Blood vessel segmentation method combining global and neighborhood information

Abstract

A blood vessel segmentation method combining global and neighborhood information is disclosed. The method, on the basis of a Unet network, proposes a Gnet segmentation network, which uses a brand new down-sampling mode and combines with a long-short term memory network at a jump joint, thereby optimizing the extraction of feature information, and reducing the loss of image information; then the network disclosed in the previous step is used as a basic segmentation framework and a parallel network Rnet for calculating neighborhood information is proposed, and a loss function of the network is used as a penalty term of the basic segmentation network to train the segment network Gnet. According to the method, the accuracy of cerebral blood vessel segmentation in MRA image processing is remarkably improved, and the problem of small blood vessel segmentation fracture can be effectively solved.

Description

technical field [0001] The invention relates to the fields of medical image processing and artificial intelligence, and is a deep learning-based cerebrovascular image segmentation method. Background technique [0002] At present, cerebrovascular disease has become one of the diseases with the highest fatality rate and the lowest recovery rate among neurosurgical diseases. Whether the segmentation of medical images is accurate or not determines whether doctors can provide reliable basis for diagnosis and treatment in clinic. Moreover, in different clinical medical fields such as neurosurgery and cardiovascular and cerebrovascular, the segmentation and reconstruction of blood vessels is very important for the diagnosis, treatment plan and evaluation of clinical outcomes. Therefore, accurate and fast segmentation of blood vessels has become one of the hotspots in medical imaging research. [0003] The existing medical image segmentation methods can be divided into two categor...

AI Technical Summary

Problems solved by technology

However, the U-net network based on CNN will inevitably bring about an increase in its calculation amount. Its excessive and redundant use of computing resources and model parameters leads to the repeated extraction of similar low-level features by the model, which reduces its segmentation ability to a certain extent.

And when the U-net network is applied to cerebrovascular segmentation, there will be phenomena such as rupture of small cerebral blood vessels and disconnection with large blood vessels, which greatly reduces the accuracy of vascular segmentation

Combined with the importance of cerebrovascular and the particularity of its location, the inaccurate phenomenon of segmentation accuracy and results has created great obstacles and challenges to current research and clinical applications.

Images