| 主权项 |
1. A method of generating a category model for classifying medical images, comprising:
providing a plurality of anatomical medical images imaging at least one internal organ or area of a body; in each of said plurality of anatomical medical images identifying a plurality of image patches, each image patch is represented by a plurality of repeatable multidimensional features in a pixel area of a respective anatomical medical image and image coordinates of said pixel area in said respective anatomical medical image, wherein said plurality of repeatable multidimensional features of each of the plurality of image patches comprises at least one noise reduction based coefficient and a gray level feature of the image patch; for each of said image patches, weighting at least some of the noise reduction based coefficient, the gray level feature, and the image coordinates, wherein said weighting is tuned experimentally on a cross-validation set; generating a plurality of visual words by clustering said plurality of image patches represented by said weighted at least one noise reduction based coefficient, gray level feature, and image coordinates of each one of said plurality of image patches; modeling a category model mapping a relation between said plurality of visual words; outputting said category model adapted to categorize a pathology in a new anatomical medical image based on a new set of image patches from said new anatomical medical image and the coordinates of pixel areas of said new set of image patches in said new anatomical medical image, wherein at least one of said providing, generating, modeling, and outputting is performed by at least one processor. |
