Grant: Discovery Grants Program - Individual
McGill University
$69,000.00
Causal Representation Learning for Trustworthy Temporal Medical Image Analysis
FY 2024-25
Summary
Building deep learning (DL) models for effectively representing temporal trajectories over sequences of patient images, typically acquired over intervals of months or years, could lead to breakthroughs in understanding and predicting disease progression, and in identifying future responders to treatments, yet remains a completely open area of research. Despite advances made in computer vision, temporal models for natural images (e.g video analysis, face aging) are not readily adaptable to real-world clinical contexts due to the unique challenges they present, such as the high dimensionality of the images, and the relatively subtle differences in images irregularly sampled over time. In order to deep learning model to be trustworthy enough to be adopted in the clinic, the system needs to communicate the basis for the predictions, the degree of confidence in the results, and to feel convinced that the models lead to generalizable and equitable predictions.
The long term objectives of the proposed research program are to foster a unique, multidisciplinary research environment where innovative, state-of-the-art advances at the intersection of machine learning, computer vision and medical image analysis are developed for a wide range of challenges clinical contexts, where they have the potential to lead to concrete improvements in medical diagnosis and patient care. The specific objectives are: (1) To develop the foundations of causal representation models that will learn, for the first time, to model the temporal trajectories of long-term complex diseases in latent space, from high dimensional medical images acquired from patients on and off treatments. This includes predictions of future disease evolutions and treatment effects. (2) To develop new explainable, causal deep learning models, resulting in the discovery of patient-specific imaging markers that are predictive of future disease activity, disease evolutions or treatment response. This includes uncertainty estimates for counterfactual future patient trajectories and treatment response. (3) To develop theoretical and practical DL models that embed fairness and equity in the context of medical image analysis, with the objective of identifying and overcoming hidden biases affecting the accuracy of predictions and impeding their generalization across different datasets.
In addition to the described theoretical advances, the new machine learning models will have the potential to lead to significant improvements in patient care in real, clinical domains. My research team has unique access to the largest, unified proprietary, clinical trial dataset of MRI images (and clinical information) acquired from patients with multiple sclerosis, on and off various treatments, in which to test the methods. New advances in machine learning could permit the identification of potential responders to different treatments, leading to the first image-based deep learning models for personalized medicine.
Program
Discovery Grants Program - Individual
Recipient
Location
Quebec, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
RGPIN-2023-05152
Name nom
Arbel, Tal
Department département
Electrical and Computer Engineering - Electrical and Computer Engineering
Organization id
61
Institution établissement
McGill University
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2023
Award amount
69000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
Area of application code
1102
Area of application group en
Health, education and social services
Area of application group fr
Santé publique, éducation et services sociaux
Area of application en
Biomedical engineering
Area of application fr
Génie biomédical
Research subject code
2800
Research subject group en
Artificial intelligence
Research subject group fr
Intelligence artificielle
Research subject en
Artificial intelligence (computer vision, use2603)
Research subject fr
Intelligence artificielle (vision artificielle, utiliser 2603)
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Causal Representation Learning for Trustworthy Temporal Medical Image Analysis
Keywords
computer vision; medical image analysis; uncertainty estimation; bayesian deep learning; deep representation learning; explainable deep learning models; domain adaptation; causal inference; temporal models; image-based precision medicine
Application summary
Building deep learning (DL) models for effectively representing temporal trajectories over sequences of patient images, typically acquired over intervals of months or years, could lead to breakthroughs in understanding and predicting disease progression, and in identifying future responders to treatments, yet remains a completely open area of research. Despite advances made in computer vision, temporal models for natural images (e.g video analysis, face aging) are not readily adaptable to real-world clinical contexts due to the unique challenges they present, such as the high dimensionality of the images, and the relatively subtle differences in images irregularly sampled over time. In order to deep learning model to be trustworthy enough to be adopted in the clinic, the system needs to communicate the basis for the predictions, the degree of confidence in the results, and to feel convinced that the models lead to generalizable and equitable predictions.
The long term objectives of the proposed research program are to foster a unique, multidisciplinary research environment where innovative, state-of-the-art advances at the intersection of machine learning, computer vision and medical image analysis are developed for a wide range of challenges clinical contexts, where they have the potential to lead to concrete improvements in medical diagnosis and patient care. The specific objectives are: (1) To develop the foundations of causal representation models that will learn, for the first time, to model the temporal trajectories of long-term complex diseases in latent space, from high dimensional medical images acquired from patients on and off treatments. This includes predictions of future disease evolutions and treatment effects. (2) To develop new explainable, causal deep learning models, resulting in the discovery of patient-specific imaging markers that are predictive of future disease activity, disease evolutions or treatment response. This includes uncertainty estimates for counterfactual future patient trajectories and treatment response. (3) To develop theoretical and practical DL models that embed fairness and equity in the context of medical image analysis, with the objective of identifying and overcoming hidden biases affecting the accuracy of predictions and impeding their generalization across different datasets.
In addition to the described theoretical advances, the new machine learning models will have the potential to lead to significant improvements in patient care in real, clinical domains. My research team has unique access to the largest, unified proprietary, clinical trial dataset of MRI images (and clinical information) acquired from patients with multiple sclerosis, on and off various treatments, in which to test the methods. New advances in machine learning could permit the identification of potential responders to different treatments, leading to the first image-based deep learning models for personalized medicine.