Liver image segmentation method based on deep learning

Abstract

The invention discloses a liver image segmentation method based on deep learning, and the method comprises the steps: firstly obtaining an abdomen CT image as a data set, carrying out the marking of a liver image, obtaining a liver mask image, and carrying out the preprocessing of the data set; constructing a liver image segmentation model comprising a coarse segmentation neural network and a fine segmentation neural network, training the coarse segmentation neural network by using the training set and the corresponding liver mask image, and training the fine segmentation neural network by using an image block sequence cut after channel superposition of a coarse segmentation result and a preprocessed data set, and finally obtaining a segmentation result. According to the method, the coarse segmentation neural network is used for processing the abdomen CT image at the small resolution to obtain the overall coarse segmentation result of the liver, then the fine segmentation neural network is used for obtaining the fine segmentation result at the full resolution based on the coarse segmentation result, overall and local combination is achieved, and the liver segmentation effect is improved.

Description

technical field [0001] The invention belongs to the field of medical image processing, and in particular relates to a liver image segmentation method based on deep learning. Background technique [0002] Liver cancer is a malignant disease with a high morbidity and mortality rate in the world. Traditional medicine's judgment of the size, shape, and location of liver tumors relies on radiologists to analyze the CT scans of each patient one by one. Therefore, automatic liver and tumor segmentation technology has very important clinical application value. Automatic liver segmentation from enhanced CT volume scans is a very challenging task due to the low contrast of the liver to its immediate neighbors. To solve this problem, radiologists typically inject contrast agents to enhance CT scans to see tumors clearly, but this can greatly increase noise in images of areas of the liver. Compared with liver segmentation, tumors have different locations, shapes, sizes, and numbers in...

AI Technical Summary

Problems solved by technology

The segmentation method based on image grayscale can get good segmentation results when the grayscale difference between the target and the background in the image is obvious, but when the difference is not obvious, the segmentation accuracy will be greatly reduced; the segmentation method based on shape structure, Segmentation is based on the prior information of the target structure in the image, which is not suitable for the special situation of the liver shape; most of the segmentation methods based on texture features need to combine the relevant knowledge of machine learning and pattern recognition, and can use more representative images in the image. characteristic image features, the results obtained are also very close to the results of manual segmentation

However, the extraction of texture features relies on feature descriptors. However, these descriptors are designed by experts, and the features that can be expressed are limited. Moreover, there are many types of feature descriptors, and it is urgent to study which descriptors should be selected for different images. problem, while developing a new feature extraction operator is very time-consuming and requires a certain amount of knowledge and experience

In addition, some medical image segmentation methods require some preprocessing such as image registration, and the preprocessing results will directly affect the segmentation results.

Images