Incremental diagnostic value of coronary CT angiography-derived pericoronary adipose tissue radiomics for identifying hemodynamically significant coronary stenosis.

BACKGROUND

Pericoronary adipose tissue (PCAT) radiomics may complement coronary computed tomography angiography (CCTA)-based anatomical and physiological assessment. This study aimed to evaluate its incremental diagnostic value for identifying hemodynamically significant coronary stenosis.

METHODS

This retrospective study screened 21,438 consecutive patients and included 81 patients with 107 target vessels who underwent CCTA, invasive coronary angiography, and invasive fractional flow reserve (FFR) measurement within 30 days. Patients were allocated at the patient level to training and validation cohorts in a 7:3 ratio. A radiomics score (Radscore) was developed from lesion-specific PCAT features using training-only preprocessing and least absolute shrinkage and selection operator logistic regression. Models incorporating diameter stenosis (DS), CCTA-derived FFR (CT-FFR), fat attenuation index (FAI), and Radscore were evaluated against invasive FFR ≤0.80 using patient-cluster bootstrap inference.

RESULTS

The training and validation cohorts comprised 57 and 24 patients with 75 and 32 vessels, respectively, without patient overlap. In validation, Radscore alone had an area under the receiver operating characteristic curve (AUC) of 0.684 (95% confidence interval [CI], 0.451-0.873). CT-FFR+ Radscore showed higher discrimination than CT-FFR alone (AUC, 0.879 vs. 0.762; difference, 0.117; 95% CI, 0.005-0.229; P = 0.041), as did DS+ Radscore vs. DS alone (AUC, 0.812 vs. 0.619; difference, 0.193; 95% CI, 0.018-0.367; P = 0.030). The FAI comparison was not significant (P = 0.238). At training-derived cutoffs, CT-FFR+ Radscore had lower sensitivity (50.0% vs. 78.6%) and negative predictive value (69.6% vs. 81.3%) but higher specificity (88.9% vs. 72.2%).

CONCLUSIONS

Adding Radscore was associated with higher discrimination for CT-FFR and DS in internal validation but did not uniformly improve threshold-dependent classification. Independent external validation is warranted.