Incremental Predictive Value of Deep Learning-Quantified Coronary Atherosclerotic Volume: A SCAPIS Cohort Analysis.
Authors
Affiliations (8)
Affiliations (8)
- Department of Clinical Sciences, Danderyd Hospital, Karolinska Institutet, Stockholm, Sweden. Electronic address: [email protected].
- Department of Biomedical Engineering and Health Systems, KTH Royal Institute of Technology, Stockholm, Sweden.
- Department of Molecular and Clinical Medicine, Institute of Medicine, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden; Region Västra Götaland, Sahlgrenska University Hospital, Department of Clinical Physiology, Gothenburg, Sweden.
- Department of Clinical Sciences in Malmö, Lund University, Malmö, Sweden.
- Department of Medical Sciences and Uppsala Clinical Research Center, Uppsala University, Uppsala, Sweden.
- Department of Clinical Sciences, Danderyd Hospital, Karolinska Institutet, Stockholm, Sweden.
- Department of Public Health and Clinical Medicine, Umeå University, Umeå, Sweden.
- CMIV (Center for Medical Image Science and Visualization), Linköping University, Linköping, Sweden; Department of Health, Medicine, and Caring Sciences, Linköping University, Linköping, Sweden.
Abstract
Coronary plaque volume assessed by coronary computed tomography angiography (CTA) may refine risk assessment beyond established risk models. This study aims to evaluate whether deep learning-derived coronary plaque volume from coronary CTA improves prediction beyond established risk models and imaging markers in a general population. Participants were randomly selected men and women 50-64 years of age from a population-based cohort, without prior atherosclerotic cardiovascular disease, with available SCORE2 (Systematic Coronary Risk Evaluation 2) data and coronary CTAs of acceptable quality. Total plaque volume (TPV) and noncalcified plaque volume (NCPV) were quantified using automated deep learning software and stratified into 6 categories. Participants were followed for coronary heart disease death or myocardial infarction over a median of 7.8 years. Among 23,314 participants, 287 coronary events occurred. Risk of events increased stepwise with increasing TPV and NCPV, including among participants with coronary artery calcium score = 0 and in those without manually detected atherosclerosis. In Cox regression analyses, adding TPV to a model including SCORE2 and segment involvement score significantly improved model performance (C-statistic: 0.798 vs 0.783; P = 0.022) and net reclassification index (0.093; 95% CI: 0.041-0.146). HRs increased with higher TPV categories, reaching 6.36 (95% CI: 3.00-13.52) in the highest category. Adding presence of stenosis ≥50% or any segment with only noncalcified plaque did not materially improve the C-statistic. Replacing TPV with NCPV yielded similar results. Automated deep learning-quantified TPV improved prediction of first coronary events beyond SCORE2, coronary artery calcium score, and manually derived coronary CTA measurements, and identified even small plaque volumes associated with increased risk in a general population without established atherosclerotic cardiovascular disease.