A Lymph Node Detection Method Based on Improved Segnet Segmentation Network

Abstract

The invention discloses a lymph node detection method based on an improved SegNet segmentation network, comprising: dividing a lymph node image data set into a training set and a test set; constructing a SegNet segmentation network based on a hole convolution operation; using the training set to train the SegNet segmentation network, Taking the minimization of the sine-cosine cross-entropy loss function as the network optimization objective function, the SegNet segmentation network is optimized; the trained SegNet segmentation network is used to identify and segment the lymph nodes in the lymph image to be identified. The invention uses dilated convolution to extract features, increases its receptive field area without increasing the amount of extra calculation, avoids the loss of down-sampling information, and solves the problem of lowering the resolution of sampled images. Through the sine-cosine cross-entropy loss function, the samples with small prediction errors are given less weight than the cross-entropy loss function, which solves the problem of unbalanced training positive and negative samples. The segmentation result is post-processed by Markov random field to further refine the edge part of the segmented object.

Description

technical field [0001] The invention belongs to the technical field of image segmentation, and more particularly relates to a lymph node detection method based on an improved SegNet segmentation network. Background technique [0002] Traditional medical image segmentation techniques can be divided into three categories: (1) segmentation based entirely on images, where all the information required for segmentation comes from the image itself; (2) methods based on target models, which add the required segmentation The prior information of the object (eg, the shape information of the object). Commonly, there are segmentation methods based on graphs (Atlas); (3) hybrid methods, first use image-based information for preliminary segmentation, and then achieve further segmentation of the target based on prior constraint information. The semantic segmentation network based on deep learning technology belongs to the first category of segmentation technology, that is, the information...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the defects of the existing technology, the purpose of the present invention is to solve the technical problem that the existing technology does not fully consider the imbalance of positive and negative samples in the training network, and fails to segment the target from the perspective of multi-scale and multi-resolution

Images