A Cedars-Sinai and Amsterdam UMC-led team developed a machine learning system that analyzes quantitative plaque in coronary CTA exams to identify patients with ischemia and impaired myocardial blood flow (MBF), potentially creating an alternative to current methods.
The researchers trained the ML model using invasive FFR data from 254 patients (484 FFR vessels) to predict ischemia and impaired MBF by analyzing plaque data in CCTA exams.
They then tested it with CCTAs from 208 patients (581 vessels) who also underwent invasive FFR and H2O PET exams, finding that the CCTA ML scores:
- Predicted FFR-defined ischemia far more accurately than standard CCTA stenosis evaluations, while rivaling FFRCT assessments (AUCs: 0.92 vs. 0.84 & 0.93)
- Predicted PET-based impaired MBF more accurately than standard CCTA stenosis evaluations and FFRCT assessments (AUCs: 0.80 vs. 0.74 & 0.77)
Because the ML scoring system operates locally, the authors highlighted its potential to quickly assess high-risk patients before invasive coronary angiography (avoiding off-site processing delays) or to assess low-risk patients at earlier stages, helping to improve ICA efficiency and accuracy.
The researchers plan to continue to develop their CCTA plaque AI solution, including adding more plaque features and CCTA metrics, and potentially seeking regulatory approval depending on the results of future validation studies.
CCTA plaque AI is already one of the hottest segments on the commercial side of imaging AI, and this study highlights similar advances in academic centers, while showing that CCTA plaque AI can quickly and accurately predict both ischemia and lower MBF.