Additional Value of Machine-Learning Computed Tomographic Angiography-Based Fractional Flow Reserve Compared to...

Background: Machine-learning-based computed-tomography-derived fractional flow reserve (CT-FFR_ML) obtains a hemodynamic index in coronary arteries. We examined whether it could reduce the number of invasive coronary angiographies (ICA) showing no obstructive lesions. We further compared CT-FFR_ML-derived measurements to clinical and CT-derived scores. Methods: We retrospectively selected 88 patients (63 ± 11years, 74% male) with chronic coronary syndrome (CCS) who underwent clinically indicated coronary computed tomography angiography (cCTA) and ICA. cCTA image data were processed with an on-site prototype CT-FFR_ML software. Results: CT-FFR_ML revealed an index of >0.80 in coronary vessels of 48 (55%) patients. This finding was corroborated in 45 (94%) patients by ICA, yet three (6%) received revascularization. In patients with an index ≤ 0.80, three (8%) of 40 were identified as false positive. A total of 48 (55%) patients could have been retained from ICA. CT-FFR_ML (AUC = 0.96, p ≤ 0.0001) demonstrated a higher diagnostic accuracy compared to the pretest probability or CT-derived scores and showed an excellent sensitivity (93%), specificity (94%), positive predictive value (PPV; 93%) and negative predictive value (NPV; 94%). Conclusion: CT-FFR_ML could be beneficial for clinical practice, as it may identify patients with CAD without hemodynamical significant stenosis, and may thus reduce the rate of ICA without necessity for coronary intervention.