Correlation of machine learning computed tomography-based fractional flow reserve with instantaneous wave free ratio...

Fractional flow reserve based on coronary CT angiography (CT-FFR) is gaining importance for non-invasive hemodynamic assessment of coronary artery disease (CAD). We evaluated the on-site CT-FFR with a machine learning algorithm (CT-FFR_ML) for the detection of hemodynamically significant coronary artery stenosis in comparison to the invasive reference standard of instantaneous wave free ratio (iFR^®).

This study evaluated patients with CAD who had a clinically indicated coronary computed tomography angiography (cCTA) and underwent invasive coronary angiography (ICA) with iFR^®-measurements. Standard cCTA studies were acquired with third-generation dual-source computed tomography and analyzed with on-site prototype CT-FFR_ML software.

We enrolled 40 patients (73% males, mean age 67 ± 12 years) who had iFR^®-measurement and CT-FFR_ML calculation. The mean calculation time of CT-FFR_ML values was 11 ± 2 min. The CT-FFR_ML algorithm showed, on per-patient and per-lesion level, respectively, a sensitivity of 92% (95% CI 64-99%) and 87% (95% CI 59-98%), a specificity of 96% (95% CI 81-99%) and 95% (95% CI 84-99%), a positive predictive value of 92% (95% CI 64-99%), and 87% (95% CI 59-98%), and a negative predictive value of 96% (95% CI 81-99%) and 95% (95% CI 84-99%). The area under the receiver operating characteristic curve for CT-FFR_ML on per-lesion level was 0.97 (95% CI 0.91-1.00). Per lesion, the Pearson's correlation between the CT-FFR_ML and iFR^® showed a strong correlation of r = 0.82 (p < 0.0001; 95% CI 0.715-0.920).

On-site CT-FFR_ML correlated well with the invasive reference standard of iFR^® and allowed for the non-invasive detection of hemodynamically significant coronary stenosis.