Artificial Intelligence to Revolutionize Child Behavioral Diagnostics
Business + ManagementHealthcareChild Behavioral DiagnosticsEast 2019HealthcareXAIposted by Elizabeth Wallace, ODSC May 4, 2020 Elizabeth Wallace, ODSC
jThe CDC estimates that a variety of disorders affect children in the US. ADHD could be as high as 10% of the childhood population, while speech disorders could affect between 5 and 12% of young children. And these aren’t the only conditions researchers are exploring. The use of AI to predict and issue guidance for a growing body of patients is in its infancy, but a few startups are rising to the challenge. Halim Abbas of Cognoa is here to discuss how his startup is tackling the complex world of child behavioral diagnostics in his AIX talk “Artificial Intelligence to Revolutionize Child Behavioral Diagnostics.” Let’s see what they’re doing.
[Related Article: AI in Healthcare: Use Cases for 2020]
The Challenge of Child Behavioral Diagnostics
The diagnostic methods for behavioral conditions are far less straightforward than physical issues. There are no definitive medical tests and no established genetic markers. Instead, doctors rely on a series of phenotypes as clues.
These behavioral disorders also differ widely, which also muddies the diagnostic process. The assessment methods fall short and hail from an age before the age of information. They’re time intensive and extensive. They rely on training and are often too simplistic.
In many cases, the diagnosis uses questionnaires. They fall short in predictive value because they’re too linear and require too much on false positives to catch enough of the true diagnoses.
Early intervention can have dramatic results on the outcome of the child, but the diagnostic process often impedes this quick intervention. Abbas believes there is a better, faster way to begin this diagnostic process.
The Obstacles to AI Adoption
AI diagnostics are more timely and may even be more accurately predictive than old school methods. It’s generally less time-intensive and less labor heavy. However, there are quite a few obstacles to this type of adoption.
Health data is notoriously private. Not only are there governmental regulations in place, but patients and practitioners themselves are resistant to the risks of having data available in this method.
There is also a big difference between traditional results in other industries and the rate of error in healthcare. This is often the difference between life and death, or at the very least, quality of life and not. This raises the bar for effective diagnosis and performance.
Also, the definitions of conditions continue to evolve and grow. This presents noise to the machine and will require new methods for solving these sorts of issues.
How Does Cognoa Approach These Obstacles?
Cognoa sends the first modality directly to the parent. They can fill out a small questionnaire to gain insights into the child’s behavior. They can also use film with directed activities. The company uses deep learning analysis for the same clues that experts use in diagnosis.
The third modality is direct interaction. They film the clues by interacting with the child. And finally, they send a questionnaire to the physician for a final assessment of the child. These assessments go into a multi-module fusion model for rendering diagnostics.
The company wants to build something both accessible and fast without sacrificing reliability. These four different areas of assessments complement each other and provide insights into the long term behavior of the child.
Cognoa is aiming for as high as 95% accuracy using these application features and is currently waiting on the data to reveal if they’ve reached their target. This application could prove invaluable for families who traditionally had to wait years for a correct diagnosis and could improve outcomes across the board.
The Future of AI Predictive Models in Behavioral Diagnostics
Abbas believes that a role reversal is on the horizon. When we think of applying healthcare now, we take the clinician’s labels and historical records to see if we can mimic the physicians’ choices. That’s going to change.
Instead, we can take the complex decision trees of AI and apply them in reverse to the process of diagnostics. We can consider millions of patients at the same time and make inferences. It’s easier to contribute to the advancement of clinical science.
In many decades to come, AI’s further influence is clear. While some may believe that AI will replace doctors, that isn’t quite what’s going to happen. Instead, the transformation will provide new systems and new ways of thinking instead of just simplistic views of replacing doctors.
We may find ourselves with a suite of services that happen more efficiently. A comprehensive set of everyday items like mobile or wearables will give us the chance to add new elements of our diagnostics. We’ll be able to stitch together a complete picture of cognitive health instead of prescriptive, on-demand services.
[Related Article: Machine Learning and Compression Systems in Communications and Healthcare]
This allows the future to be individualized more fully for each child. Each child could reach their potential seamlessly using these integrated AI systems in a way that could never happen before.