Featured Publications

Nature Medicine cover featuring smartwatch monitoring of heart failure

Remote Monitoring of Heart Failure Exacerbations using a Smartwatch

We demonstrate that Apple Watch data can be used to predict peak oxygen uptake in heart failure patients using a causal attention transformer. Our results demonstrated for the first time that wearable-derived daily pVO2 strongly correlated with clinical exercise testing, and that changes in pVO2 were predictive of subsequent healthcare events. External validation confirmed these findings generalize across platforms, providing scalable monitoring for longitudinal heart failure research.

ProbMED medical multimodal learning framework

ProbMED: A Probabilistic Framework for Medical Multimodal Binding

We present ProbMED, a novel probabilistic framework that enables robust multimodal learning in healthcare by binding diverse medical data types including images, text, and physiological signals. Our approach addresses uncertainty quantification and enables more reliable clinical decision support through principled probabilistic modeling of multimodal medical embeddings.

Research on shortcut learning and AI generalization

Measuring the Generalizability of AI Models in Healthcare

Unstructured datasets, including medical imaging, electrocardiograms, and natural language data, are gaining attention with advancements in deep convolutional neural networks and large language models. However, estimating the generalizability of these models to new healthcare settings without extensive validation on external data remains challenging. We propose an open source, bias-corrected external accuracy estimate that better estimates external accuracy to within 4% on average by measuring and calibrating for bias induced shortcut learning.

Nature Medicine cover featuring AI-guided radiotherapy planning

AI-guided Prospective Cancer Radiotherapy

We describe an AI model that generates radiation treatment plans for prostate cancer patients. AI was successfully deployed in a standard-of-care clinical setting, providing gains in efficiency, improved treatment quality, and an understanding of physicians’ decision-making and perceptions of AI use when patient care is at stake.

Director

Chris McIntosh

Chris McIntosh

  • Lead AI Scientist and Senior Scientist: University Health Network's Peter Munk Cardiac Centre
  • Associate Professor: Department of Medical Biophysics, Department of Computer Science, Medical Imaging, University of Toronto
  • Faculty Affiliate, Vector Institute
  • Contact information

Students and Staff

Siham Amara-Belgadi

Siham Amara-Belgadi
PhD Student

Sangwook Kim

Sangwook Kim
PhD Student

Sejin Kim

Sejin Kim
PhD Student

Bhavish Verma

Bhavish Verma
PhD Student

Max You

Max You
PhD Student

Gerd Bizi

Gerd Bizi
MSc Student

Jisoo Lee

Jisoo Lee
PhD Student

Saemee Choi

Saemee Choi
Visiting Student

Lab Alumni

William (Yuan) Gao

William (Yuan) Gao
PhD, Medical Biophysics; Applied Scientist, Amazon

Balagopal Unnikrishnan

Balagopal Unnikrishnan
PhD, Computer Science; PiHealth

Aly Khalifa

Aly Khalifa
Physics Resident, Princess Margaret Cancer Centre

Cathy Ongly

Cathy Ongly
Risk Modelling, RBC

Overview

Working together

Innovation doesn't happen in a vacuum. Jointly embedded in Canada's largest hospital network and the University of Toronto, our research brings together experts and trainees in computer science, medicine, biology, and physics to solve impactful problems at the intersection of artificial intelligence and medicine.

Medical domains: Our work includes a variety of data modalities (biological, wearable, structured reporting, and imaging) and critical diseases such as heart disease and cancer. We work closely with clinicians to understand and ultimately augment patient care with AI.

AI technologies: Solving clinical problems means pushing and extending the boundaries of AI and machine learning technology including semi-supervised learning, domain adaptation/model generalization, meta-learning, and multi-modal data fusion.

Data: We work on datasets of all shapes and sizes, from rare diseases with rich structured data but few patients, to diverse imaging datasets in the tens of thousands.

Keep reading below for example projects.

Wearable device biomarkers

What if we understood the patient journey instead of the Coles Notes?

Wearable technologies offer a unique glimpse into patient function and biology outside of episodes of care. If AI is the present, wearbles with AI are the next frontier. Working with leading cardiologists Heather Ross and Yas Moyaedi, the Ted Rogers Centre for Heart Research, and a key partnership with Apple we are developing AI technology to asses the role of wearbles in heart failure.

Key Publications:

In the News:

  • March 2026: Toronto Star - Your smartwatch could predict heart failure a week before it lands you in hospital, Toronto study says.
  • March 2026: Global News - UHN researchers exploring how smartwatches help patients living with heart failure detect early warning signs.
  • June 2025: Canadian Healthcare Technology - Exploring how smartwatch technology and AI are revolutionizing remote heart failure monitoring, offering new possibilities for early intervention and personalized patient care.
  • Fall 2020: University of Toronto - Groundbreaking collaboration between University of Toronto researchers and Apple to develop innovative smartwatch capabilities for heart failure prevention and early detection through advanced wearable AI.

Multimodal Generative Models and Multitask Learning

Multimodal healthcare AI research

What if a model could simultaneously understand multiple sources of medical data and multiple medical tasks?

Multimodal large language models in healthcare represent a cutting-edge fusion of advanced linguistic capabilities and diverse data sources. By seamlessly integrating text, images, electrocardiograms, and more these models harbor great potential to enhance clinical care through more robust and generalizable machine learning.

Key Publications:

Automation in Radiation Therapy

The new standard of care

An experienced clinical team can more than double a cancer patient's odds of survival. In particular, radiation therapy (RT) is a key part of the treatment for 40% of all cancer patients with the potential to be highly curative and effective given the right treatment plan. However, a major barrier in that effectiveness is the expertise and experience of the patient's clinical team in his or her particular cancer. We believe that AI can help bring the expertise of the world's best clinical teams to every patient in every hospital, saving lives and costly resources.

Working with clinical partners (Tom Purdie, Ale Berlin, and Leigh Conroy) we have developed the world's first computer vision treatment system that we piloted to treat prostate cancer patients at Princess Margaret Cancer Centre, with superior results to our standard of care for the majority of patients. Cancer treatments are generated by the AI in minutes, instead of the hours-to-days taken under the typical manual process, increasing both efficiency and quality of care. Follow-up studies are expanding the models to breast cancer, brain tumours in young adults, and more. Through a successful licensing deal of this Canadian technology to RaySearch Laboratories we are also supporting deployment across Canada and around the world.

Key Publications:

In the News:

  • October 2022: UroToday - ASTRO 2022 conference coverage highlighting clinical integration of machine learning for curative-intent radiation treatment of patients with prostate cancer, showcasing real-world implementation of AI-driven treatment planning.
  • 2021: MedPage Today - Comprehensive coverage on therapeutic radiology and machine learning advancements, exploring the intersection of AI technology and clinical radiation therapy practice.
  • 2021: Hospital Healthcare - Examining physician perspectives and challenges in adopting machine learning algorithms for treatment planning in prostate cancer, addressing clinical adoption barriers and solutions.
  • June 2021: ScienceDaily - Research breakthrough showcasing fast AI-driven radiotherapy planning capabilities, demonstrating significant improvements in treatment efficiency and quality.
  • August 2018: Forbes - Early coverage of groundbreaking AI technology capable of designing comprehensive radiation therapy treatment plans for cancer patients in just twenty minutes, revolutionizing traditional planning workflows.

AI Model Generalizability in Healthcare

Illustration of model performance and generalization

How do we know if our AI will work in the real world?

The promise of AI in healthcare hinges on models that can generalize across different hospitals, patient populations, and clinical settings. However, unstructured medical data—from imaging scans to ECGs to clinical notes—presents unique challenges for model validation and deployment.

Our research addresses a critical gap: how to estimate whether an AI model trained at one institution will perform reliably at another, without requiring extensive external validation studies. We've developed novel bias-corrected methods that can predict external accuracy within 4% on average by identifying and calibrating for problematic "shortcut learning" that models often exploit.

This work enables more confident deployment of AI systems across healthcare networks and helps ensure that the benefits of machine learning reach patients regardless of where they receive care.

Key Publications:

In the News:

  • November 2024: EMJ Podcast on AI - Discussing the advancement of AI from "bench-to-bedside," while exploring concerns around bias and fairness when working with diverse patient populations and the critical importance of ensuring equitable healthcare AI deployment.
  • September 2024: UHN Research - Predicting the performance of AI tools and exploring potential biases, highlighting breakthrough research in estimating AI model reliability across different healthcare settings without extensive external validation.

Image Segmentation

Medical image segmentation research

Coming soon to a clinic near you

Image segmentation is the gold standard for computational morphology and a key step in the radiation therapy process. A long sought after technology, automated segmentation methods are recently achieving near clinical performance, paving the way for clinical studies and real-world deployment. Our research focuses on developing robust deep learning approaches for multi-organ segmentation that can handle the variability encountered across different imaging protocols and patient populations.

Building on our success in automated treatment planning, we are now conducting clinical studies to evaluate the integration of AI-driven segmentation tools into routine clinical workflows. These studies are examining both the accuracy and efficiency gains of automated contouring in radiation therapy, with early results showing potential savings for clinicians while maintaining high-quality treatment planning standards. However, many challenges remain, including handling edge cases, ensuring consistent performance across diverse patient anatomies, and seamlessly integrating these tools into existing clinical systems.

Jobs Help Wanted"Jobs Help Wanted" by Innov8social is licensed under CC BY 2.0

Do you want to change the shape of AI and healthcare?

  • Postdoctoral Fellowships

    We are seeking exceptional postdoctoral fellows to join our interdisciplinary team developing AI solutions that directly impact patient care. Ideal candidates have strong expertise in machine learning and deep learning, with demonstrated experience in healthcare applications involving medical signals, imaging, or wearable data.

    Our postdocs work on cutting-edge projects from multimodal AI models to clinical deployment of radiation therapy planning systems. We offer competitive funding, mentorship opportunities, and the chance to publish in top-tier venues while making real-world clinical impact.

    Interested candidates should submit a CV, cover letter describing research interests and career goals, your most relevant publication, and contact information for three references to Chris McIntosh with the subject line "Postdoctoral Fellowship Application".

  • Graduate Students (MSc/PhD): Healthcare AI at the University of Toronto

    Develop machine learning methods with a path to clinical use. Our lab at the University of Toronto and University Health Network works on multimodal and vision-language models, AI reliability and generalization, medical imaging, radiotherapy, and wearable monitoring. Students collaborate with computer scientists and clinicians, including teams at University Health Network's Peter Munk Cardiac Centre.

    Recent student-led work includes AlphaRAD (ECCV 2026), SilverLining (WACV 2026), ProbMED (ICCV 2025), and smartwatch monitoring of heart failure exacerbations (Nature Medicine, 2026). Our alumni have moved into roles at Amazon, PiHealth, RBC, and Princess Margaret Cancer Centre. Explore our projects and meet the team.

    We welcome applicants with strong programming skills and a foundation in machine learning, computer vision, or related fields. Prior healthcare experience is valuable but not required. In your inquiry, describe the research questions that interest you and how your experience connects to our work.

    Apply through Medical Biophysics (preferred) or Computer Science, according to your background and interests. Consult the department pages for current requirements and deadlines. To discuss research fit before applying, email Chris McIntosh with a brief research statement and your CV.

  • Summer Research Opportunities

    Experience the excitement of healthcare AI research through our summer internship program. Work alongside graduate students and postdocs on projects ranging from wearable device biomarkers to automated medical imaging analysis. Our summer students often contribute to publications and gain valuable experience for graduate school applications.

    We welcome undergraduate students with programming experience and coursework in machine learning, computer science, or related fields. Interest in healthcare applications and strong problem-solving skills are essential.

    Apply through the MBP summer student program or contact us directly to discuss project opportunities and eligibility for other summer research programs.

Affiliations