Cerebrovascular image segmentation method based on multi-attention dense connection generative adversarial network

Abstract

The invention discloses a cerebrovascular image segmentation method based on a multi-attention dense connection generative adversarial network. The network structure mainly comprises three parts: 1, asegmentation generation network S: based on a U-shaped structure, image feature information is reserved to the maximum extent in a contraction path by using a densely connected convolution layer so as to accurately segment a blood vessel detail contour, and discrimination features are highlighted by using an attention module in an expansion path so as to alleviate confusion of similar regions during segmentation; 2, a segmentation discrimination network D: standard real segmentation is introduced as a condition, a foreground cerebrovascular region is ninput to a densely connected convolutionlayer under the shielding of a segmentation image, and concentrating is carried out on extracting of cerebrovascular region features to distinguish generate segment and truly segment; and 3, deep enhancement of an adversarial loss function: employing deep supervision: a Wasserstein distance adversarial loss, a generative network, an error attention type weighted cross entropy loss are used to focuse on solving a class imbalance problem and reducing a segmentation error. The method can be widely applied to computer-aided diagnosis, treatment and the like of cerebrovascular diseases.

Description

Technical field [0001] The invention relates to a cerebrovascular image segmentation method based on a multi-attention dense connection generation confrontation network, in particular to a cerebrovascular magnetic resonance image segmentation method based on a multi-attention dense connection generation confrontation network (MA-DenseGAN), which belongs to digital image processing , Pattern recognition and medical imaging engineering technology. It mainly involves Convolutional Neural Network (CNN) and Generative Adversarial Strategy (GAN), which can be widely used in computer-aided diagnosis and treatment systems for cerebrovascular diseases. Background technique [0002] As a specific application research branch of medical image segmentation, cerebrovascular image segmentation technology can assist the diagnosis and analysis of clinical cerebrovascular diseases, facilitate the planning and execution of treatment plans after the diagnosis of the disease, and effectively realize ...

AI Technical Summary

Problems solved by technology

[0004] In cerebrovascular imaging, on the one hand, there are large differences in the shape and grayscale of cerebrovascular vessels. Specifically, the diameter and brightness of different vascular branches, especially the aorta and the end of the blood vessels, are significantly different, making it difficult to realize the image of the cerebrovascular region. Complete segmentation; on

Images