Neural Network-Based Assessment of Coronary Angiogram Flow

Adequate blood flow through the coronary tree is critical for maintaining cardiac perfusion. Coronary angiography has the potential to provide rich and dynamic hemodynamic information, however, current strategies to assess flow through the coronary vessels depend on either subjective expert opinion (TIMI flow grade) or laborious frame-by-frame anatomical analysis (TIMI frame count and quantitative flow ratio). Here we present a strategy for automated characterization of bulk flow through the coronary tree using the right coronary artery as an example. We leverage the AngioNet neural network to generate sequential segmentations of angiograms, create time series of the summed segmentation intensities (i.e., a contrast intensity profile), and quantitatively characterize the filling and washout phases of these intensity profiles. We demonstrate that AngioNet-derived frame counts and normalized mean filling slopes of contrast intensity profiles correlate well with manual frame counts and flow grades in both our derivation and validation datasets. Furthermore, the generated washout dynamics appear to provide different information to the traditional frame count and flow grade metrics, which only deal with the initial filling phase, suggesting that washout dynamics of the contrast intensity profiles may capture novel information about the coronary microcirculation.