New Algorithm for Medical Image Segmentation
Author Information
Author(s): Lin Jian, Yu Xiaomin, Wang Zhengxian, Ma Chaoqiong
Primary Institution: Sichuan Academy of Medical Science and Sichuan Provincial People’s Hospital, Chengdu, China
Hypothesis
Can a source-free domain transfer algorithm improve medical image segmentation accuracy without accessing source data?
Conclusion
The SFDT-RSS algorithm significantly improves segmentation accuracy in medical images by addressing style shifts and error diffusion.
Supporting Evidence
- SFDT-RSS achieved segmentation accuracy improvements of 2.83%, 2.64%, 3.21%, 3.01%, and 3.32% on five datasets.
- The algorithm enhances generalization capability and reduces reliance on a single image style.
- Experimental results demonstrate significant improvements in Dice coefficients for unsupervised cross-domain segmentation tasks.
Takeaway
This study created a new way to help computers better understand medical images without needing old examples, making it easier to spot things like tumors.
Methodology
The study developed the SFDT-RSS algorithm, which uses a pre-training strategy and adaptive confidence regulation to enhance medical image segmentation.
Limitations
The algorithm's performance may be limited by the inability to access source domain data during training.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website