Nucleus detection method for pathological images based on visual language large model
By iteratively training the student network through self-training knowledge distillation and generating prompt word sequences using a large visual language model, the training optimization difficulties in cell nucleus detection in pathological images are solved, and efficient cell nucleus detection results are achieved.
CN118298422BActive Publication Date: 2026-07-21BEIHANG UNIV
View PDF 2 Cites 0 Cited by
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2024-03-18
- Publication Date
- 2026-07-21
AI Technical Summary
Technical Problem
Existing methods for detecting cell nuclei in pathological images based on large visual language models face challenges in training and optimization. The detection network cannot effectively inherit the pre-trained feature extraction capabilities, leading to suboptimal results.
Method used
A self-trained knowledge distillation method is adopted to generate prompt word sequences using a large visual language model, and the student network is iteratively trained through self-trained knowledge distillation to improve detection performance.
Benefits of technology
It achieves efficient cell nucleus detection without labeling, improves detection results, and optimizes the performance of the detection network.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN118298422B_ABST
Abstract
The present application relates to a kind of pathological image cell nucleus detection methods based on visual language large model, the method is based on existing pre-training visual language large model and language large model such as GLIP, BLIP, GPT etc., pathological image cell nucleus detection method, the method is a new automatic generation prompt flow, the degree of consideration in the description of adjective of the cell nucleus to be detected, and relevance ranking is introduced, improve the effect of zero sample detection cell nucleus detection.In training method, self-training knowledge distillation, the pre-training feature extraction capability of GLIP is migrated to the network structure more suitable for detection, and finally a more optimal cell nucleus detection network is obtained, and the training of the network does not need any labeling.Compared with the previous unsupervised pathological cell nucleus detection method, the final detection effect of the method of the present application is better.
Need to check novelty before this filing date? Find Prior Art