Anatomical and Functional Assessment of Coronary Artery Disease Using Machine Learning

A computer-implemented method for computing Fractional Flow Reserve (FFR) from angiography data, the method comprising:

• receiving, by one or more processors, two-dimensional (2D) angiography image data of a vessel inspection region containing a coronary vessel tree;
• generating, by the one or more processors, a three-dimensional (3D) geometric vessel tree model from the 2D angiography image data;
• extracting, by the one or more processors, flow data from dynamic angiography images depicting progression of contrast agent through the coronary vessel tree;
• applying, by the one or more processors, the 3D geometric vessel tree model and the extracted flow data to a computational fluid dynamics model;
• and
• computing, by the one or more processors, an FFR value for at least one vessel in the coronary vessel tree based on output from the computational fluid dynamics model;
• determining, by the one or more processors, a state of vessel occlusion for the at least one vessel based on the computed FFR value;
• and
• generating, by the one or more processors, an indication of coronary artery disease based on the determined state of vessel occlusion.

A computer-implemented method for extracting flow data from dynamic angiography for use in FFR determination, the method comprising:

• receiving, by one or more processors, dynamic angiography image data comprising a plurality of angiography images captured over a sampling time period, the images depicting transport of a contrast agent through a coronary vessel tree;
• applying, by the one or more processors, a vessel segmentation machine learning model to each angiography image to generate a series of 2D segmented vessel images;
• counting, by the one or more processors, vessel pixels in each 2D segmented vessel image to generate an array of vessel pixel counts;
• normalizing, by the one or more processors, the array of vessel pixel counts;
• and
• extracting, by the one or more processors, flow data for the coronary vessel tree based on the normalized array of vessel pixel counts.

A computer-implemented method for determining a state of vessel occlusion using a graph-theory based reduced order model, the method comprising:

• receiving, by one or more processors, a 1D segmented vessel tree model representing a coronary vessel tree;
• receiving, by one or more processors, boundary conditions on flow and pressure extracted through data assimilation;
• generating, by the one or more processors, a dense graph of discrepancies between ground truth data and a low-fidelity model of flow, wherein the ground truth data comprises one of in silico simulations of 3D Navier-Stokes equations or in vivo data acquired in a catheterization laboratory;
• deriving, by the one or more processors, a reduced order model from the dense graph of discrepancies;
• and
• computing, by the one or more processors, FFR, instantaneous wave-free ratio (iFR), or quantitative flow ratio (QFR) for the coronary vessel tree using the reduced order model.

A computer-implemented method for training a vessel segmentation machine learning model using synthetic angiograms, the method comprising:

• generating, by one or more processors, a plurality of synthetic angiogram images simulating coronary vessel trees with varying stenosis characteristics;
• augmenting, by the one or more processors, the synthetic angiogram images through geometric transformations comprising one or more of:
• horizontal flip, vertical flip, rotation, zoom, or shear;
• combining, by the one or more processors, the augmented synthetic angiogram images with clinical angiogram images to form a training dataset;
• and
• training, by the one or more processors, a vessel segmentation machine learning model using the training dataset.