Related Experiment Video
Updated: Aug 31, 2025
04:25
Author Spotlight: Bridging Gaps in Anatomy and Establishing a Foundation for Algorithmic Studies
Published on: December 15, 2023
2.8K
A Weakly Supervised Brain Tumor Segmentation Strategy Based on Multi-level Sub-category and Membership Matrix.
Zi-Wei Li1, Shi-Bin Xuan1,2, Li Wang1
1School of Artificial Intelligence, Guangxi Minzu University, Daxue East Road 188, Nanning, China.
Current Medical Imaging
|August 22, 2022
Summary
This study introduces a new weakly supervised learning strategy for brain tumor segmentation using multi-level sub-categories and membership matrices to improve class activation mapping (CAM) accuracy.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Class activation mapping (CAM) is a common method for weakly supervised semantic segmentation.
- However, CAM struggles to accurately delineate brain tumor boundaries in medical images.
Purpose of the Study:
- To enhance the performance of CAM for brain tumor segmentation.
- To develop a weakly supervised learning strategy that improves boundary fitting.
Main Methods:
- Implemented a multi-level sub-category strategy for intensive data classification, enabling deeper feature learning.
- Integrated fuzzy clustering and a membership matrix into the model, combined with CAM to create a novel loss function.
Main Results:
- The proposed method significantly improved CAM performance on the BraTS2019 brain tumor dataset.
- Achieved a 17.1% improvement over the baseline using the dice similarity coefficient and a 9% improvement over recent studies.
Conclusions:
- The developed methods train networks using image-level labels to better mine target boundary information.
- This approach enables CAM to more accurately fit brain tumor borders in medical images.
