Explainable Deep Learning for Myocardial Perfusion Imaging with a Grad-CAM-Based Loss Function
(2026) Medical Imaging 2026: Computer-Aided Diagnosis In Progress in Biomedical Optics and Imaging - Proceedings of SPIE 13926.- Abstract
As deep learning methods become more frequent in medical imaging, model explainability becomes essential in order to ensure trust. Gradient-weighted class activation mapping (Grad-CAM) is a widely used technique for visualizing which regions of an input contribute to a network’s decision. In this work, we introduce a novel Grad-CAM–based loss function that leverages prior knowledge of target regions to constrain and improve the relevance assigned by Grad-CAM. The approach is applied to myocardial perfusion imaging (MPI), a standard examination method for evaluating suspected coronary artery disease (CAD). When CAD is suspected, invasive coronary angiography is usually performed to assess the degree of stenosis using quantitative... (More)
As deep learning methods become more frequent in medical imaging, model explainability becomes essential in order to ensure trust. Gradient-weighted class activation mapping (Grad-CAM) is a widely used technique for visualizing which regions of an input contribute to a network’s decision. In this work, we introduce a novel Grad-CAM–based loss function that leverages prior knowledge of target regions to constrain and improve the relevance assigned by Grad-CAM. The approach is applied to myocardial perfusion imaging (MPI), a standard examination method for evaluating suspected coronary artery disease (CAD). When CAD is suspected, invasive coronary angiography is usually performed to assess the degree of stenosis using quantitative coronary angiography (QCA). A convolutional neural network (CNN) was trained to predict QCA values for the three main coronary arteries of the heart (LAD, RCA, LCx) from MPI. When training the CNN with the proposed loss function, area under the receiver operating characteristics curve (AUC) values of 79.2%, 93.1%, 86.2% were achieved for LAD, RCA, and LCx respectively. The proposed loss function remains competitive to standardized functions, while also producing heatmaps which less frequently assign relevance to incorrect regions.
(Less)
- author
- Karlsson, Jennie
LU
; Heyden, Anders
LU
; Åström, Karl
LU
; Overgaard, Niels Christian
LU
; Dahl, Julia
; Davidsson, Anette
; Figueroa, Miguel Ochoa
and Arvidsson, Ida
LU
- organization
-
- LTH Profile Area: AI and Digitalization
- Computer Vision and Machine Learning (research group)
- LU Profile Area: Natural and Artificial Cognition
- eSSENCE: The e-Science Collaboration
- ELLIIT: the Linköping-Lund initiative on IT and mobile communication
- LU Profile Area: Proactive Ageing
- LTH Profile Area: Engineering Health
- Lund Laser Centre, LLC
- LTH Profile Area: Photon Science and Technology
- LU Profile Area: Nature-based future solutions
- LU Profile Area: Light and Materials
- Stroke Imaging Research group (research group)
- publishing date
- 2026-04-02
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- keywords
- Convolutional neural networks, Explainability, Grad-CAM, Myocardial perfusion imaging
- host publication
- Medical Imaging 2026 : Computer-Aided Diagnosis - Computer-Aided Diagnosis
- series title
- Progress in Biomedical Optics and Imaging - Proceedings of SPIE
- editor
- Wismuller, Axel and Deserno, Thomas Martin
- volume
- 13926
- article number
- 1392629
- publisher
- SPIE
- conference name
- Medical Imaging 2026: Computer-Aided Diagnosis
- conference location
- Vancouver, Canada
- conference dates
- 2026-02-15 - 2026-02-19
- external identifiers
-
- scopus:105041009737
- ISSN
- 1605-7422
- 2410-9045
- ISBN
- 9781510697898
- DOI
- 10.1117/12.3086120
- language
- English
- LU publication?
- yes
- id
- 4d7b1036-f52a-4cf9-9a78-efbb0ff6613d
- date added to LUP
- 2026-08-24 14:19:53
- date last changed
- 2026-08-25 03:05:46
@inproceedings{4d7b1036-f52a-4cf9-9a78-efbb0ff6613d,
abstract = {{<p>As deep learning methods become more frequent in medical imaging, model explainability becomes essential in order to ensure trust. Gradient-weighted class activation mapping (Grad-CAM) is a widely used technique for visualizing which regions of an input contribute to a network’s decision. In this work, we introduce a novel Grad-CAM–based loss function that leverages prior knowledge of target regions to constrain and improve the relevance assigned by Grad-CAM. The approach is applied to myocardial perfusion imaging (MPI), a standard examination method for evaluating suspected coronary artery disease (CAD). When CAD is suspected, invasive coronary angiography is usually performed to assess the degree of stenosis using quantitative coronary angiography (QCA). A convolutional neural network (CNN) was trained to predict QCA values for the three main coronary arteries of the heart (LAD, RCA, LCx) from MPI. When training the CNN with the proposed loss function, area under the receiver operating characteristics curve (AUC) values of 79.2%, 93.1%, 86.2% were achieved for LAD, RCA, and LCx respectively. The proposed loss function remains competitive to standardized functions, while also producing heatmaps which less frequently assign relevance to incorrect regions.</p>}},
author = {{Karlsson, Jennie and Heyden, Anders and Åström, Karl and Overgaard, Niels Christian and Dahl, Julia and Davidsson, Anette and Figueroa, Miguel Ochoa and Arvidsson, Ida}},
booktitle = {{Medical Imaging 2026 : Computer-Aided Diagnosis}},
editor = {{Wismuller, Axel and Deserno, Thomas Martin}},
isbn = {{9781510697898}},
issn = {{1605-7422}},
keywords = {{Convolutional neural networks; Explainability; Grad-CAM; Myocardial perfusion imaging}},
language = {{eng}},
month = {{04}},
publisher = {{SPIE}},
series = {{Progress in Biomedical Optics and Imaging - Proceedings of SPIE}},
title = {{Explainable Deep Learning for Myocardial Perfusion Imaging with a Grad-CAM-Based Loss Function}},
url = {{http://dx.doi.org/10.1117/12.3086120}},
doi = {{10.1117/12.3086120}},
volume = {{13926}},
year = {{2026}},
}