Automated opportunistic cardiovascular risk assessment in non-small cell lung cancer patients on routine chest CT using an optimised nnU-net framework
File(s)
Published version
Author(s)
Type
Journal Article
Abstract
Background
Cardiovascular disease (CVD) and non-small cell lung cancer (NSCLC) are the global leading causes of overall and cancer-related deaths, respectively. NSCLC patients have a higher CVD risk than the general population which is frequently underdiagnosed. Coronary artery calcification (CAC), a marker of CVD, is commonly detected on routinely acquired CT from NSCLC work-up but often not reported. We present an automated CAC assessment tool validated for NSCLC patients using a deep learning-based framework to provide a non-invasive CVD screening opportunity without incurring extra workload or radiation exposure.
Methods
We trained nnU-Net models on ungated, unenhanced chest CTs (n = 97) from Stanford AIMI dataset, and tested them on three mutually independent datasets: (1) ungated unenhanced CTs from AIMI (n = 95), (2) attenuation correction CTs from PET-CT scans of NSCLC patients at our institution (ICHNT, n = 87; age 67.8 ± 10.1 years; M:F 174:113), and (3) CAC-negative scans from TCIA (n = 50); and used the best performing model to produce CAC segmentations, post-processed with TotalSegmentator, to stratify patients into CVD risk groups, informing the need for dedicated cardiac clinic assessment.
Results
For a CAC threshold of 100, the model achieved accuracy: 83.6%, sensitivity: 91.9%, specificity: 70.8%, positive predictive value (PPV): 82.9%, negative predictive value (NPV): 85.1%, F1-score: 0.87, kappa coefficient: 0.65 and Area Under the Receiver Operating Characteristic Curve (AUC) score of 0.899. For a threshold of 400, accuracy: 84.5%, sensitivity: 90.9%, specificity: 79.5%, PPV: 77.6%, NPV: 91.8%, F1-score: 0.84, and kappa coefficient: 0.69 as well as an AUC of 0.926.
Conclusion
Our optimised deep learning model can benefit NSCLC patients by providing CVD risk information from their routine CT scans which may not acted upon otherwise, thus enabling a practical opportunistic screening solution for these patients.
Cardiovascular disease (CVD) and non-small cell lung cancer (NSCLC) are the global leading causes of overall and cancer-related deaths, respectively. NSCLC patients have a higher CVD risk than the general population which is frequently underdiagnosed. Coronary artery calcification (CAC), a marker of CVD, is commonly detected on routinely acquired CT from NSCLC work-up but often not reported. We present an automated CAC assessment tool validated for NSCLC patients using a deep learning-based framework to provide a non-invasive CVD screening opportunity without incurring extra workload or radiation exposure.
Methods
We trained nnU-Net models on ungated, unenhanced chest CTs (n = 97) from Stanford AIMI dataset, and tested them on three mutually independent datasets: (1) ungated unenhanced CTs from AIMI (n = 95), (2) attenuation correction CTs from PET-CT scans of NSCLC patients at our institution (ICHNT, n = 87; age 67.8 ± 10.1 years; M:F 174:113), and (3) CAC-negative scans from TCIA (n = 50); and used the best performing model to produce CAC segmentations, post-processed with TotalSegmentator, to stratify patients into CVD risk groups, informing the need for dedicated cardiac clinic assessment.
Results
For a CAC threshold of 100, the model achieved accuracy: 83.6%, sensitivity: 91.9%, specificity: 70.8%, positive predictive value (PPV): 82.9%, negative predictive value (NPV): 85.1%, F1-score: 0.87, kappa coefficient: 0.65 and Area Under the Receiver Operating Characteristic Curve (AUC) score of 0.899. For a threshold of 400, accuracy: 84.5%, sensitivity: 90.9%, specificity: 79.5%, PPV: 77.6%, NPV: 91.8%, F1-score: 0.84, and kappa coefficient: 0.69 as well as an AUC of 0.926.
Conclusion
Our optimised deep learning model can benefit NSCLC patients by providing CVD risk information from their routine CT scans which may not acted upon otherwise, thus enabling a practical opportunistic screening solution for these patients.
Date Issued
2026-04-08
Date Acceptance
2026-02-23
Citation
BMC Medical Imaging, 2026, 26 (1)
ISSN
1471-2342
Publisher
BMC
Journal / Book Title
BMC Medical Imaging
Volume
26
Issue
1
Copyright Statement
© TheAuthor(s) 2026. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
License URL
Identifier
10.1186/s12880-026-02252-z
Subjects
6%
sensitivity: 91.9%
specificity: 70.8%
positive predictive value (PPV): 82.9%
negative predictive value (NPV): 85.1%
F1-score: 0.87
kappa coefficient: 0.65 and Area Under the Receiver Operating Characteristic Curve (AUC) score of 0.899. For a threshold of 400
accuracy: 84.5%
sensitivity: 90.9%
specificity: 79.5%
PPV: 77.6%
NPV: 91.8%
F1-score: 0.84
Artificial intelligence
Cardiovascular disease
Lung cancer
Computed tomography
Opportunistic screening
Publication Status
Published
Article Number
ARTN 179
Date Publish Online
2026-03-03