Applying Machine Learning to Preoperative Treatment Plans in Prostate Brachytherapy
By Connie Jeske Crane
Alexandru Mihai Nicolae
When it comes to machine learning and medical therapies, successful implementation comes with an additional and rather interesting challenge – namely, winning over a medical community that’s understandably apprehensive about adopting computerized methods in treatment planning. (“How might this affect patient outcomes?” “What about reliability and continuity of care?”)
A team of Ryerson researchers working in conjunction with Sunnybrook Hospital’s Odette Cancer Centre recently had notable success applying machine learning, specifically with prostate cancer treatment. One of the most commonly diagnosed cancers in men, according to Prostate Cancer Canada, 21,000 men in this country are diagnosed with prostate cancer every year. Without adequate treatment, more than 4,000 will die from complications related to the disease.
So how could machine learning make a difference? Researcher Alexandru Mihai Nicolae (MSc Biomedical Physics ’16) and his co-supervisors, Ananth Ravi, an affiliate scientist with the Sunnybrook Research Institute, and Carl Kumaradas, a Ryerson physics professor, pinpointed a key opportunity – specifically the pre-planning and delivery of brachytherapy treatment.
Alexandru explains: “One of the most effective forms of treatment for prostate cancer is prostate brachytherapy, a form of internal radiation therapy where small radioactive seeds are implanted directly within the cancer. Both the American Brachytherapy Society (ABS) and the American Society of Clinical Oncology (ASCO) recommend that patients with localized prostate cancer be offered this type of definitive curative treatment.” A huge challenge though, as Nicolae explains, involves resourcing. “This highly critical and difficult task can take experts several hours to adequately perform…and significantly impacts both the clinical workflow and the quality of the final patient outcome.” The ever-increasing need for skilled clinicians to deliver brachytherapy puts a huge strain on available resources and subsequently, patient access to care.
Working out of Sunnybrook Hospital’s Odette Cancer Centre, Nicolae, Ravi and Kumaradas, focused on the ultra-time-consuming manual planning process for optimizing the pattern of radioactive seeds within the prostate during brachytherapy. Using machine learning-based methods, they tackled the development of an automated approach to treatment planning for brachytherapy cases to potentially alleviate previous shortcomings. “The system uses a database of prior, expertly treated prostate brachytherapy plans that led to excellent patient outcomes and learns how to perform these on new patient cases,” explains Alexandru.
One of the main benefits with the newly developed machine learning-based approach is speed. “Plans of comparable quality to those of experts can be planned in less-than a few minutes, with consistent quality,” says Alexandru. Add to that, improved access: “Many Canadians would not otherwise be able to receive prostate brachytherapy,” he says. Among other benefits, Alexandru mentions the ability to deliver treatment planning expertise to patients at remote centres with consistent quality, as well as knowledge transfer between centres, possible cost reductions and freeing up clinicians for other critical tasks.
Visual comparison of qualitative treatment plans features. Contours include the prostate (red), urethra (yellow), and
rectum (dark blue). Abbreviations: BT Z brachytherapist; Inf. Z inferior; ML Z machine learning; Ref Z reference plan;
Sup. Z superior
Without a doubt, the team’s research was promising. Yet extensive follow-up testing and communication was also crucial in building up credibility and acceptance. Alexandru says, “Many clinicians and staff were apprehensive about even testing out the system at first. Through repeat usage much of this apprehension has been replaced with more enthusiasm for the additional time computerized planning frees up in the clinic.”
Of his time at Ryerson, Alexandru recalls first and foremost the connections, brilliant minds and high quality research within the Physics department. “Many of these connections have become close friends or collaborators over time and continue to contribute to brain storming and new idea creation.”
The team’s machine learning-based research studies were funded by grants from Telus Ride-for-Dad and the Ontario Consortium for Adaptive Interventions in Radiation Oncology (OCAIRO).