Bladder Wall and Tumor Segmentation Using a New Framework
Author Information
Author(s): Wei Jie, Zheng Yao, Huang Dong, Liu Yang, Xu Xiaopan, Lu Hongbing, Ahn Kwang Woo, Baselli Giuseppe
Primary Institution: School of Biomedical Engineering, Air Force Medical University
Hypothesis
Can a semi-supervised multi-region framework improve the segmentation of bladder wall and tumors in MRI?
Conclusion
The proposed method achieves promising results in segmenting both the bladder wall and tumor, enhancing clinical decision-making.
Supporting Evidence
- The average Dice Similarity Coefficients (DSCs) for the bladder wall and tumor were 0.8351 and 0.9175, respectively.
- Extensive experimental validation demonstrated promising results in segmentation accuracy.
- The proposed method minimizes reliance on annotated data while improving segmentation performance.
Takeaway
This study created a new way to help doctors see bladder tumors and walls better in MRI scans, which can help with treatment.
Methodology
The study used a semi-supervised multi-region framework with wall-enhanced self-supervised pre-training to improve segmentation accuracy.
Potential Biases
Some data had poor annotation quality, affecting the model's learning and performance.
Limitations
The model construction currently ignores clinical prior knowledge of bladder cancer and relies on the quality of annotations.
Participant Demographics
The study involved 52 subjects diagnosed with bladder cancer, including cases of non-muscle-invasive and muscle-invasive bladder cancer.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website