Related Experiment Video
Updated: Jul 25, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Multi-ConDoS: Multimodal Contrastive Domain Sharing Generative Adversarial Networks for Self-Supervised Medical Image
IEEE Transactions on Medical Imaging
|June 28, 2023
Summary
This study introduces Multi-ConDoS, a novel method for self-supervised medical image segmentation that overcomes domain shift and multimodality issues. It achieves superior performance with minimal labeled data, outperforming existing methods.
Area of Science:
- Medical Imaging
- Computer Vision
- Machine Learning
Background:
- Self-supervised medical image segmentation faces challenges with domain shift and limited use of multimodal data.
- Existing methods often struggle to generalize across different data distributions or integrate information from multiple imaging modalities.
Purpose of the Study:
- To develop an effective multimodal contrastive self-supervised medical image segmentation method.
- To address the limitations of domain shift and multimodality in existing approaches.
Main Methods:
- Proposed Multimodal Contrastive Domain Sharing (Multi-ConDoS) generative adversarial networks.
- Integrated multimodal contrastive learning, CycleGAN, and Pix2Pix for domain translation.
- Introduced novel domain sharing layers for learning domain-specific and shared information.
Main Results:
- Multi-ConDoS significantly outperforms state-of-the-art self-supervised and semi-supervised methods using only 5% or 10% labeled data.
- Achieved comparable or superior performance to fully supervised methods with significantly less labeled data.
- Ablation studies confirmed the effectiveness of all proposed improvements.
Conclusions:
- Multi-ConDoS demonstrates superior medical image segmentation performance with a drastically reduced labeling workload.
- The method effectively leverages multimodal data and mitigates domain shift issues in self-supervised learning.
