Detection of calcified plaques: comparison between coronary CT angiography and thin-slice non-contrast CT with deep learning-aided image registration.
Authors
Affiliations (12)
Affiliations (12)
- Department of Radiology, Charité-Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, Berlin, Germany. [email protected].
- Department of Radiology, Charité-Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, Berlin, Germany.
- Inria, Université Côte d'Azur, Epione team, Sophia Antipolis, France.
- Institute of Medical Sciences, National University of Medical Sciences, Ulaanbaatar, Mongolia.
- Coronary Care Unit, Shastin's Third State Central Hospital, Ulaanbaatar, Mongolia.
- Department of Radiology, Medical Imaging Centre, Semmelweis University, Budapest, Hungary.
- German Cancer Research Center (DKFZ), Heidelberg, Germany.
- Heidelberg University, Heidelberg, Germany.
- Department of Artificial Intelligence, Fraunhofer Heinrich Hertz Institute, Berlin, Germany.
- Department of Electrical Engineering and Computer Science, Technical University of Berlin, Berlin, Germany.
- BIFOLD (Berlin Institute for the Foundations of Learning and Data), Berlin, Germany.
- BIH (Berlin Institute of Health), DHZC (German Heart Center of the Charité) and DZHK (German Center for Cardiovascular Research), Berlin, Germany.
Abstract
To investigate whether coronary CT angiography (CCTA) misses calcified plaques detected by thin-slice non-contrast CT (NCCT). This study included patients from two sites in the DISCHARGE trial for whom both 0.5 mm thin-slice NCCT and CCTA were available. Plaques on CCTA were defined as missed if they showed no spatial overlap with NCCT-detected plaques after deep learning-aided co-registration. Comparisons of plaque volume, density, and local coronary luminal attenuation between plaques missed and those detected by CCTA were performed using the Mann-Whitney U-test. In addition, the presence of these plaques on standard calcium scoring CT was assessed. Interobserver agreement was assessed using the intraclass correlation coefficient and Bland‒Altman analysis. This study included 45 patients (40% female, mean age 62 ± 11 years), in whom CCTA missed 37.6% of calcified plaques detected by NCCT (121/322). Missing calcified plaques on CCTA misclassified 8.9% of patients (4/45) as having no plaques. Compared with detected plaques, plaques missed by CCTA were both significantly smaller in volume (3.0 mm³ [IQR, 1.5-4.9] vs. 9.2 mm³ [IQR, 4.3-21.9], p < 0.001) and had lower density (212.7 HU [IQR, 174.5-242.4] vs. 292.7 HU [IQR, 243.2-361.3], p < 0.001). Only 44.0% of plaques (53/121) missed by CCTA were detected by standard calcium scoring CT. Interobserver analysis demonstrated excellent agreement for calcified plaque volume on CCTA (ICC = 0.91) and NCCT (ICC = 0.98). CCTA missed more than one-third of coronary calcified plaques that are identifiable on co-registered thin-slice NCCT, which suggests an underutilized role of thin-slice NCCT in clinical practice. Question Accurate detection of all coronary plaques is crucial for risk stratification. CCTA misses calcified plaques, which are detectable by thin-slice non-contrast CT (NCCT). Findings CCTA misses over one-third of calcified plaques, nearly half of which are also missed by calcium scoring CT. NCCT detected these calcified plaques. Clinical relevance Deep learning-aided registration enables multimodal CCTA-NCCT assessment, improving detection of calcified plaques overlooked by CCTA alone and providing more accurate plaque burden quantification that may support better clinical decision-making, which should be investigated in future studies.