White matter high-signal segmentation method based on multi-scale fusion and attention splitting

Abstract

The invention discloses a white matter high-signal segmentation method based on multi-scale fusion and attention splitting, and belongs to the technical field of brain magnetic resonance image processing. According to the technical key points, the method comprises the following steps: obtaining a white matter high-signal FLAIR image data set, and carrying out the division and preprocessing of the data set; constructing and training the white matter high-signal segmentation model based on multi-scale fusion and split attention, and obtaining the white matter high-signal segmentation model when the training meets a termination condition; and inputting each image in the test set into the trained white matter high-signal segmentation model based on multi-scale fusion and split attention for testing. The beneficial effects are that the white matter high-signal segmentation method based on multi-scale fusion and split attention can effectively improve the WMH segmentation accuracy, and especially has good recognition capability for small lesions. The method has reference significance for segmentation of other medical images and also has positive significance for promoting computer-aided diagnosis of brain diseases.

Description

technical field [0001] The present invention relates to the technical field of brain magnetic resonance image processing, in particular to a 3D U-Net full convolution model using multi-scale convolution and distraction, which is used to solve the problem of white matter hyperintensities (WMH) Automatically segment questions. Background technique [0002] White matter hyperintensity (WMH) refers to the bright local area that appears on T2-weighted images of MRI and FLAIR images, also known as white matter lesions. WMH is commonly seen in brain magnetic resonance images (MRI) of patients with neurodegenerative diseases (e.g., dementia, Alzheimer's disease), stroke, and cerebral small vessel disease, as well as brain structures in healthy older adults over the age of 70 middle. Previous studies have shown that the size, location, number and shape of WMH can provide valuable information for exploring the etiology and development of brain diseases and evaluating the effect of t...

AI Technical Summary

Problems solved by technology

However, most of the existing deep learning methods directly adopt the classic fully convolutional neural network U-Net model in the field of image segmentation or simply improve it when solving the WMH segmentation problem, such as simply adding skip connections, or in the bottleneck layer or All layers use attention indiscriminately. These seemingly improved methods do not improve the segmentation effect significantly due to lack of pertinence. Therefore, the accuracy of segmentation is not high, there are small lesions that are missed, and the segmentation of lesion boundaries is not accurate. The problem

The reason is that the model does not fully consider the characteristics of WMH when designing, which makes the feature extraction ability of the model insufficien

Images