Related Experiment Video
Updated: Apr 5, 2026
Hybrid µCT-FMT imaging and image analysis
Published on: June 4, 2015
User-preference alignment with uncertainty-aware interactive rectification for liver organ and tumor segmentation and
Guangyuan Zhao1, Yang Wang2, Chen Gong3
1Institute of Organ Transplantation, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology; Key Laboratory of Organ Transplantation, Ministry of Education, NHC Key Laboratory of Organ Transplantation; Key Laboratory of Organ Transplantation, Chinese Academy of Medical SciencesOrgan Transplantation Clinical Medical Research Center of Hubei Province, Wuhan, China.
We developed a new AI framework for accurate liver and tumor segmentation in CT scans. This tool reduces manual effort by letting clinicians select preferred segmentation options, improving efficiency and accuracy.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Accurate liver and tumor segmentation in CT images is crucial for cancer management.
- Manual segmentation is time-consuming and inconsistent.
- Existing automated methods struggle with clinical variability and image uncertainties.
Purpose of the Study:
- To introduce a novel framework, UAIR (User-Preference Alignment with Uncertainty-Aware Interactive Rectification), for efficient and adaptive liver and tumor segmentation.
- To reduce the human interaction cost in segmentation tasks.
- To improve the accuracy and adaptability of automated segmentation models.
Main Methods:
- UAIR quantifies model uncertainty to generate a small set of diverse segmentation candidates.
- Clinicians interactively select the most suitable candidate, guiding iterative refinement.
- This selection-based approach avoids laborious pixel-level corrections.
Main Results:
- UAIR achieved superior accuracy (DSC 0.776) compared to manual positional prompting (DSC 0.685) on a multi-center CT dataset.
- The framework demonstrated reduced prompting efforts and improved segmentation efficiency.
- Validation on a large-scale dataset confirmed the framework's robustness.
Conclusions:
- UAIR offers a clinically viable solution for rapid and robust liver and tumor segmentation.
- The framework integrates seamless human guidance, aligning with specific clinical preferences.
- UAIR enables efficient quantitative analysis for cancer diagnosis and treatment planning.
