Cell segmentation data set expansion method based on diffusion model

Through the method based on the diffusion model, unconditional and conditionally controlled diffusion models are trained to generate new cell images and their segmentation masks, which solves the problems of high manual annotation cost and low algorithm annotation accuracy in cell segmentation data set expansion in the prior art, and achieves more reliable data expansion and improved model robustness.

CN119941751APending Publication Date: 2025-05-06HARBIN UNIV OF SCI & TECH
0 Cites 0 Cited by

Patent Information

Application Number
CN202510011775.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-01-05
Publication Date
2025-05-06

AI Technical Summary

Technical Problem

The prior art has problems such as high manual annotation cost and low algorithm annotation accuracy in the expansion of cell segmentation data sets, resulting in the generated cell segmentation mask that does not match the image, and additionally increases the accuracy loss.

Method used

Using a diffusion model-based method, the unconditional cell segmentation mask diffusion model and conditional control cell image diffusion model are trained to generate new cell images and their segmentation mask data, thereby expanding the original data set.

Benefits of technology

The disadvantages of manual and algorithm labeling are effectively avoided, and the generated cell images and their masks are more credible, increasing the diversity of the data set, reducing the risk of overfitting, and improving the robustness of the model.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a cell segmentation data set expansion method based on a diffusion model, and belongs to the technical field of digital image processing. According to the method, a condition control embedded cell segmentation mask and a U-Net-based cell data set generation network are designed, cell segmentation mask features are embedded into a diffusion model, and a cell segmentation mask generation diffusion network is constructed. A data set expansion model is obtained through training by taking a cell image as a noise adding and restoring image and taking a segmentation mask as a condition control image. The model generates a corresponding cell image by using a segmentation mask to expand a data set, generates a segmentation mask through an unconditional diffusion model, and performs training by using a segmentation mask image in an original data set. Experiments show that the method can effectively expand the cell data set and improve the generalization ability of the cell segmentation model. Compared with an unexpanded data set, the model trained by using the expanded data set has higher precision on the verification set.
Need to check novelty before this filing date? Find Prior Art