Vessel identification using shape and motion mapping for coronary angiogram sequences

摘要

Embodiments of the invention relate to a method, system, and computer program product to automate image classification with respect to coronary vessels in an angiography sequence. Two primary elements are employed, including training and recognition. Training pertains to the pre-processing images and extracting salient features that characterize the appearance of coronary arteries under different viewpoints. Recognition pertains to extraction of features from a new image sequence and determining a classification boundary for the new image from previously classified and labeled image sequences.

主权项

1. A computer program product comprising a computer readable storage device having computer readable program code stored therewith, which when executed causes a computer to implement the method comprising:
detecting and labeling regions of interest in at least a subset of images of the sequence, wherein the region of interest contain a coronary vessel; estimating shape and motion in at least the subset of images of the image sequence, including extracting centerline curves of coronary vessels and computing optical flow within an angiogram sequence; sampling the extracted centerline curves to generate feature points; sampling a surrounding region of the generated feature points to capture shape and motion context of a depicted vessel; forming a feature set for classification of the depicted vessel, wherein features of the represented vessels are constructed from pyramid histograms of shape and motion context; building a first pyramid histogram representation of a shape of the vessel shape and a second pyramid histogram representation of motion of the vessel; and forming a matrix from combined shape and motion vectors, wherein a first dimension in the matrix corresponds to features at each keypoint and a second dimension represents one aspect of the shape context or one aspect of the motion context; determining a set of characteristics separating data in feature space, where the feature set is derived from pyramid histograms of the samplings, and the training employs angiogram sequences with known viewpoint labels; and returning a classification result for a new sequence of images based on a recognized shape and motion of the vessel, including fitting the multiclass support vector machine to computed features and determining where a vessel feature lies in the feature space against a classification boundary.