Pancreas CT image segmentation method based on integrated deep convolutional neural network

Abstract

The invention discloses a pancreas CT image segmentation method based on an integrated deep convolutional neural network. A two-stage segmentation framework from coarse to fine is adopted to accurately segment pancreas in a CT image. The method comprises the following steps: firstly, constructing a CNN (Convolutional Neural Network) of a three-dimensional U-shaped encoding and decoding structure introducing an attention module and cross-level dense connection, namely, applying a Unet model as an identification network to two stages of pancreas image segmentation; in the coarse segmentation stage, downsampling normalization preprocessing is performed on an original image, and then a plurality of data blocks are randomly taken as input of a network for training so as to obtain a pancreas coarse segmentation result; in the fine segmentation stage, using a bounding box including a pancreas region, and taking image blocks in the bounding box region for training; during identification, a pancreas region being determined by using a coarse segmentation result, and then prediction being performed by using fine segmentation so as to obtain a fine segmentation result. And finally, voting anddeciding the results of the two stages are carried out to obtain a segmentation result. According to the method, the problem of manual labeling is solved, and an ideal segmentation result is obtained.

Description

technical field [0001] The invention belongs to the medical field, and in particular relates to a CT image segmentation method. Background technique [0002] Pancreatic cancer is a highly malignant tumor of the digestive system. Its early clinical symptoms are relatively hidden, and most of them are discovered at an advanced stage, so the prognosis is often poor. Although the overall incidence of pancreatic cancer is low among all malignant tumors, But the mortality rate is in the forefront, and the incidence rate has an upward trend in recent years. Using the existing computer-aided diagnosis system to automatically and accurately segment the pancreas in CT images can greatly facilitate the assessment of pancreatic diseases. However, due to the particularity of the pancreas, the segmentation effect is often far from the ideal situation. The difficulty of pancreas segmentation mainly lies in: the serious category imbalance problem, that is, the proportion of pancreas in the...

AI Technical Summary

Problems solved by technology

However, due to the particularity of the pancreas, the segmentation effect is often far from the ideal situation. The difficulty of pancreas segmentation mainly lies in: the serious category imbalance problem, that is, the proportion of pancreas in the entire CT image is often less than 1%. At the same time, the anatomy of the pancreas is relatively complex, and there is a visually blurred interclass boundary relative to other tissues

For example, Roth et al. used the high representation ability of the CNN model to image features to effectively segment the pancreatic tissue in the CT image. They took the lead in using the Full Convolution Network (FullConvolution Network, FCN) for the segmentation of the pancreas, and demonstrated the CNN model. Potential in medical image segmentation, but whether it is a CNN model based on 2D convolution or 3D convolution, there are certain defects

The CNN model based on 2D convolution can only process two-dimensional CT image slices, and cannot learn the three-dimensional information contained in the overall CT image.

The CNN model based on 3D convolution uses 3D data blocks as input. This method fully pays attention to the context information in 3D CT images, but 3D image features often need to occupy a huge video memory, thus limiting the size of the input data. At the same time, due to the unbalanced category of pancreas segmentation, randomly selected 3D image blocks often cannot contain enough positive samples to effectively train the network.

Images