A 3D medical image segmentation method based on an attenuation attention network
By designing a decaying attention network and combining ResConv blocks and a top-k perceptual attention mechanism, the problems of insufficient global modeling in CNNs and loss of local details in Transformers are solved, achieving efficient segmentation of 3D medical images and improving the accuracy of multi-organ and heart segmentation.
CN122115857APending Publication Date: 2026-05-29TAIZHOU INST OF SCI &TECH NUST
View PDF 0 Cites 0 Cited by
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TAIZHOU INST OF SCI &TECH NUST
- Filing Date
- 2026-02-12
- Publication Date
- 2026-05-29
AI Technical Summary
Technical Problem
The lack of global modeling in CNNs and the loss of local details in Transformers in existing technologies limit the segmentation accuracy of complex anatomical structures or small lesions.
Method used
A 3D medical image segmentation method based on decaying attention network is designed. It adopts a symmetric U-shaped encoding and decoding structure and combines ResConv block, decaying spatial activation attention block and top-k perceptual attention mechanism block to achieve efficient fusion of local features and global dependencies.
Benefits of technology
It improves the segmentation accuracy of complex 3D medical images, especially the accuracy of segmentation of multiple organs and the heart.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN122115857A_ABST
Abstract
The application relates to the technical field of medical image segmentation, in particular to a 3D medical image segmentation method based on an attenuation attention network, which comprises the following steps: inputting a 3D medical image into an encoder, inputting the features extracted by the encoder into a decoder, performing feature restoration on the decoder, and outputting a final segmentation mask; judging the pixel region of the medical image by using the segmentation mask to obtain a 3D medical image segmentation result; the encoder layer and the decoder both comprise a layered multi-attention layer, the layered multi-attention layer comprises a ResConv block for extracting fine-grained texture features, an attenuation spatial activation attention block for performing global information enhancement on the input features and the features output by the ResConv block, and a top-k perception attention mechanism block; and the top-k perception attention mechanism performs global interaction on the input features and the output features of the attenuation spatial activation attention block. The application aims to solve the problems of insufficient CNN global modeling and local detail loss of the Transformer in the prior art.
Need to check novelty before this filing date? Find Prior Art