A Multimodal Nasopharyngeal Carcinoma Image Segmentation Method Based on Attention and Graph Convolution

By proposing a multimodal nasopharyngeal carcinoma image segmentation method based on attention and graph convolution, attention feature maps of each modality are generated and feature fusion is performed, which solves the accuracy problem of single-modal segmentation methods and achieves adaptive and efficient nasopharyngeal carcinoma image segmentation.

CN115187613BActive Publication Date: 2026-03-31GUANGDONG UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-07-12
Publication Date
2026-03-31

AI Technical Summary

Technical Problem

Existing nasopharyngeal carcinoma image segmentation methods mainly target a single modality, making it difficult to achieve accurate segmentation results, and are greatly affected by the doctor's experience and subjectivity.

Method used

A multimodal nasopharyngeal carcinoma image segmentation method based on attention and graph convolution is adopted. It generates attention feature maps for each modality, and uses multimodal feature fusion recalibration technology to generate multimodal fusion feature maps with cross-modal interaction. Finally, graph convolution is used to generate the final feature map to capture the long-distance dependencies of the global context.

Benefits of technology

It achieves adaptive and accurate segmentation of nasopharyngeal carcinoma images, improving segmentation accuracy and efficiency while reducing reliance on physician experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115187613B_ABST
    Figure CN115187613B_ABST
Patent Text Reader

Abstract

The application provides a kind of multi-modal nasopharyngeal carcinoma image segmentation method based on attention and graph convolution, comprising the following steps: S1: obtaining nasopharyngeal carcinoma image, and inputting nasopharyngeal carcinoma image into image segmentation model;S2: extracting visual feature map of each modality of nasopharyngeal carcinoma image using shared encoder;S3: each modality of channel attention feature map is generated according to the visual feature map of each modality;S4: each modality of spatial attention feature map is generated according to the channel attention feature map of each modality;S5: the spatial attention feature map of each modality is fused and recalibrated to obtain multi-modal fusion feature map;S6: generating final feature map according to multi-modal fusion feature map using graph convolution;S7: the final feature map is input into decoder to obtain segmentation result.The application provides a kind of multi-modal nasopharyngeal carcinoma image segmentation method based on attention and graph convolution, solves the problem that current nasopharyngeal carcinoma image segmentation method is only segmented for single modality.
Need to check novelty before this filing date? Find Prior Art