Pericoronary Radiomics Signature for Non-Culprit Lesion Progression and Revascularization Decision in NSTE-ACS

Background/Objectives: This study aimed to establish a CCTA-based radiomics model of perivascular adipose tissue (PCAT) to identify high-risk NCLs in patients with NSTE-ACS, potentially facilitating early risk stratification. Methods: In this prospective cohort of 542 NSTE-ACS patients, pericoronary adipose tissue (PCAT) radiomic features of non-culprit lesions (NCLs) were extracted from baseline coronary CTA. Patients were assigned to training (n = 379) and validation (n = 163) cohorts. Machine learning algorithms were applied to develop a radiomics signature (Rad model) to predict 4-year NCL-related major adverse cardiovascular events (MACEs). A combined clinic-radiomics model was constructed to enhance predictive performance. Additionally, the association between the baseline Rad model and longitudinal non-calcified plaque progression (ΔNCPV%/year) was evaluated in a subcohort (n = 60) with serial CCTA. Results: Over a median 4.0-year follow-up, NCL-related MACE occurred in 84 patients (15.5%). The Rad model (comprising nine features) independently predicted MACE (adjusted hazard ratio, 1.988; 95% CI, 1.753-2.254; p < 0.001). In the validation cohort, the combined model yielded higher discrimination for 4-year MACE than the clinical model alone (AUC, 0.793 vs. 0.703; p < 0.05). In the serial CCTA subgroup, a higher baseline Rad model was significantly associated with annualized non-calcified plaque volume progression (standardized β, 0.477; p < 0.001). Conclusions: A CCTA-based PCAT radiomics model is associated with future NCL-related MACE and accelerated plaque progression in patients with NSTE-ACS. This approach may serve as a non-invasive tool for individualized risk stratification.