Related Experiment Video
Updated: May 9, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.6K
Structural uncertainty estimation for medical image segmentation
Bing Yang1, Xiaoqing Zhang1, Huihong Zhang1
1Department of Computer Science and Engineering, Southern University of Science and Technology, Shenzhen, 518055, China.
Medical Image Analysis
|May 1, 2025
Summary
This study introduces SU-ASM, a new method for precise medical image segmentation and uncertainty estimation. It uses structural information to improve accuracy and reduce errors in diagnostic assistance.
Area of Science:
- Medical Imaging
- Computer Vision
- Machine Learning
Background:
- Precise segmentation and uncertainty estimation are vital for medical diagnostic assistance, aiding error identification and correction.
- Current pixel-wise uncertainty methods overlook global context and cause attention interference, leading to inaccuracies and confusion.
- Existing approaches struggle with comprehensive analysis, necessitating improved methods for reliable medical image interpretation.
Purpose of the Study:
- To propose a novel structural uncertainty estimation method, SU-ASM, integrating global shape information for enhanced segmentation and uncertainty estimation.
- To address limitations of pixel-wise methods by incorporating global context and reducing attention interference.
- To improve the accuracy and reliability of medical diagnostic assistance through advanced segmentation and uncertainty quantification.
Main Methods:
- Developed SU-ASM, a method combining Convolutional Neural Networks (CNN) and Active Shape Models (ASM).
- Incorporated multi-task learning for improved ASM initialization and shape optimization.
- Utilized Combined Boundary Probability (CBP) and Key Landmark Template Matching (KLTM) for enhanced boundary reliability and shape template selection.
Main Results:
- SU-ASM demonstrated superior performance in segmentation and uncertainty estimation across cardiac ultrasound, ciliary muscle, and chest X-ray datasets.
- The method effectively incorporates global shape information, overcoming limitations of pixel-wise approaches.
- Validated SU-ASM's efficacy on diverse medical imaging datasets, confirming its robustness.
Conclusions:
- SU-ASM offers a significant advancement in precise medical image segmentation and uncertainty estimation.
- The structural uncertainty approach improves diagnostic assistance by providing more reliable segmentation and error identification.
- SU-ASM outperforms existing methods, paving the way for more accurate and dependable medical diagnostic tools.
