UNSUPERVISED TRAINING FOR AN ATLAS-BASED REGISTRATION

摘要

A method includes obtaining a single training image from a set of training images in a data repository. The method further includes generating an initial tissue class atlas based on the obtained single training image. The initial tissue class atlas includes two or more different tissue class images corresponding to two or more different tissue classes. The method further includes registering the remaining training images of the set of training images to the initial tissue class atlas. The method further includes generating a quality metric for each of the registered images. The method further includes evaluating the quality metric of each of the registered image with a predetermined evaluation criterion. The method further includes identifying a sub-set of images from the set of training images that satisfy the evaluation criterion. The method further includes generating a subsequent tissue class atlas based on the identified sub-set of the set of training images.

主权项

1. A method, comprising:
obtaining a single training image from a set of training images in a data repository; generating an initial tissue class atlas based on the obtained single training image, wherein the initial tissue class atlas includes two or more different tissue class images corresponding two or more different tissue classes; registering the remaining training images of the set of training images to the initial tissue class atlas; generating a quality metric for each of the registered images; evaluating the quality metric of each of the registered image with a predetermined evaluation criterion; identifying a sub-set of images from the set of training images that satisfy the evaluation criterion; and generating a subsequent tissue class atlas based on the identified sub-set of the set of training images.