A bidirectional cross-connected convolutional neural network for image segmentation

Abstract

A bidirectional cross-connected convolutional neural network for image segmentation to perform simultaneous and accurate segmentation of different objects of interest in multimodal medical images by introducing two different networks into the existing BiO‑Net segmentation network Branches and a new cross-skip connection to achieve effective extraction of different interest targets, segmentation experiments based on public fundus images show that: the present invention can effectively extract the optic disc and optic cup regions in the fundus image, and the acquisition is better than U-Net and BiO Segmentation performance of existing networks such as ‑Net.

Description

technical field [0001] The invention specifically relates to the technical field of image segmentation and target detection, in particular to a bidirectional cross-connected convolutional neural network for image segmentation. Background technique [0002] Image segmentation is a technique that divides the entire image into several independent local areas based on imaging characteristics such as grayscale distribution and tissue contrast. This technology can be used for tasks such as the understanding and analysis of medical images, the detection and positioning of lesions, and the measurement and evaluation of morphological characteristics, so it has important clinical diagnosis and academic research value. Based on this, a large number of image segmentation algorithms have been proposed. These algorithms can be roughly divided into unsupervised and supervised segmentation algorithms according to different image evaluation strategies. The unsupervised segmentation algorit...

AI Technical Summary

Problems solved by technology

However, although this U-Net network has good segmentation performance, it is difficult to deal with the boundary area of ​​the target and has a large boundary detection error because (a) the network uses image downsampling operations many times, although speeding up The detection efficiency of convolution features will greatly reduce the image resolution, resulting in the blurring of the target boundary and the loss of a large amount of texture information; (b) the segmentation network only uses one-way jump connections to establish the connection between the encoding and decryption convolution modules , which is not conducive to the detection and integration of multi-level and multi-dimensional image information

In order to overcome the shortcomings of the U-Net network, various improvements have been made to construct networks such as M-Net, BiO-Net and U-Net++; however, these networks use fewer jump connections, which are not enough There are many deficiencies in alleviating the information loss problem caused by multiple downsampling

Images