Deep Learning From Coronary Computed Tomography Angiography for Atherosclerotic Plaque and Stenosis Quantification and Cardiac Risk Prediction

Abstract

Background: Atherosclerotic plaque quantification from coronary computed tomography angiography (CCTA) enables accurate assessment of coronary artery disease burden and prognosis. We sought to develop and externally validate a deep learning (DL) system for CCTA-derived measures of plaque volume and stenosis severity. Methods: This was an international multicenter study of patients undergoing CCTA at 11 sites. A novel DL convolutional neural network was trained to segment coronary plaque in 921 patients (5,045 lesions). The DL network was then applied to an independent test set, which included an external validation cohort of 175 patients (1,081 lesions) as well 50 patients (84 lesions) assessed by intravascular ultrasound (IVUS) within one month of CCTA. Thereafter, we evaluated the prognostic value of DL-based plaque measurements for fatal or nonfatal myocardial infarction (MI) in 1,611 patients from the prospective SCOT-HEART trial. Findings: In the overall test set, there was excellent agreement between DL and expert reader measurements of total plaque volume (intraclass correlation coefficient [ICC] 0·964) and percent diameter stenosis (ICC 0·879; both p<0.001). When compared with IVUS, there was excellent agreement for DL total plaque volume (ICC 0·949) and minimal luminal area (ICC 0·904). The average per-patient DL plaque analysis time was 5.7 seconds versus 25-30 minutes taken by experts. Over a median follow-up of 4·7 years, MI occurred in 41 patients (2·5%) from the SCOT-HEART trial. DL-based total plaque volume ≥238·5mm 3 was associated with an increased risk of MI (HR 5·36, 95% CI 1·70-16·86; p=0·004) after adjustment for the presence of DL-based obstructive stenosis (HR 2·49, 95% CI 1·07-5·50; p=0·009) and the cardiovascular risk score (HR 1·01, 95% CI 0·99-1·04; p=0·35). Interpretation: A novel externally validated DL system provides rapid measurements of plaque volume and stenosis severity from CCTA which agree closely with expert readers and IVUS and carry prognostic value for future MI. Funding: National Heart, Lung, and Blood Institute. Declaration of Interest: Outside of the current work, S.C., P.S., and D.D. received software royalties from Cedars-Sinai Medical Center. D.B, P.S., and D.D. hold a patent (US8885905B2 in USA and WO patent WO2011069120A1, Method and System for Plaque Characterization). All other authors have nothing to declare. Ethical Approval: This work was approved by the Institutional Review Boards and all patients provided written informed consent.