Artificial intelligence-powered automatic coronary computed tomography angiography plaque quantification: comparison against optical coherence tomography.
Authors
Affiliations (6)
Affiliations (6)
- Biomedical Instrument Institute, School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai 200030, China.
- Department of Cardiology, Fujian Medical University Union Hospital, Fuzhou, China.
- Department of Radiology, Shanghai Jiao Tong University Affiliated Ruijin Hospital, Shanghai, China.
- College of Health Science and Technology, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
- Division of Cardiovascular Medicine, Radcliffe Department of Medicine, University of Oxford, Oxford, UK.
- Institute of Medical Robotics, Shanghai Jiao Tong University, Shanghai, China.
Abstract
Coronary computed tomography angiography (CCTA) enables a non-invasive, comprehensive assessment of coronary artery disease, and artificial intelligence (AI) offers the potential to improve CCTA image interpretation. This study aimed to evaluate the performance of an AI-powered method for automatic plaque quantification from CCTA, with optical coherence tomography (OCT) as reference standard. Patients who underwent CCTA within 6 months prior to OCT were retrospectively enrolled. AI-assisted automatic plaque quantification was performed on CCTA with specific plaque composition classification based on adaptive Hounsfield unit thresholds. Qualitative high-risk plaque features were also assessed. Automated co-registration of CCTA and OCT was performed with the link of invasive coronary angiography. A total of 91 patients with 153 co-registered lesions were evaluated. The AI-assisted automatic CCTA analysis showed significant correlations with OCT for quantifying plaque volume/burden and different plaque compositions (all <i>P</i> values <0.001); of which, the correlation coefficient for plaque volume was 0.84. Vulnerable plaque, defined as lipid-to-cap ratio >0.33 on OCT, was identified in 39 (25.5%) lesions. CCTA-derived plaque volume >82.5 mm<sup>3</sup> [odds ratio (OR), 9.39], maximal plaque burden >76.4% (OR, 3.70), lipidic tissue volume >16.3 mm³ (OR, 4.42), all <i>P</i> < 0.001, and high-risk plaque features ≥2 (OR, 2.70, <i>P</i> = 0.009) were independent predictors of OCT-derived vulnerable plaques. The average time for automatic CCTA plaque quantification was 1.8 min per patient. The novel AI-powered method facilitated fully automatic plaque quantification and correlated well with co-registered OCT.