91��ɫ

Ali Sadeghi-Naini Chris Robinson

Talk to any researcher in the artificial intelligence (AI) space and their excitement for the possibilities of how it could transform many aspects of health care is palpable, and for good reason. They are developing ethical AI tools that can be integrated into clinical elements in ways that could bring us that much closer to precision medicine, where treatments would be customized to each patient and play a powerful role in improving outcomes.

AI can analyze huge, data-rich medical images and enormous quantities of data much faster than a human, but also sometimes better, observing the tiniest of details or changes that doctors cannot see, but that could necessitate a different course of patient treatment. These AI tools can also provide outstandingly accurate predictive analyses.

When it comes to cancer and liver transplant patients, it could mean the difference between a poor outcome and a higher survival rate. 91��ɫ researchers of the Lassonde School of Engineering and Divya Sharma of the Faculty of Science are developing AI tools for specific tasks that in some cases give clinicians information they otherwise would not have, with real-world implications for patients.

An associate professor, Sadeghi-Naini is developing AI tools coupled with imaging for brain, ovarian and breast cancers to characterize, monitor and predict different biological processes.

“We can scan these patients ahead of time using state-of-the-art ultrasound..."

He is the principal investigator of a new research project in collaboration with Women’s College Hospital with funding from the New Frontiers in Research Fund to develop a cost-effective, accessible AI platform to analyze the digital pathology images of ovarian cancer. The goal is to determine whether the patient has a genetic condition called homologous recombination deficiency without performing expensive genomic testing.

“It is an important factor in determining if the patient can benefit from available targeted therapies or not, but currently it requires genomic instability analysis to find out that is costly and not always accessible,” says Sadeghi-Naini, director of the Quantitative Imaging and Biomarkers Laboratory at 91��ɫ. That project is just beginning.

“I am also leading projects in collaboration with Sunnybrook Health Sciences Centre to develop AI frameworks that analyze digital pathology images of routine biopsy samples to predict treatment outcomes for individual breast cancer patients before they go through chemotherapy, to predict their response to treatment. It shows very promising results.”

Innovation 91��ɫ and Sunnybrook, where Sadeghi-Naini is a cross-appointed scientist, are currently in partnership to commercialize a couple of those tools for use.

In about 30 per cent of cases, chemotherapy does not work to shrink tumours effectively, as is the case with some high-risk breast cancers, but currently this is often determined months later after the completion of chemotherapy. “We can scan these patients ahead of time using state-of-the-art ultrasound to acquire raw signal data that after signal processing will generate quantitative ultrasound parametric images.”

The tool can then analyze those images deeper, faster and in more detail, as well as predict patient response to chemo before or shortly after it starts. It is important information that would allow oncologists to choose different treatment options if a chemo regimen is predicted not to work, which could significantly alter the survival rate of those patients who do not respond well.

“Studies show that the response of patients to upfront chemotherapy is linked to survival. Good responders show significantly better survival compared to poor responders,” says Sadeghi-Naini.

He is also working on an AI solution to a different problem, this time for brain cancer patients.

Following stereotactic radiotherapy for brain tumours, there is an up to 25 per cent chance a patient will experience radiation necrosis, a complication that can occur months to years later and is difficult for doctors to discern from brain tumour recurrence or progression.

“The problem here is that on the standard anatomical imaging, they appear very similar to each other,” he says. “That’s a challenge because radiation necrosis and tumour progression are two quite different things with different treatment approaches.”

In a recent study involving more than 90 patients with 230 brain tumours, Sadeghi-Naini and the team developed an AI platform that can analyze images of the brain using a new advanced MRI technique. Manually analyzing tumours on this multi-channel MRI is complicated, but with AI-guided methods it is much easier to distinguish between radiation necrosis, tumour progression or tumour recurrence.

“Our AI tools will not only help to predict but also improve long-term health outcomes for transplant patients by reducing disparities.”

He also leads development of an AI system to streamline analysis of repeated MRI scans for each brain cancer patient, a faster process that can better monitor and categorize tumour changes from one scan to another. In addition, he is working on an AI platform that can analyze early imaging of brain tumours and detect features invisible to the human eye, but that can provide information on the long-term outcome of the tumour, which may require a change in treatment.

“These are all cost-effective AI decision support tools for oncologists that inform personalized treatments and streamline their daily workflow, ultimately contributing to better patient care,” says Sadeghi-Naini. His research has garnered funding from the Natural Sciences and Engineering Research Council of Canada (NSERC), the Canadian Institutes of Health Research (CIHR), the National Research Council of Canada, the Terry Fox Foundation and others.

Sadeghi-Naini does not think AI can replace oncologists or radiologists, but says, “it can provide valuable complementary information, improve accessibility to precision therapeutics, save time and resources, and streamline and triage more complicated cases for expert review,” all providing added benefits to patients.

Sharma, an assistant professor and co-principal investigator on two recent CIHR project grants worth close to $3 million, is creating more equitable access to liver transplants through a national framework and developing a multimodal AI tool to improve success rates following liver transplantation.

The goal of the five-year national framework project with the University Health Network (UHN) and others is to understand the roadblocks to preventing fair access for all patients on the liver transplant waitlist, create an ethical, data-driven AI model framework to ensure equal access to donor organs going forward, and improve post-transplant outcomes.

“As part of developing an equitable AI-driven framework, we will include diverse voices in its development process. We will also analyze data on liver disease and transplants for all patients, regardless of race, socioeconomic status, sex and gender, to identify and address inequalities,” says Sharma, who leads 91��ɫ’s IMPACT-AI lab and is a scientist at UHN.

“Our AI tools will not only help to predict but also improve long-term health outcomes for transplant patients by reducing disparities.”

Some three million Canadians from all sectors of society are affected by liver disease. Although transplants can be life saving for those with end-stage liver disease, access is not equal and about 5,000 patients die from end-stage liver disease annually.

“Building trust and understanding around the new AI technology is an important piece of our project. By talking with patients, doctors and technology experts about what our AI model will do and how it can improve the process and the outcome, it can help ensure the adoption and clinical success of the framework,” says Sharma, who earned a 2025 Petro-Canada Emerging Innovator Award and a New Frontiers in Research Fund grant to develop genomic data-driven generative AI for pancreatic cancer.

“Ensuring the framework model is ethical from the beginning is key in reversing inequities to liver transplants some patients currently experience across Canada.”

"We will also analyze data on liver disease and transplants for all patients, regardless of race, socioeconomic status, sex and gender, to identify and address inequalities."

However, once a patient receives a liver transplant there is a high potential for serious complications. Up to 25 per cent of recipients will develop graft fibrosis or scarring from immunosuppressant medications or through organ rejection. Sharma’s second five-year project with UHN hopes to address this by developing a multimodal AI tool to predict patients at high risk of graft scarring.

“Our AI tools will not only help to predict but also improve long-term health outcomes for transplant patients by reducing disparities.”

Sharma says they previously used clinical and laboratory data from about 2,000 transplant recipients to develop a mathematical model to diagnose the condition. They will now expand the model’s capabilities using pathology and ultrasound imaging data so that it can also predict the future risk of scarring. The hope is it will lead to earlier diagnosis, and the development of better prevention and treatment strategies to improve outcomes.

The work of both projects are designed to have clinical benefits in hospitals and transplant centres that will result in improved and more equitable patient care. Sharma is also co-first author on a recent paper in the journal , which highlights the team’s work with GraftIQ, a neural network model designed to be a non-invasive diagnostic tool for liver graft injury.

These are some of the ways 91��ɫ researchers are capitalizing on the ability of well-designed, ethical and safe AI tools to provide real health benefits to patients, now and into the future.