An automatic medical image segmentation method based on multi-path attention fusion

Abstract

The invention belongs to the technical field of medical image processing and computer vision, and in particular relates to a medical image automatic segmentation method based on multi-path attention fusion. The images in the training set are augmented, and the images in the validation set and the augmented training set are normalized; the images in the training set are input into the multi-path attention fusion network model, and the segmentation results are obtained under the guidance of the cross entropy loss function. Figure; select the model with the highest accuracy in the verification set, input the test set into the multi-path attention fusion network loaded with this model, and output the segmentation result map of the image; the present invention solves the problem that the existing network in the medical image segmentation process cannot be used in the encoder. It can effectively improve the quality of features at different scales, and it is difficult to control the inter-layer dependencies between low-level structural features and high-level semantic features of the network, resulting in poor segmentation results.

Description

technical field [0001] The invention belongs to the technical field of medical image processing and computer vision, in particular to an automatic medical image segmentation method based on multi-path attention fusion. Background technique [0002] Medical images play a key role in medical treatment and diagnosis. The goal of computer-aided diagnosis (CAD) systems is to provide physicians with accurate interpretations of medical images to enable better treatment of large numbers of patients. Also, automated processing of medical images results in reduced time, cost and errors of human-based processing. One of the main areas of research in this field is medical image segmentation, which is a key step in numerous medical imaging studies. [0003] Like other areas of search in computer vision, deep learning networks achieve excellent results and outperform non-deep techniques in medical imaging. Deep neural networks are mainly used for classification tasks, where the output ...

AI Technical Summary

Problems solved by technology

However, in the hierarchical transformation in the U-Net network, the learning process of different pooling levels often shares the same data path, so the generated multi-scale feature map may not be fully distinguished as expected. Due to the existence of the pooling layer, the encoder Losing part of the spatial information, the single U-Net network uses simple two consecutive 3×3 convolutional layers. This standard convolutional layer is difficult to preserve rich spatial information. Another disadvantage is that the U-Net network uses It is a simple skip connection, which only splices the features of the encoder and the features of the decoder, and the features output from the decoder will be redundant, which will affect the final segmentation result

Images