Research  >  About SRI  >  News & events  >  Research News

Sunnybrook researchers developing new artificial intelligence platform for precision medicine

April 22, 2021

SHARE

A $485,676 grant from the Canada Foundation for Innovation (CFI) will help to establish the Artificial Intelligence Platform for Precision Medicine (AIPPM) at Sunnybrook Research Institute.

The project, which includes a new state-of-the-art computing system designed to handle massive amounts of imaging data, will enable researchers to carry out transformative, translational work in the field of deep learning applied to medical imaging and image-guided therapy.

“At Sunnybrook we collect an enormous amount of data in the form of medical images, radiologists and pathologists reports and treatment outcomes,“ says Dr. Anne Martel, a senior scientist in physical sciences at Sunnybrook Research Institute and principal investigator of the project. “AI can help us to discover new connections and insights within this data that will allow us to predict patient outcomes, guide therapy decisions and improve the delivery of therapy.”

The research team, led by Dr. Martel, will use anonymized patient data — in the form of X-rays, CT scans, MR images and digital pathology — to develop and train artificial intelligence (AI) models that will eventually be used to solve pressing clinical problems across multiple clinical areas at Sunnybrook. The platform will be a significant step forward for clinical-grade AI systems in Canada.

“AI systems for medical data have typically been slow to progress for a few reasons — medical images are much larger than those traditionally used to train deep learning algorithms, very large datasets are needed to capture the variability in patient populations, and there is a need to link the imaging to clinical data whilst still protecting personal health information,” says Dr. Martel, who is also a professor of Medical Biophysics at the University of Toronto.

The team at Sunnybrook has already developed innovative AI approaches for a range of medical applications. The computational power provided by the new computer systems will enable the team to accelerate experiments and move past existing memory constraints to explore new deep learning architectures. This capability will be particularly useful for applications involving large data volumes, such as digital pathology applications where a single image is made up of thousands of individual tiles.

The team of collaborators working on the AIPPM project includes scientists and clinicians from a wide range of programs at Sunnybrook. Their expertise in translational research will ensure that these advances are applied to a number of clinical challenges across disorders and systems, and will be used to improve patient care. The team at Sunnybrook is also collaborating with computer scientists at the Vector Institute to translate state-of-the-art deep learning methods from the computer vision community into the medical imaging domain.

» Identifying patients at risk of neurodegenerative disease

Dr. Maged Goubran, a scientist in physical sciences at SRI, is working with Dr. Sandra Black and Dr. Jennifer Rabin to build and validate a prognostic tool to predict patient-specific rates of cognitive decline and identify at-risk individuals. The tool will incorporate a diverse set of imaging, clinical and genomic data. “The new AI platform will accelerate and augment our ability to develop clinically-focused deep learning models, as it will provide the processing capacity and compute resources required for training based on large-scale multidimensional and multimodal datasets,” says Dr. Goubran.

» Predicting risk of fracture in patients following radiation

Researchers in the Holland Bone and Joint and Odette Cancer programs are using spinal imaging from individuals with spinal metastases to train a computer model to help predict individuals who might be at risk of spinal fracture following radiotherapy. “If we can identify these patients ahead of time, using AI, we can direct them toward spine stabilizing surgeries as soon as possible, says Dr. Michael Hardisty, a junior scientist in the Holland Bone and Joint Research Program.

» Improved targeting for focused ultrasound treatments

Dr. Meaghan O’Reilly, also a scientist in the Holland Bone and Joint Research Program, and Dr. Kullervo Hynynen, vice president of Research and Innovation and focused ultrasound pioneer, are working on projects that will use AI models to adjust focused ultrasound treatments in real time. Focused ultrasound is an image-guided technology that uses ultrasound energy to target areas without the use of scalpels or incisions in to the skin. This work will specifically focus on treatment planning and monitoring for interventions in the brain and spinal cord. “The models will enable us to modify treatments, accounting for patient motion and allowing for more accurate dose delivery,” says Dr. O’Reilly.

» Better risk stratification in breast cancer patients

Dr. William Tran, a scientist in the Odette Cancer Research Program, plans to use AI to customize therapy for patients with breast cancer, by predicting their response to treatment. Dr. Tran and his team will use both medical imaging and clinical data to build and validate a model which will stratify patients by risk and guide them toward customized treatment strategies, improving overall outcomes — including reduction in tumour recurrences, survival and overall better surveillance programs. “The funding for the AIPPM will unify researchers into a common pathway, where complementary expertise can be leveraged; for example, image analysis, data science, machine learning, epidemiology, and input from medical specialties,” says Dr. Tran, who is also a radiation therapist at Sunnybrook.

» Real-time therapy guidance with AI

Similarly, Dr. Angus Lau, a scientist in physical sciences and the Odette Cancer Research Program, and his team will be using AI to modify radiotherapy plans in real time for patients treated on the MR-Linac, the first-of-its-kind hybrid machine which allows doctors to target tumours and monitor their response to treatment simultaneously. The team aims to rapidly segment MRI scans of tumours and nearby sensitive organs that need to be avoided during radiotherapy, use AI to speed up treatment pre-planning in response to patient motion, and to estimate uncertainty in radiation treatments, allowing for physician intervention. Although Dr. Lau and his team have already done some preliminary work, he says the AIPPM project will allow them to quickly progress. “The advanced graphic processing capabilities from the AIPPM project will let us train more complex models that will lead to more accurate doses delivered to patients,” says Dr. Lau.