Blake VanBerlo, a PhD candidate at the Cheriton School of Computer Science, is one of five recipients at Waterloo of a prestigious Vanier Canada Graduate Scholarship. Valued at $150,000 over three years, Vanier Scholarships recognize doctoral students who have exhibited academic excellence, research potential and leadership.
Blake received the scholarship in recognition of his research in which he is training self-supervised machine learning models to analyze unlabelled lung ultrasound data, with the hope of improving the accuracy and speed at which medical practitioners can interpret lung ultrasound results.
Lung ultrasound imaging is an essential diagnostic tool because it’s fast, inexpensive, and safe compared with radiation-based imaging devices such as CT scanners and X-ray machines, Blake says. “You can wheel these units right to a patient’s bedside and use them as a screening tool to narrow down your differential diagnosis quickly.”
Such diagnostic tools are more important than ever, as many patients have experienced lingering respiratory effects after a COVID-19 infection. Unfortunately, Blake explains, “there aren’t a lot of trained professionals out there who know how to interpret that ultrasound data.”
Blake’s unique academic trajectory ideally suits him to tackle this problem. As an undergrad, he entered Western University as a Schulich Leader excited about science and planning to go to medical school. He ended up majoring in software engineering, then completed two years of medical school before realizing he was more interested in the problem-solving and research side of medicine than the clinical side.
Before he left medical school, however, he befriended Dr. Robert Arntfield, a professor of medicine who is interested in the possible impacts of machine learning on ultrasound interpretation. Blake is a founding member of Deep Breathe, Dr. Arntfield’s project aiming to use AI to improve lung ultrasound diagnostics. In 2020, this line of inquiry brought Blake to the University of Waterloo for a master’s of computer science funded by a Vector Scholarship in Artificial Intelligence. He is continuing this work in his PhD.
Blake’s PhD research will help automate the diagnostic process, increasing the accessibility of lung ultrasounds as a diagnostic tool by training a machine learning model to recognize important patterns and features in lung ultrasounds. His work also addresses a secondary problem with the technology: much of the ultrasound data that is available to researchers is unlabelled. “There’s a lot of data,” he says, “but often no clinicians went back to label it.”
Blake addresses this problem by employing a self-supervised learning model. He provides his model with the unlabelled data, and it looks for existing patterns or categories without being told what to look for. Then, he provides his self-supervised model with a more specific problem — for example, looking for anomalies that indicate lung tissue damage — with the hope that the previous self-supervised learning will improve its efficacy.
“We want to produce better automated lung ultrasound systems that are accurate and trustworthy,” Blake says. “If we have more automation in lung ultrasound interpretation, then these tools can be used more frequently in diagnostics in critical care settings.”
“Winning the Vanier lets me devote my attention to a space where I think there are not enough people working, and it gives me the time to give it my all,” he says. “That would not be possible without this kind of support.” In turn, he says, he would never have been able to achieve what he has without the unwavering encouragement of his wife Olivia, “the most supportive human being that I’ve ever met.”
He is also grateful to his PhD supervisors, Professor Jesse Hoey at the Cheriton School of Computer Science and Professor Alexander Wong in the Department of Systems Design Engineering, along with Dr. Arntfield and the Deep Breathe team.
“It’s very meaningful,” he says, “to have the ability to pursue a research question born out of a very real need.”