Method and System for Assessing Vessel Obstruction Based on Machine Learning

A method to train a vessel obstruction assessment (VOA) model, comprising:

• obtaining a training database that includes volumetric imaging datasets for multiple patients and corresponding coronary artery disease (CAD) related reference values, the volumetric image data sets being for a target organ that includes a vessel of interest, the CAD related reference values corresponding to one or more points along a vessel of interest within the corresponding imaging data set;
• and
• for at least a portion of the volumetric image data sets and corresponding CAD related reference values,
• extracting an axial trajectory extending along of a vessel of interest (VOI) within the corresponding volumetric image dataset,
• creating a three-dimensional (3D) multi-planer reformatted (MPR) image based on the corresponding volumetric image dataset and the axial trajectory of the VOI, the MPR image extending along the axial trajectory of the VOI, and
• training a machine learning-based vessel obstruction assessment (VOA) model based on the MPR images, the training further comprising extracting, from the MPR images, features characterizing a CAD related parameter along the axial trajectory within the VOI.

A system for assessing a vessel obstruction, comprising:

• memory configured to store a volumetric image dataset for a target organ that includes a vessel of interest;
• and
• at least one processor that, when executing program instructions stored in the memory, is configured to:
• extract an axial trajectory extending along of a vessel of interest (VOI) within the volumetric image dataset,
• create a three-dimensional (3D) multi-planer reformatted (MPR) image based on the volumetric image dataset and the axial trajectory of the VOI, and
• extract a VOI parameter from the MPR image utilizing a machine learning-based vessel obstruction assessment (VOA) model.

A system to train a vessel obstruction assessment (VOA) model, comprising:

• memory configured to store a training database that includes volumetric imaging datasets for multiple patients and corresponding coronary artery disease (CAD) related reference values, the volumetric image data sets being for a target organ that includes a vessel of interest, the CAD related reference values corresponding to one or more points along a vessel of interest within the corresponding imaging data set;
• and
• at least one processor that, when executing program instructions stored in the memory, are configured to:
• for at least a portion of the volumetric image data sets and corresponding CAD related reference values,
• extract an axial trajectory extending along of a vessel of interest (VOI) within the corresponding volumetric image dataset,
• create a three-dimensional (3D) multi-planer reformatted (MPR) image based on the corresponding volumetric image dataset and the axial trajectory of the VOI, the MPR image extending along the axial trajectory of the VOI, and
• train a machine learning-based vessel obstruction assessment (VOA) model based on the MPR images, which involves extracting, from the MPR images, features characterizing a CAD related parameter along the axial trajectory within the VOI.