Grant: Discovery Grants Program - Individual
McMaster University
$36,000.00
Building clinically-relevant, drift-tolerant AI models for continual usage in medical imaging
FY 2024-25
Summary
Medical artificial intelligence (AI) applications that include clinical and administrative applications are on the rise given the seminal improvements in AI, including machine learning (ML) and deep learning (DL). However, the actual adoption of AI for clinical applications is very low. Only 13% of all AI developments move into production. This discrepancy between AI's promise and practice (called the AI Chasm) has become a topic of interest among AI researchers and practitioners. This is leading to concerns regarding adoption of AI in healthcare. In this proposal, we focus on two major issues that pose barriers to the practical application of AI, i.e., the chasm.
First, it is important that the clinical context (why, where and how the AI is used in the clinical workflow) is incorporated into the technical design of an AI-model to facilitate its application and adoption in an existing clinical workflow/process. Second, it is important to mitigate the effect of data drift (deviations from the narrow data-related assumptions) in medical AI applications. This is required to determine measures for changes (e.g., changes in appearance of a tumor in X-ray when captured with a GE machine instead of a SIEMENS machine) in input data and incorporating drift-tolerant mechanisms for deploying AI systems for continual usage.
My proposed 5-year research program addresses these two fundamental issues (deficiency of clinical context and presence of data drift) from the point-of-view of natural sciences and engineering (NSE) by developing novel technology in medical imaging-based AI systems for breast cancer (BC). The choice of a disease is important to define the clinical context. My specific objectives are:
(1) To design novel clinically contextual technology to compare mammograms (2D breast X-ray) over time - the task of comparing multiple mammograms (from the same patient over time), though routinely performed by radiologists, remains underexplored as an engineering problem, despite the potential broad utility.
(2) To design approaches for identifying and measuring data drift in the imaging modalities used in BC - this is important to mitigate the effect of the drift.
(3) To mitigate the data drift in the imaging modalities used in BC - this is important for the continual usage of the deployed AI models, including the models we develop for the first objective.
Our work will contribute by developing a dataset along with novel architectures and models in deep learning to facilitate the use of assistive and accurate AI-models in the interpretation of longitudinal mammograms. This work will adapt NSE principles needed to bridge the AI Chasm for BC imaging. This will catalyze the adoption of AI-models in a high-stake environment and support their continual usage, ultimately improving patientcare in Canada. The program will train HQPs to solve engineering problems for healthcare and equip them with unique skillsets that will socioeconomically benefit Canada.
Program
Discovery Grants Program - Individual
Recipient
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
RGPIN-2024-06758
Name nom
Saha, Ashirbani
Department département
Oncology
Organization id
27
Institution établissement
McMaster University
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
36000
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
1206
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Engineering
Area of application fr
Génie
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
Building clinically-relevant, drift-tolerant AI models for continual usage in medical imaging
Keywords
biomedical image analysis and processing; mammogram; longitudinal medical image; distribution drift; artificial intelligence; machine learning
Application summary
Medical artificial intelligence (AI) applications that include clinical and administrative applications are on the rise given the seminal improvements in AI, including machine learning (ML) and deep learning (DL). However, the actual adoption of AI for clinical applications is very low. Only 13% of all AI developments move into production. This discrepancy between AI's promise and practice (called the AI Chasm) has become a topic of interest among AI researchers and practitioners. This is leading to concerns regarding adoption of AI in healthcare. In this proposal, we focus on two major issues that pose barriers to the practical application of AI, i.e., the chasm.
First, it is important that the clinical context (why, where and how the AI is used in the clinical workflow) is incorporated into the technical design of an AI-model to facilitate its application and adoption in an existing clinical workflow/process. Second, it is important to mitigate the effect of data drift (deviations from the narrow data-related assumptions) in medical AI applications. This is required to determine measures for changes (e.g., changes in appearance of a tumor in X-ray when captured with a GE machine instead of a SIEMENS machine) in input data and incorporating drift-tolerant mechanisms for deploying AI systems for continual usage.
My proposed 5-year research program addresses these two fundamental issues (deficiency of clinical context and presence of data drift) from the point-of-view of natural sciences and engineering (NSE) by developing novel technology in medical imaging-based AI systems for breast cancer (BC). The choice of a disease is important to define the clinical context. My specific objectives are:
(1) To design novel clinically contextual technology to compare mammograms (2D breast X-ray) over time - the task of comparing multiple mammograms (from the same patient over time), though routinely performed by radiologists, remains underexplored as an engineering problem, despite the potential broad utility.
(2) To design approaches for identifying and measuring data drift in the imaging modalities used in BC - this is important to mitigate the effect of the drift.
(3) To mitigate the data drift in the imaging modalities used in BC - this is important for the continual usage of the deployed AI models, including the models we develop for the first objective.
Our work will contribute by developing a dataset along with novel architectures and models in deep learning to facilitate the use of assistive and accurate AI-models in the interpretation of longitudinal mammograms. This work will adapt NSE principles needed to bridge the AI Chasm for BC imaging. This will catalyze the adoption of AI-models in a high-stake environment and support their continual usage, ultimately improving patientcare in Canada. The program will train HQPs to solve engineering problems for healthcare and equip them with unique skillsets that will socioeconomically benefit Canada.