Related Experiment Video
Updated: Jun 9, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Source free domain adaptation for kidney and tumor image segmentation with wavelet style mining
Yuwei Yin1,2, Zhixian Tang1, Zheng Huang2
1Department of Nephrology, Jinshan District Central Hospital affiliated to Shanghai University of Medicine & Health Sciences, The College of Medical Technology, Shanghai University of Medicine & Health Sciences, Shanghai, People's Republic of China.
This study introduces a new unsupervised domain adaptation framework for kidney cancer segmentation using CT images. The method improves segmentation accuracy and generalization, addressing limited labeled medical data challenges.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Kidney cancer diagnosis and treatment rely on accurate segmentation of tumors from CT images.
- Limited availability of labeled medical imaging data hinders the development of intelligent diagnostic systems.
- Unsupervised domain adaptation (UDA) offers a potential solution to leverage unlabeled data for improved segmentation.
Purpose of the Study:
- To develop a novel UDA framework for kidney and tumor CT image segmentation.
- To overcome the scarcity of labeled medical data in kidney cancer imaging.
- To enhance the accuracy and generalization capabilities of automated segmentation models.
Main Methods:
- A two-phase UDA framework involving a generation and an adaptation phase.
- Wavelet-based style mining generator for creating class-specific, source-like images to aid domain alignment.
- Contrastive domain extraction and compact-aware domain consistency modules for feature and output level adaptation using data augmentation.
Main Results:
- The proposed UDA framework demonstrated superior performance in kidney and tumor segmentation tasks.
- Achieved higher accuracy and improved generalization capabilities compared to existing state-of-the-art UDA methods.
- Validated the framework's effectiveness in addressing domain shift challenges in medical image segmentation.
Conclusions:
- The novel UDA framework offers significant advantages for kidney and tumor segmentation in CT images.
- The approach effectively mitigates the impact of limited labeled data, paving the way for more robust AI in medical diagnostics.
- This work highlights the potential of advanced UDA techniques for improving automated analysis of radiological scans.
