An Auto Region of Interest Segmentation Method for Cardiac CT Images Based on the Contradiction Labeling Method

Abstract

The whole heart segmentation of medical CT images is of great significance for assisting doctors in the diagnosis of cardiovascular diseases and guiding doctors surgery. Due to the difficulty of labeling medical heart images and the complexity of samples, accurate segmentation of the entire tissue structure of CT cardiac images is still a challenging topic. This paper proposed an automatic method based on neural network learning to segment the region of interest that is meaningful to the doctor. The new method is based on the contradiction law of one of the basic laws of logic. It trains the neural network by labeling small data sets with certain contradictions, so it is named the contradiction labeling method. The method uses contradiction labeling set training, forcing the neural network to better learn the overall characteristics of the medical image, thereby improving the generalization ability of the model training. The experimental results show that the neural network can be used to automatically generate and generate the foreground part of the medical image (that is, the main area of interest of the doctor) by means of a small amount of low-precision labeling. Since there is no need to require large data sets to be annotated, and the labeling accuracy does not need to be too high, the proposed method also satisfies the requirement of requiring as little time as possible.