Device, method, equipment and medium for segmenting damaged tissue based on surgical video

By using polygonal cues to generate sparse and dense features in laparoscopic surgical videos, and combining multi-scale features with long-term memory, the problem of insufficient adaptability to complex morphological features and inter-frame temporal alignment ability of damaged tissue segmentation models is solved, and accurate segmentation of damaged tissue is achieved.

CN122244080APending Publication Date: 2026-06-19NAT UNIV OF DEFENSE TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
NAT UNIV OF DEFENSE TECH
Filing Date
2026-03-25
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing medical image segmentation models struggle to adapt to the complex and varied morphological features of damaged tissues in laparoscopic surgery videos. Their coarse cueing patterns lack guidance and they lack the ability to align temporal features between frames, resulting in poor segmentation performance.

Method used

A surgical video-based damaged tissue segmentation device is used. The interactive module determines polygonal cues, generates sparse and dense cues features, and combines multi-scale feature generation and long-term memory features to achieve accurate segmentation of damaged tissue.

Benefits of technology

It improves the accuracy and stability of damaged tissue segmentation, adapts to significant morphological and positional changes of damaged tissue in surgical videos, enhances the ability to perceive geometric information from polygon cues, and constructs stable temporal features for surgical videos.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122244080A_ABST
    Figure CN122244080A_ABST
Patent Text Reader

Abstract

This application discloses a device, method, equipment, and medium for segmenting damaged tissue based on surgical videos. When segmenting damaged tissue, this solution can determine sparse and dense cue features based on polygonal cues. Sparse cue features can be matched with different learnable embeddings according to geometric semantic types, avoiding geometric ambiguity caused by polygonal cue structures. Dense cue features are generated through the distance feature map of a dense mask image, enhancing the ability to perceive geometric information from polygonal cues during damaged tissue segmentation. This application aligns and fuses the features of the current frame image with those of adjacent frames to obtain target multi-scale features, improving the perception of temporal features in adjacent frames and better adapting to significant morphological and positional changes in damaged tissue between adjacent frames due to surgical operations. By fusing target multi-scale features with long-term memory features and combining sparse and dense cue features, accurate damaged tissue segmentation results can be obtained.
Need to check novelty before this filing date? Find Prior Art