Utility of Artificial Intelligence Plaque Quantification: Results of the DECODE Study

Abstract

BackgroundArtificial Intelligence Plaque Analysis (AI-QCPA, HeartFlow) provides, from a CCTA, quantitative plaque burden information including total plaque and plaque subtype volumes. We sought to evaluate the clinical utility of AI-QCPA in clinical decision making. MethodsOne hundred cases were reviewed by 3 highly experienced practicing cardiologists who are SCCT level 3 CCTA readers. Patients had varying levels of calcium (median CACS: 99.5) and CAD-RADS scores. Initial management plan for each case was a majority decision based upon patient demographics, clinical history, and CCTA report. AI-QCPA was then provided for each patient, and the plan was reconsidered. The primary endpoint was the reclassification rate (RR). In a secondary analysis of 40 cases, the above process was repeated but the initial plan was based upon review of the actual CCTA images. ResultsRR following AI-QCPA review was 66% (66/100) of cases (95% CI, 56.72%-75.28%). RR ranged from 47% in cases with CACS 0 to 96% in cases with CACS >400, and from 40% in CAD-RADS 1 cases to 94% in CAD-RADS 4 cases. RR was higher in cases with coronary stenoses ≥50% (89.5%) vs cases with stenoses <50% (51.6%). RR was 39% in cases with LDL <70 mg/dL vs 70% in LDL ≥70 mg/dL. Following review of the CCTA images rather than the CCTA report, the RR was 50% (95% CI of 34.51% - 65.49%). The primary reclassification effect was to intensify preventative medical therapy. ConclusionsAdding AI-QCPA to CCTA alone leads to a change in clinical care in two-thirds of patients.