Cross-domain adaptive spine CT image segmentation method

Through the diffusion model, forward diffusion and reverse denoising are performed in CT image segmentation, combined with time embedding and adaptive frequency domain enhancement, the problem of domain differences in multi-source CT data sets is solved, and high-precision cross-domain adaptive segmentation is achieved, reducing costs and improving stability.

CN119888237AActive Publication Date: 2025-04-25JIANGSU SHIYU INTELLIGENT MEDICAL TECH CO LTD
4 Cites 1 Cited by

Patent Information

Application Number
CN202510356764.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-03-25
Publication Date
2025-04-25
Estimated Expiration
2045-03-25

AI Technical Summary

Technical Problem

When the prior art processes domain differences in multi-source CT data sets, the cross-domain adaptive feature alignment effect is poor, resulting in poor segmentation effect of target domain data. The multi-source data fusion and denoising process in the diffusion model lacks an efficient processing mechanism, resulting in unstable performance.

Method used

The forward diffusion model is used to perform forward diffusion and reverse denoising processes. By gradually adding noise and using encoder and decoder to build a model architecture, combining time embedding and adaptive frequency domain enhancement, cross-domain feature alignment and segmentation are achieved.

Benefits of technology

The segmentation accuracy between multi-source CT data sets is significantly improved, especially in the target domain that has not been seen, the segmentation effect is better than traditional methods, and the dependence on a large amount of labeled data is reduced, the cost of data annotation and model training is improved, and the stability of the model is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119888237A_ABST
    Figure CN119888237A_ABST
Patent Text Reader

Abstract

The invention provides a cross-domain self-adaptive spine CT image segmentation method. The method comprises the following steps of performing a forward diffusion process on data; constructing a model architecture; a reverse denoising process; and optimizing a loss function. The method has the beneficial effects that the working performance can be improved: the segmentation precision of the frame between multi-source CT data sets is greatly improved, particularly, better cross-domain adaptation can be realized on a domain without a target, and the segmentation effect is superior to that of a traditional method; by using a diffusion model and a comparative learning mechanism, the scheme can effectively reduce the dependence on a large amount of annotated data on data sets from other sources, and the cost of data annotation and model training is reduced; the cross-domain feature alignment effect is more stable through domain-guided asymmetric contrast learning, and the performance fluctuation of the model on different data sets is reduced.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Hyperspectral image multi-source domain adaptive classification method based on diffusion model

    CN118247668A

  • Image restoration method of enhanced conditional diffusion model based on dual-domain interactive Transform

    CN118537249A

  • Unsupervised domain adaptation method for medical image segmentation

    CN118735948A

  • Diffusion model and generative adversarial network-based image segmentation method and apparatus

    WO2025050542A1

Cited By

  • Vertebral bone tissue segmentation method combined with structure-guided conditional diffusion network

    CN121121125A