Jeff Sawalha Harnesses the Power of Machine Learning and Computer Science to Spot Early Signs of Bipolar Disorder and PTSD

“I’m two years into my PhD now. But it feels like it’s closer to 10 years since I was last on campus, so my perception of time is completely warped,” he says.

“For a while I was going to the Glenrose (Rehabilitation Hospital) each week to work with some people on a military study, but because of COVID it was shut down. So now I’m working from home 95% of the time.”

Sawalha’s only escape is an occasional game of shinny hockey with a few friends at one of the city’s outdoor rinks. A recent cold snap interrupted that, but now that milder weather has returned, he’s eager to get back out on the ice.

Sawalha’s path to the Graduate Program in Psychiatry had a few twists and turns as well.

“When I first got to university I had no clue what I wanted to pursue but I slowly started gravitating towards courses in Neuroscience, Psychology and Pharmacology. That’s where I garnered my love for mental health and the field of Psychiatry.”

While completing his Master’s degree in Neuroscience, Sawalha had to take courses in coding and graduate-level statistics. Much to his surprise he loved both.

“It’s just like working on a big puzzle to me. So Computer Science and Machine Learning fascinated me, and it became the partnering of those two things that I really loved.”

He is now applying those skills along with his “intermediate understanding” of how the brain works to his PhD research, which encompasses three broad areas.

“I focused on these three areas for several reasons. One was to expand my knowledge in the field of Computer Science and Machine Learning. The second was about accessibility. I wanted to hit the ground running with my PhD and get datasets that were easily accessible,” he explains.

“And finally, these three types of domains are being studied now to improve clinical diagnosis and prognostic outlooks, so there’s also a clinical aspect to it potentially.”

Sawalha’s first area of research involved using Machine Learning (ML) applications to better understand the underlying conditions of psychiatric disorders. Specifically, he uses computational modeling to identify early biological or cognitive indicators that may manifest during the early stages of a given illness.

His research was done with the collaboration and support of the Computational Psychiatry Group, which is affiliated with the Alberta Machine Intelligence Institute, the flagship group of the U of A Signature Area AI4Society. The A14 Society is one of the U of A’s five Signature Research Areas, and is focused on AI (Artificial Intelligence) and its applications.

“Identifying these markers are crucial for two reasons. First, being able to detect these alterations can allow clinicians to quickly assess and treat mental illnesses before they progress,” he says. “Secondly, we can use these indicators to help validate and classify various illnesses, improving on the accuracy of our current approaches.”

Using this approach, Sawalha examined neurocognitive data in patients with Bipolar Disorder (BD), specifically the correlation between certain neurocognitive deficits and the number of manic or psychotic episodes in first-episode and chronic-BD participants.

“The overall objective of this study was to identify individuals in the beginning stages of Bipolar Disorder by learning the more nuanced cognitive deficits that chronic-BD participants display in their neurocognitive tests,” he says.

“With the help of ML models, we were able to determine which individuals resembled a more progressive stage of BD versus a control group, and which cognitive domains were most affected by manic or psychotic episodes.”

Sawalha’s second area of research involved using functional neuroimaging for children with anxiety disorders. Specifically, he employed a dataset of 45 children, aged five to nine, who were shown images of fearful, angry and neutral faces while they were under an fMRI (functional magnetic resonance imaging) scanner.

“Our goal was to classify anxious and non-anxious children based strictly on their neuroimaging data. Using advanced computational algorithms, we were successfully able to classify children with various anxiety disorders. More interestingly, our model identified a relatively unknown area of the brain that best distinguished these individuals,” he says.

Sawalha believes the study may help researchers to ultimately relate functional brain measures to pediatric diagnoses in anxiety disorders, and also help to generate new therapeutic insights.

With his first two studies completed, Sawalha is now focused on his third and final area of research, which is the primary focus of his PhD thesis.

It involves studying the speech patterns of Canadian military veterans afflicted with PTSD (post-traumatic stress disorder). The project is being done in conjunction with Alberta Innovates, IBM’s Center for Advanced Studies, Alberta Health Services and the Department of National Defense.

“There is growing evidence that changes in the quality of the voice may be indicative of mental illness. Further, semantic or linguistic content may also be affected by one’s internal mental state,” he explains.

“For example, someone suffering from Major Depressive Disorder (MDD) may speak in a lower volume and reduced pitch, or in a more negatively valenced manner (i.e., expressing more negative emotions) compared to someone without MDD. Our goal is to develop a computational model that can distinguish veterans with PTSD from those who don’t have it.”

Once such a model is refined, Sawalha hopes to develop a screening tool that predicts PTSD onset in new military personnel, and secondly, that uses voice data to monitor and predict symptom severity for personnel in remote locations who don’t have access to primary care.

“With the recent pandemic, it’s vital that members can be reached remotely. This may be made possible by acquiring voice samples via mobile devices. In these professions, workers face repetitive occupational stress which can increase the risk of developing PTSD. By creating a screening tool, we can prevent personal suffering and large medical costs for a line of work that demands it.”

Sawalha hopes to finish his PhD this year. After that, he aims to pursue a career in industry as a data scientist.
“I’d love for it to be in the field of medicine, particularly mental health, because I think that is where the greatest need for it exists right now. If I end up in another field where I can apply what I’ve learned from computer science I’d be happy with that too. But my first priority would be mental health and Psychiatry.”

He is also hoping to stay in Alberta – and he says it’s not just because the Edmonton Oilers have Connor McDavid, the world’s best hockey player, on their roster.

“It would be nice to stay in Alberta. I’m comfortable here. And it’s nice that Edmonton is one of the hotbeds for AI (Artificial Intelligence) right now. We have lots of companies here including huge organizations like Google, Amazon and Huawei, so if there is an opportunity here for me that would be fantastic.”