Image segmentation method and device based on dual condition compatible neural network

Abstract

The invention relates to an image segmentation method and device based on a dual condition compatible neural network. The method comprises the following steps: S1, obtaining image data which has the same structural mode as a to-be-segmented image and has a part of labels as training data; S2, constructing a main segmentation network and a dual segmentation network, inputs of the main segmentation network and the dual segmentation network comprising target data and partial annotation data providing conditional prior, and outputs being segmentation results of the target data; S3, respectively determining loss functions of the main segmentation network and the dual segmentation network; S4, training the main segmentation network and the dual segmentation network by using the training data; and S5, carrying out image segmentation: taking the to-be-segmented image as target data, inputting the target data and the partial annotation data providing condition priori into the main segmentation network, and outputting a segmentation result. Compared with the prior art, accurate image segmentation can be realized under the condition of a small amount of training data.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an image segmentation method and device based on a dual condition compatible neural network. Background technique [0002] In the field of medical imaging, the annotation of medical images requires a lot of labor costs and is extremely difficult to obtain. Partially annotated medical images serving specific medical purposes are more common. For example, in the diagnosis of myocardial activity, only the myocardium is annotated in cardiac MRI images; in the diagnosis of right ventricular abnormalities, only the right ventricular region is often annotated. Partially supervised medical image segmentation can save doctors' annotation time and make full use of existing partially annotated image data. [0003] At present, in the existing partially supervised segmentation methods, the methods of ignoring the unlabeled structure or treating the unlabeled structure as the backgro...

AI Technical Summary

Problems solved by technology

However, this method ignores the information of the unlabeled structure, and it is difficult to achieve accurate segmentation with a small amount of training data.

Images