Automatic segmentation of coronary plaques in coronary CT angiography using neural networks
Mahdi Moosavi, Keno Bressem, Rafael Adolf, Anastasiya Valentik, Albrecht Will, Eva Hendrich, Martin Hadamitzky
Abstract
Rapid and accurate detection of coronary plaques on CCTA is critical for timely CAD diagnosis but is limited by reader workload and interobserver variability. Our objective was to evaluate the effectiveness of machine learning (ML) based on automated segmentation of coronary plaques in coronary computed tomography angiography (CCTA).
Introduction
Coronary artery disease (CAD) remains a major cause of preventable mortality, particularly among older individuals, with its incidence continuing to rise globally [1,2]. Accurate estimation of disease burden is crucial for identifying patients at high risk of major adverse cardiac events [3].
Materials and methods
In this study, to reduce the selection bias, we enrolled 1642 consecutive patients who underwent CCTA for suspected coronary artery disease (CAD) between 2014 and 2018 in German Heart Center of Munich, TUM University Hospital.
Results
From a total of 2090 plaques across 1112 vessels in our test dataset, the neural network model correctly identified 1772 plaques (TP) (Fig 3). It failed to identify 318 plaques (FN) and mistakenly marked 382 parts of coronary arteries as plaques (FP) that had no overlap with ground truth annotations (Fig 4).
Discussion
In this study, we developed and evaluated a neural network model based on nnU-Net for automated detection and segmentation of coronary plaques in CCTA images.
Conclusions
Our nnU-Net-based model demonstrates strong segmentation performance for medium-to-large and calcified coronary plaques in CCTA, with high negative predictive value for ruling out significant disease. Performance remains limited for small and non-calcified lesions, and variability in image quality reduces reliability in challenging cases.
Citation: Moosavi M, Bressem K, Adolf R, Valentik A, Will A, Hendrich E, et al. (2026) Automatic segmentation of coronary plaques in coronary CT angiography using neural networks. PLoS One 21(2): e0343887. https://doi.org/10.1371/journal.pone.0343887
Editor: Redoy Ranjan, James Cook University Hospital, UNITED KINGDOM OF GREAT BRITAIN AND NORTHERN IRELAND
Received: August 26, 2025; Accepted: February 12, 2026; Published: February 24, 2026
Copyright: © 2026 Moosavi et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: To support reproducibility, we provide an anonymized sample dataset together with the source code and model weights in a public repository (https://github.com/MM-DHM/nnUNet-Coronary-CTA-Segmentation). The main analysis was performed on clinical coronary CT angiography scans without individual patient consent. Bavarian and German data protection regulations permit the use of such data for in-house scientific research only. Patient-level data cannot be made publicly available without prior patient consent. External researchers may request controlled on-site access to the anonymized study dataset for research purposes. Requests should be addressed to the institutional point of contact at the Department of Cardiovascular Radiology and Nuclear Medicine, TUM University Hospital German Heart Center. Contact: Department of Cardiovascular Radiology and Nuclear Medicine TUM University Hospital German Heart Center Lazarettstraße 36, 80636 Munich, Germany Email: rad-sekretariat@dhm.mhn.de. Further information is provided by the local Institutional Review Board (Ethikkommission der Technischen Universität München, Ismaninger Straße 22, 81675 München).
Funding: The author(s) received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
