CT image pectoral muscle segmentation method based on improved UNet model

Abstract

The invention relates to a CT image pectoral muscle segmentation method based on an improved UNet model, and the method comprises the steps: carrying out the training of a target UNet neural network model based on each chest CT sample image and a corresponding mask image, obtaining a pectoral muscle recognition model, and achieving the recognition of each type of pectoral muscle region in a to-be-analyzed chest CT image; the improved UNet model applies dilated convolution in an encoder, and adopts two dilated convolution alternate operations with different expansion rates, so that the detail feature information of the local area of pectoral muscles is retained, the convolution receptive field is increased, and the global information of the image is obtained; a batch normalization layer is added between a convolution layer and a pooling layer, so that the network convergence speed is improved, and the over-fitting phenomenon is relieved; and a DUpsampling structure is applied in a decoder, and detail feature information of pectoral muscle edges is fully captured, so that the accuracy and the stability of CT image pectoral muscle positioning segmentation under the real complex condition are improved by the whole method, and the problems of false positive, missing detection and the like existing when a traditional UNet model is used for pectoral muscle segmentation are reduced.

Description

technical field [0001] The invention relates to a CT image chest muscle segmentation method based on an improved UNet model, and belongs to the technical field of image recognition and classification. Background technique [0002] Medical image segmentation can provide very important meaning and value for accurate identification, detailed analysis, and reasonable diagnosis of diseases, and its key task is to segment the region of interest (such as organs or lesions) in medical images. Compared with natural images, medical images have their own characteristics such as low resolution, low contrast, and target dispersion. Therefore, medical image segmentation methods have higher requirements for accuracy and stability. [0003] Traditional medical image segmentation techniques include threshold-based methods, edge-based methods, region-based methods, and specific theory-based methods. meet the requirements of practical applications. SUMMARY OF THE INVENTION [0004] The tec...

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved by the present invention is to provide a CT image pectoralis muscle segmentation method based on the improved UNet model, which can improve the accuracy and stability of CT image pectoralis muscle positioning and segmentation under complex real-world conditions, and reduce the risk of using the traditional UNet model for pectoralis muscle segmentation. False positives and missed detections

Images