorous clinical trials with one clinician specialist evaluating every child to determine their diagnostic criteria for autism and a separate, independent clinician specialist confirming the diagnosis, to arrive at an 81% predictive value, ensuring children meet, miss or are in the indeterministic category of autism diagnosis.

Moving on, Cognoa is diving into personalized treatment. The company aims to identify the phenotypic subtypes of autism that will give better granularity to predict autism and generate the most appropriate therapy for those who have been diagnosed.

“The goal of Cognoa is to redefine what is autism,” Dr. Taraman says. “We are looking at whether there’s a biologically finite number of autism and we can phenotypically map them in a high-dimensional space using advanced tools. We are not just concentrating on diagnostic but examining the interesting patterns of behaviors that may be a telltale sign for other behavioral conditions like ADHD, speech and language impairment, childhood anxiety and so on.

Thus far, the only thing that has been approved by the FDA for autism treatment is an anti-psychotic drug. We are putting children, as young as three years of age on anti-psychotic medication because that’s the only thing that’s ever been approved. So, we hope to use AI, augmented reality, interactive emotion recognition paradigms as the main therapeutics.”

Cognoa is planning to make Canvas Dx widely available in the US later this year.