Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

Carleton University

$33,000.00

Concept-based explanations for machine learning in medical imaging

EF 2024-25

Résumé
Hardly a day goes by without hearing about a new `smart' object to make our lives easier: appliances that text us when our clothes are clean and dry; thermostats that adjust themselves. Artificial intelligence ("AI") has helped to make certain tasks more convenient. And while mistakes can occur, it is rarely a cause for serious concern. After all: humans make mistakes, too. AI has also made its way into medicine. It is now possible to train computers with hundreds or thousands of images (like X-rays) and `teach' it to provide the correct diagnosis. Some of these AI programs are extremely accurate; sometimes even more accurate than well-trained specialist doctors. But what can we do if the computer gets a diagnosis wrong? Mistakes in health care can be fatal. When they are we need to be able to find out what went wrong so that we can make sure it never happens again. This proposal sets out to open the black box so that we may learn more about how these programs work. To do this, the proposed program brings together insights and expertise from computer science and elsewhere to cope with the data we collect in medical imaging. Now when a highly trained doctor looks at medical images, they look for patterns and give their diagnosis. But if they turn out to be wrong, they can return to the image and re-visit each step in their decision making. They might say: "that tumour looked round and smooth, so that convinced me it was not a danger to the patient." The doctor's (human) colleague might point out an area that the first doctor missed. A rough patch they didn't see before. The two doctors will discuss the relative concerns posed by this `new' information. Would it change the doctor's diagnosis? There would be debate and, with luck, they would reach a consensus and make a plan. While AI programs are exciting, we still have a lot to learn about how they reach decisions. If someone wanted to explain a mistake in diagnosis, the "explanation" would amount to a blackboard full of millions of equations. As long as we lack the language to explain the "reasoning" underlying artificial intelligence decisions, we cannot trust them with our lives. We are partnering with specialist doctors who will help identify differences between how human and AI experts reach decisions. The proposed work will help unite the best of human and computer learning so that we can improve diagnostic performance. This program will help to pave the way for AI diagnoses that are accountable. This work will provide a number of exciting training opportunities to explore the inner workings of AI programs. While this will directly benefit medical imaging, understanding AI programs is also urgently needed for other sensitive applications, such as security (e.g., face detection), or for object detection in drone surveillance imaging.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
RGPIN-2021-04224
Name nom
Thornhill, Rebecca
Department département
Systems and Computer Engineering
Organization id
23
Institution établissement
Carleton University
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2021
Award amount
33000
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
1211
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Medical and health sciences
Area of application fr
Sciences médicales et de la santé
Research subject code
1901
Research subject group en
Biomedical engineering
Research subject group fr
Génie biomédical
Research subject en
Biomedical technology
Research subject fr
Technologie biomédicale
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
Concept-based explanations for machine learning in medical imaging
Keywords
machine learning; deep learning; medical imaging; radiomics; imaging; pattern recognition; bioinformatics; computer-aided diagnosis; artificial intelligence; explainable ai
Application summary
Hardly a day goes by without hearing about a new `smart' object to make our lives easier: appliances that text us when our clothes are clean and dry; thermostats that adjust themselves. Artificial intelligence ("AI") has helped to make certain tasks more convenient. And while mistakes can occur, it is rarely a cause for serious concern. After all: humans make mistakes, too. AI has also made its way into medicine. It is now possible to train computers with hundreds or thousands of images (like X-rays) and `teach' it to provide the correct diagnosis. Some of these AI programs are extremely accurate; sometimes even more accurate than well-trained specialist doctors. But what can we do if the computer gets a diagnosis wrong? Mistakes in health care can be fatal. When they are we need to be able to find out what went wrong so that we can make sure it never happens again. This proposal sets out to open the black box so that we may learn more about how these programs work. To do this, the proposed program brings together insights and expertise from computer science and elsewhere to cope with the data we collect in medical imaging. Now when a highly trained doctor looks at medical images, they look for patterns and give their diagnosis. But if they turn out to be wrong, they can return to the image and re-visit each step in their decision making. They might say: "that tumour looked round and smooth, so that convinced me it was not a danger to the patient." The doctor's (human) colleague might point out an area that the first doctor missed. A rough patch they didn't see before. The two doctors will discuss the relative concerns posed by this `new' information. Would it change the doctor's diagnosis? There would be debate and, with luck, they would reach a consensus and make a plan. While AI programs are exciting, we still have a lot to learn about how they reach decisions. If someone wanted to explain a mistake in diagnosis, the "explanation" would amount to a blackboard full of millions of equations. As long as we lack the language to explain the "reasoning" underlying artificial intelligence decisions, we cannot trust them with our lives. We are partnering with specialist doctors who will help identify differences between how human and AI experts reach decisions. The proposed work will help unite the best of human and computer learning so that we can improve diagnostic performance. This program will help to pave the way for AI diagnoses that are accountable. This work will provide a number of exciting training opportunities to explore the inner workings of AI programs. While this will directly benefit medical imaging, understanding AI programs is also urgently needed for other sensitive applications, such as security (e.g., face detection), or for object detection in drone surveillance imaging.