A semi-supervised method for removing metal artifacts in multi-spectral ct images

By combining a semi-supervised average teacher-student network and the U-net model with voxel-level contrastive learning and an unsupervised loss function, the problem of removing metal artifacts in multi-energy spectral CT images is solved, achieving efficient artifact removal while preserving tissue details, and is suitable for medical image processing.

CN116664429BActive Publication Date: 2026-07-03TIANJIN TIANXIN MICROSYSTEM INTEGRATION RES INST CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
TIANJIN TIANXIN MICROSYSTEM INTEGRATION RES INST CO LTD
Filing Date
2023-05-29
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively remove metal artifacts from multi-energy spectral CT images, especially since unsupervised learning models suffer from poor training stability and loss of tissue structural details, while semi-supervised methods are still immature.

Method used

A semi-supervised average teacher-student network is adopted, combined with the U-net model and sequence feature attention module. Through voxel-level contrastive learning and unsupervised loss function, the artifact information of low-energy and high-energy images is exchanged, and the loss function is reduced to improve the model's artifact removal effect.

Benefits of technology

It improves the removal of artifacts in multi-spectral CT images, preserves more tissue detail information, and provides high-quality image input for subsequent medical diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116664429B_ABST
    Figure CN116664429B_ABST
Patent Text Reader

Abstract

This invention relates to a semi-supervised method for removing metal artifacts in multi-energy spectral CT images. Based on a semi-supervised average student-teacher model, the U-net network structure incorporates a slice feature attention module, enhancing the perception of deep feature information between channels and improving the feature connections between different monoenergy maps of multi-energy spectral CT slices. Simultaneously, in unsupervised learning, the student model, combined with voxel-level contrastive learning, makes the model more focused on artifact regions, enabling artifact information interaction between low-energy and high-energy maps. This allows for better removal of artifacts in low-energy maps while preserving more tissue detail information. Consequently, it provides high-quality CT images for subsequent medical diagnostic tasks such as image classification and segmentation.
Need to check novelty before this filing date? Find Prior Art