Related Experiment Video
Updated: May 27, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.6K
An accurate and trustworthy deep learning approach for bladder tumor segmentation with uncertainty estimation
Jie Xu1, Haixin Wang2, Min Lu3
1School of Information Technology and Management, University of International Business and Economics, Beijing 100029, China.
Computer Methods and Programs in Biomedicine
|February 15, 2025
Summary
This study introduces a bladder tumor segmentation model that provides confidence scores for AI predictions, enhancing diagnostic reliability. The BSU model improves accuracy and efficiency for urologists in bladder cancer diagnosis.
Area of Science:
- Artificial Intelligence in Medicine
- Medical Imaging Analysis
- Oncology Diagnostics
Background:
- Deep learning models for bladder cancer diagnosis show promise but lack reliability assessment.
- Evaluating the trustworthiness of neural network predictions is crucial for clinical adoption.
Purpose of the Study:
- To develop a trustworthy AI-based tumor segmentation model for bladder cancer.
- To provide confidence information alongside segmentation predictions.
Main Methods:
- Proposed a novel bladder tumor segmentation with uncertainty estimation (BSU) model.
- Utilized Test Time Augmentation (TTA) and Test Time Dropout (TTD) to estimate aleatoric and epistemic uncertainty.
- Evaluated the model on internal and external cystoscopy datasets.
Main Results:
- BSU model achieved Dice coefficients of 0.766 (internal) and 0.848 (external).
- Attained high accuracy rates of 0.950 (internal) and 0.954 (external).
- Demonstrated superior performance compared to state-of-the-art methods, with clinical validation of uncertainty estimation's practical value.
Conclusions:
- The BSU model visualizes segmentation confidence, aiding urologists.
- Enhances precision and efficiency in clinical bladder cancer diagnosis.
- Offers a valuable tool for improving diagnostic workflows.
