Grant: Discovery Grants Program - Individual
University of Calgary
$53,000.00
Causal Deep Learning for Medical Imaging Applications
FY 2024-25
Summary
Medical imaging is one of the largest data contributors in health care and holds significant importance in disease prevention, diagnosis, and treatment planning. Within this context, deep convolutional neural networks have the potential to revolutionize radiology by automating the analysis of highly complex but structured medical image datasets. However, the adoption of machine learning models in clinical radiology practice faces challenges because many existing models focus solely on the predictive or generative aspects by making use of correlations in the data without considering the underlying causal relationships. Thus, machine learning models are often not able to properly explain their decisions and have limited use to generate new knowledge about disease mechanisms in terms of causal relations.
While a few methods have been proposed that combine deep learning methods with causal modelling approaches, these examples are mostly restricted to 2D toy examples from the computer vision domain while applications in the radiology domain to date have been mostly limited to tabulated data extracted from the imaging data due to the complexity of 3D medical imaging data and associated problems of using it directly within a causal analysis framework.
Over the next 5 years, we will develop and evaluate novel causal deep learning algorithms that provide not only accurate inferences but can also be used to obtain insights into the causal relations between clinical data and high-dimensional medical images. Their development will be achieved via three specific aims. First, we will develop a flexible causal deep learning framework by combining cutting-edge deep learning architectures with causal discovery methods, to enable the discovery of causal relationships between non-imaging parameters and high-dimensional medical imaging datasets. Second, we will implement an advanced method to generate synthetic medical images with known causal and anti-causal relationships, and spurious correlations of non-imaging variables, which will be used as ground truth for testing and evaluating the effectiveness of the causal deep learning framework. Third, we will develop methods to generate counterfactual images hat can help removing non-informative biases from the imaging data. Additionally, the generation of counterfactual images will enable the identification of the structural or functional changes evident in medical images that are causally related to non-imaging parameters.
This ambitious research program aims to combine the power of deep learning and elements of causal inference to revolutionize medical imaging analysis and contribute to a better understanding of complex medical conditions for the end goal of improved patient care and treatment. Moreover, the methods developed will be flexible enough to be useful for other problems in the computer vision domain.
Program
Discovery Grants Program - Individual
Recipient
Location
Alberta, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
RGPIN-2024-05772
Name nom
Forkert, Nils Daniel
Department département
Radiology
Organization id
11
Institution établissement
University of Calgary
Province en
Alberta
Province fr
Alberta
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
53000
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
1510
Committee name en
Electrical and Computer Engineering
Committee name fr
Génie électrique et informatique
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
2804
Research subject group en
Artificial intelligence
Research subject group fr
Intelligence artificielle
Research subject en
Expert systems
Research subject fr
Systèmes experts
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 Deep Learning for Medical Imaging Applications
Keywords
medical imaging; medical image processing; deep learning; causal analysis
Application summary
Medical imaging is one of the largest data contributors in health care and holds significant importance in disease prevention, diagnosis, and treatment planning. Within this context, deep convolutional neural networks have the potential to revolutionize radiology by automating the analysis of highly complex but structured medical image datasets. However, the adoption of machine learning models in clinical radiology practice faces challenges because many existing models focus solely on the predictive or generative aspects by making use of correlations in the data without considering the underlying causal relationships. Thus, machine learning models are often not able to properly explain their decisions and have limited use to generate new knowledge about disease mechanisms in terms of causal relations.
While a few methods have been proposed that combine deep learning methods with causal modelling approaches, these examples are mostly restricted to 2D toy examples from the computer vision domain while applications in the radiology domain to date have been mostly limited to tabulated data extracted from the imaging data due to the complexity of 3D medical imaging data and associated problems of using it directly within a causal analysis framework.
Over the next 5 years, we will develop and evaluate novel causal deep learning algorithms that provide not only accurate inferences but can also be used to obtain insights into the causal relations between clinical data and high-dimensional medical images. Their development will be achieved via three specific aims. First, we will develop a flexible causal deep learning framework by combining cutting-edge deep learning architectures with causal discovery methods, to enable the discovery of causal relationships between non-imaging parameters and high-dimensional medical imaging datasets. Second, we will implement an advanced method to generate synthetic medical images with known causal and anti-causal relationships, and spurious correlations of non-imaging variables, which will be used as ground truth for testing and evaluating the effectiveness of the causal deep learning framework. Third, we will develop methods to generate counterfactual images hat can help removing non-informative biases from the imaging data. Additionally, the generation of counterfactual images will enable the identification of the structural or functional changes evident in medical images that are causally related to non-imaging parameters.
This ambitious research program aims to combine the power of deep learning and elements of causal inference to revolutionize medical imaging analysis and contribute to a better understanding of complex medical conditions for the end goal of improved patient care and treatment. Moreover, the methods developed will be flexible enough to be useful for other problems in the computer vision domain.