A Coronary CT Angiography Radiomics Model to Identify Vulnerable Plaque and Predict Cardiovascular Events

Background A noninvasive coronary CT angiography (CCTA)-based radiomics technique may facilitate the identification of vulnerable plaques and patients at risk for future adverse events. Purpose To assess whether a CCTA-based radiomic signature (RS) of vulnerable plaques defined with intravascular US was associated with increased risk for future major adverse cardiac events (MACE). Materials and Methods In a retrospective study, an RS of vulnerable plaques was developed and validated using intravascular US as the reference standard. The RS development data set included patients first undergoing CCTA and then intravascular US within 3 months between June 2013 and December 2020 at one tertiary hospital. The development set was randomly assigned to training and validation sets at a 7:3 ratio. Diagnostic performance was assessed internally and externally from three tertiary hospitals using the area under the curve (AUC). The prognostic value of the RS for predicting MACE was evaluated in a prospective cohort with suspected coronary artery disease between April 2018 and March 2019. Multivariable Cox regression analysis was used to evaluate the RS and conventional anatomic plaque features (eg, segment involvement score) for predicting MACE. Results The RS development data set included 419 lesions from 225 patients (mean age, 64 years ± 10 [SD]; 68 men), while the prognostic cohort included 1020 lesions from 708 patients (mean age, 62 years ± 11; 498 men). Sixteen radiomic features, including two shape features and 14 textural features, were selected to build the RS. The RS yielded a moderate to good AUC in the training, validation, internal, and external test sets (AUC = 0.81, 0.75, 0.80, and 0.77, respectively). A high RS (≥1.07) was independently associated with MACE over a median 3-year follow-up (hazard ratio, 2.01; <i>P</i> = .005). Conclusion A coronary CT angiography-derived radiomic signature of coronary plaque enabled the detection of vulnerable plaques that were associated with increased risk for future adverse cardiac outcomes. © RSNA, 2023 <i>Supplemental material is available for this article.</i> See also the editorial by De Cecco and van Assen in this issue.