Deep Learning, Cancer, and Better Healthcare Decisions
April 22, 2024
Matt Warkentin is committed to using AI as a force for good
Matt Warkentin knows all about academia, having obtained a PhD in epidemiology from the Dalla Lana School of Public Health at the University of Toronto. But his interest lies in applications, rather than abstractions. Currently a cancer epidemiologist and postdoctoral associate in the Oncology Outcomes (O2) program in the Department of Oncology at the University of Calgary, Matt wants his research to make a difference for living, breathing patients.
His current research focuses on lung cancer – specifically, integrating biomarkers into lung cancer screening to get a better handle on risk and using deep learning technologies to help triage and manage pulmonary nodules detected in the screening process. “I believe screening is the next frontier in reducing lung cancer morbidity and mortality for Canadians,” he says, adding that “artificial intelligence [AI] has the potential to detect lung cancer at an earlier stage, which has obvious downstream benefits for patients.” By the same token, AI can help mitigate human resource bottlenecks, such as the projected shortage of radiologists in Canada.
“Screening is the next frontier in reducing lung cancer morbidity and mortality, and AI has the potential to detect lung cancer at an earlier stage.”
Also exciting to Matt is how real-world data can answer important questions about cancer and thereby improve patient outcomes. “Real-world data is about using data from previous patients – which therapy was used, and when, to produce the best outcomes – to ensure current patients are getting the right treatment at the right time,” he explains.
THE BIG QUESTIONS
Why cancer? “With the population getting older, cancer is affecting more and more Canadians,” says Matt, who has been working in cancer research for 10 years. And why AI? Matt’s interest took root during his doctoral training, when he used radiomics – a rapidly evolving field of research concerned with the extraction of data from medical images – to help detect lung malignancies. “Medical imaging and radiology have the potential to benefit immensely from the introduction of AI decision-support tools,” he notes. “Embracing these tools helps us find cancers as early as possible and minimize its impact on Canadians.”
Matt takes special pride in his work to identify high-risk individuals who don’t meet standard eligibility criteria for lung cancer screening, such as non-smokers and light smokers, and in developing tools to manage pulmonary findings detected by screening programs. While working on his doctorate, he led a large multi-centre study that used images from international lung cancer screening programs to develop a radiomics-based model for assessing the risk of malignancy in screen-detected lung nodules. The work culminated in a 2024 publication in the internationally respected Thorax journal.
While applauding the fact that several provinces in Canada are piloting lung cancer screening, Matt recognizes the challenge in implementing and optimizing such programs. The success of his radiomics-based model in predicting lung cancer leads him to “hope it will be used to identify lung cancers as early as possible all, while minimizing excessive diagnostic work-ups.”
As for those who worry about AI “taking over,” Matt insists that the human aspect of medicine is not going away. He sees AI as a tool serving humans, rather than the converse. “If computational tools can help improve the shared decision-making between patients and physicians while optimizing patient outcomes and saving healthcare resources, it would be a shame not to use them.”