Segmentation algorithm evaluation method and device based on anatomy prior

Abstract

The invention provides a segmentation algorithm evaluation method and device based on anatomy prior, and the method comprises the steps: carrying out the shape processing of the relative position information of the coordinates of a central point of each segmentation target region according to an active shape model algorithm and a Principal analysis method, and obtaining a processed shape; according to the training sample set data, calculating an average value of the processed shape, a characteristic value of a covariance matrix and a characteristic vector value of the covariance matrix; obtaining a shape coefficient value according to the average value of the processed shape and the feature vector value of the covariance matrix; according to the shape coefficient value and the eigenvalue of the covariance matrix, obtaining the prior of the relative position stability of the total segmentation target area; obtaining the unique prior of each segmentation target area according to the segmentation deviation value of each segmentation target area; and obtaining a training model according to the priori that the relative position of the total segmentation target area is stable and the unique priori of each segmentation target area, and evaluating the current segmentation algorithm.

Description

technical field [0001] The present invention relates to the field of medical imaging, in particular to a method and device for evaluating segmentation algorithms based on anatomical prior. Background technique [0002] Medical image analysis methods based on deep learning models are receiving more and more attention. In order to achieve good analysis results, a large amount of data needs to be labeled for model training and testing. Since high-quality medical data labeling requires doctors to have rich experience, it is very important to select the most informative data for labeling by evaluating the effect of existing models when labeling resources are limited. [0003] Existing segmentation algorithm evaluation methods mainly fall into two categories. One is to directly calculate the evaluation index value based on the difference between the segmentation result and the label when there is labeling, but this method is obviously not suitable for unlabeled data. Another ty...

AI Technical Summary

Problems solved by technology

There are two problems in doing this. One is that the training time is long, which makes the iterative update of the model slow; the other is that a large number of features need to be manually designed, but the anatomical prior information of the task itself is not fully utilized.

Images