Related Experiment Video
Updated: Jan 18, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
DUR-Net+: Semi-Supervised Abdominal CT Pheochromocytoma Segmentation via Dynamic Uncertainty Rectified and Prior
IEEE Journal of Biomedical and Health Informatics
|September 12, 2025
Summary
This study introduces a novel semi-supervised framework for segmenting pheochromocytoma (a rare adrenal tumor) using CT scans. The method enhances accuracy by dynamically rectifying uncertainty and leveraging prior knowledge, improving diagnosis and treatment planning.
Area of Science:
- Urological oncology
- Medical imaging analysis
- Artificial intelligence in healthcare
Background:
- Pheochromocytoma is a rare adrenal tumor requiring accurate segmentation from CT scans for diagnosis and treatment.
- Automated segmentation is challenging due to irregular tumor shapes, variable locations/sizes, and limited annotated data.
- Existing methods struggle with the inherent difficulties in pheochromocytoma segmentation.
Purpose of the Study:
- To develop a robust semi-supervised framework for automated pheochromocytoma segmentation.
- To address challenges posed by limited annotated data and complex tumor characteristics.
- To improve the accuracy and efficiency of pheochromocytoma diagnosis and treatment planning.
Main Methods:
- A semi-supervised segmentation model with a shared encoder and multiple decoders was designed, dynamically selecting pseudo labels.
- A dynamic uncertainty rectification mechanism was implemented to prioritize reliable predictions from sparse annotations.
- An Attentional Convolution Block (ACB) was integrated for enhanced feature extraction, and SAM-Med3D prior knowledge was incorporated.
- Pseudo labels were used to generate mask prompts, automating the SAM-Med3D workflow.
Main Results:
- The proposed framework demonstrated competitive performance in segmenting pheochromocytomas across two independent datasets.
- The dynamic uncertainty rectification and SAM-Med3D integration improved segmentation accuracy with limited labeled data.
- The Attentional Convolution Block effectively utilized global and local features for better tumor recognition.
Conclusions:
- The developed semi-supervised framework offers a promising solution for accurate pheochromocytoma segmentation.
- This approach mitigates the need for extensive manual annotation, streamlining the clinical workflow.
- The method holds potential for advancing the diagnosis and treatment of this rare adrenal tumor.
