Artificial intelligence coronary computed tomography angiography for lesion specific ischaemia: invasive physiology validation across plaque burden
S B Satyabhisma, H ParkAbstract
Background/Introduction
Lesion-specific assessment of coronary ischaemia by non-invasive testing remains difficult, especially in patients with high plaque burden and calcification.
Purpose
To compare vessel-territory diagnostic performance of an artificial intelligence quantitative coronary computed tomography angiography ischaemia model (AI-QCT) with fractional flow reserve computed tomography (FFR-CT) across plaque volume strata.
Methods
This was a post-hoc pooled analysis of two prospective studies. Symptomatic patients with suspected coronary artery disease underwent coronary computed tomography angiography, FFR-CT, and invasive fractional flow reserve (FFR). Analyses were performed at the vessel-territory level. The reference standard was invasive FFR <=0.80. Performance was evaluated overall and stratified by total plaque volume, calcified plaque, and non-calcified plaque (<50, 50–150, and >150 mm3). Accuracy and AUC were reported with 2-sided bootstrapped 95% confidence intervals using cluster sampling; AUC was estimated using Mann–Whitney statistics.
Results
The pooled cohort included 513 patients and 1,480 vessel territories. Overall, AI-QCT showed higher accuracy and discrimination than FFR-CT: accuracy 81.0% (95% CI 78.7–82.9) vs 75.4% (72.8–78.0); AUC 0.86 (0.84–0.88) vs 0.77 (0.74–0.80); p<0.0001. Across total plaque volume strata (<50/50–150/>150 mm3), accuracy remained higher for AI-QCT: 90.2/79.6/74.8% vs 82.4/73.5/71.3%. In severe calcification (calcified plaque >150 mm3), accuracy was 71.2% vs 68.5% (AUC 0.78 vs 0.68; p=0.010). In severe non-calcified plaque (non-calcified plaque >150 mm3), accuracy was 77.1% vs 73.2% (AUC 0.89 vs 0.79; p<0.0001).
Conclusion(s)
In a pooled dataset using invasive physiology as the reference standard, AI-QCT outperformed FFR-CT for detecting vessel-territory ischaemia and maintained performance across increasing total, calcified, and non-calcified plaque burden.ESC 2026 Table