Related Experiment Video
Updated: Oct 9, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
CyCMIS: Cycle-consistent Cross-domain Medical Image Segmentation via diverse image augmentation
1Institute of Medical Robotics, School of Biomedical Engineering, Shanghai Jiao Tong University, No.800 Dongchuan Road, Shanghai 200240, China.
Medical Image Analysis
|December 17, 2021
Summary
This study introduces CyCMIS, a novel unsupervised domain adaptation method for medical image segmentation. It enhances deep learning model robustness against domain shift without requiring target domain annotations.
Area of Science:
- Medical Image Analysis
- Deep Learning
- Computer Vision
Background:
- Domain shift, a discrepancy between training and test data distributions, significantly degrades deep learning model performance in clinical settings.
- Effective deployment of deep learning models in healthcare requires methods robust to domain shift, especially in medical image segmentation.
- Unsupervised domain adaptation (UDA) offers a promising approach to address this challenge by leveraging unlabeled target domain data.
Purpose of the Study:
- To develop an unsupervised domain adaptation technique for robust cross-domain medical image segmentation.
- To propose a novel method, CyCMIS, that integrates diverse image translation and semantic consistency for improved generalizability.
- To address the limitations of one-to-one mapping by characterizing complex domain relationships as many-to-many mappings.
Main Methods:
- Cycle-consistent Cross-domain Medical Image Segmentation (CyCMIS) integrating online diverse image translation via disentangled representation learning.
- Characterizing cross-domain relationships using a many-to-many mapping approach.
- Implementing a novel diverse inter-domain semantic consistency loss and an intra-domain semantic consistency loss for regularization.
Main Results:
- CyCMIS demonstrated effectiveness in cross-domain medical image segmentation tasks.
- The proposed method showed robustness against domain shift without requiring target domain annotations.
- Comprehensive experiments on public datasets validated the efficacy of the CyCMIS approach.
Conclusions:
- The proposed CyCMIS method effectively addresses domain shift in medical image segmentation using unsupervised domain adaptation.
- Integrating diverse image translation and semantic consistency regularization enhances model generalizability across domains.
- CyCMIS offers a valuable solution for deploying deep learning models in clinical settings where domain shift is prevalent.
