Key result
Deep learning cMPR analysis detects vulnerable plaque with ~93% sensitivity and correlates with quantitative angiography.
Why the study?
This study aimed to develop and validate a deep learning-based image analysis method to detect and quantify coronary artery stenosis, vulnerable plaque, and calcification in Curved-Multiplanar Reconstruction images for aiding pre-operative evaluation of CAD.
Does a deep learning-based image analysis method using a 2D nnU-Net framework accurately detect and quantify coronary artery stenosis, vulnerable plaque, and calcification in Curved-Multiplanar Reconstruction (cMPR) images compared to expert annotation and conventional quantitative angiographic analysis?
Observational (n=2,978)
Does a deep learning-based image analysis method using a 2D nnU-Net framework accurately detect and quantify coronary artery stenosis, vulnerable plaque, and calcification in Curved-Multiplanar Reconstruction (cMPR) images compared to expert annotation and conventional quantitative angiographic analysis?
This automated deep learning-based image analysis method demonstrates high sensitivity and specificity for detecting vulnerable plaque and a notable correlation with conventional quantitative angiography for evaluating stenosis in cMPR scans.
Introduction This study developed and validated a deep learning-based image analysis method to detect and quantify coronary artery stenosis, vulnerable plaque, and calcification in Curved-Multiplanar Reconstruction (cMPR) images for aiding pre-operative evaluation of coronary artery disease (CAD). Methods A total of 2,978 cMPR scans, were collected based on the data utilized between 2014 and 2022 under Institutional Review Board approval from 2020 October extended through 2023 October. Each cMPR scan was reconstructed to generate 13 images per cMPR using the CT Cardiac Package by TeraRecon Inc. From these, 769 CT images of a time independent test dataset were used for evaluation. 10 years-experienced expert meticulously annotated locations of stenosis and vulnerable plaques to train and test a stenotic segmentation model (SSM) and a vulnerable plaque segmentation model (VPSM) using 2D nnU-Net framework. SSM facilitated the assessment of stenosis positioning and length as well as the determination of CAD diameter profiles through Full Width at Half Maximum analysis, leading to the calculation of a stenosis regression index (SRI). The detection of vulnerable plaques was also performed using VPSM for differentiating normal and vulnerable plaques. Sensitivity and specificity of each slice and patient were primary metrics for evaluation. Results A notable correlation of 0.6874 was established between SRI and conventional quantitative angiographic analysis. The correlation of stenosis length quantification was 0.4033. Five expert cardiologists visual scoring for aiding calcification scoring with calcium segmentation averaged 7.277 ± 2.286 on a 10-point scale. For vulnerable plaque detection, the VPSM achieved a sensitivity of 0.929 and a specificity of 0.910. Conclusion The automated quantitative image analysis developed in this study demonstrated robust potential for evaluating SRI and for detecting calcification and vulnerable plaque in cMPR scans, showing promise for application in clinical screening environments. This non-invasive, cost-effective, rapid, and user-independent method offers a significant advancement in cMPR screening, providing clinicians with a powerful tool for the detection of CAD stenosis, calcification, and vulnerable plaque.
No takes yet. Share an insight, caveat, or question.
D A E W O N Kim (2026) conducted an observational in Coronary artery disease (n=2,978). Deep learning-based image analysis (2D nnU-Net stenotic and vulnerable plaque segmentation models) vs. Conventional quantitative angiographic analysis and expert cardiologist annotation was evaluated on Sensitivity and specificity for vulnerable plaque detection and correlation with quantitative angiographic analysis (Sensitivity 0.929, Specificity 0.910; SRI correlation r=0.6874). Deep learning-based cMPR analysis achieved 0.929 sensitivity and 0.910 specificity for vulnerable plaque detection and correlated with quantitative angiography for stenosis (r=0.6874).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:
More from European Heart Journal