Related Experiment Video
Updated: Aug 23, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Multi-modal contrastive mutual learning and pseudo-label re-learning for semi-supervised medical image segmentation
Shuo Zhang1, Jiaojiao Zhang1, Biao Tian1
1State Key Laboratory of Reliability and Intelligence of Electrical Equipment, School of Health Sciences and Biomedical Engineering, Hebei University of Technology, China; Tianjin Key Laboratory of Bioelectromagnetic Technology and Intelligent Health, School of Health Sciences and Biomedical Engineering, Hebei University of Technology, China.
This study introduces a semi-supervised contrastive mutual learning (Semi-CML) framework for medical image segmentation using multi-modal data. The method significantly improves segmentation accuracy while drastically reducing annotation costs.
Area of Science:
- Medical Image Analysis
- Machine Learning
- Computer Vision
Background:
- Semi-supervised learning shows promise for medical image segmentation with limited labeled data.
- Existing multi-modal approaches often require all data modalities during both training and inference, limiting clinical applicability.
- Multi-modal data offers potential for enhanced segmentation performance compared to single-modal approaches.
Purpose of the Study:
- To develop a novel semi-supervised framework for multi-modal medical image segmentation.
- To address the limitations of coupled multi-modal models requiring data at inference.
- To improve segmentation performance across all modalities and reduce annotation burden.
Main Methods:
- Proposed a semi-supervised contrastive mutual learning (Semi-CML) framework.
- Introduced an area-similarity contrastive (ASC) loss for cross-modal learning and prediction consistency.
- Developed a soft pseudo-label re-learning (PReL) scheme to balance performance gaps between modalities.
Main Results:
- Semi-CML with PReL significantly outperformed state-of-the-art semi-supervised methods on two public multi-modal datasets.
- Achieved performance comparable to fully supervised methods using 100% labeled data.
- Demonstrated a 90% reduction in data annotation cost.
Conclusions:
- The proposed Semi-CML framework with PReL effectively leverages multi-modal data for improved medical image segmentation.
- The method offers a practical solution for reducing annotation costs while maintaining high segmentation accuracy.
- ASC loss and PReL module are effective components for enhancing semi-supervised multi-modal segmentation.
