Related Experiment Video
Updated: Jul 9, 2025
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
48.1K
QMLS: quaternion mutual learning strategy for multi-modal brain tumor segmentation
Zhengnan Deng1, Guoheng Huang1, Xiaochen Yuan2
1School of Computer Science and Technology, Guangdong University of Technology, Guangzhou, 510006, People's Republic of China.
Physics in Medicine and Biology
|December 7, 2023
Summary
This study introduces a novel quaternion mutual learning strategy (QMLS) for brain tumor segmentation using multi-modal MRI. QMLS effectively leverages complementary information from different MRI modalities, improving segmentation accuracy with fewer parameters.
Area of Science:
- Medical imaging analysis
- Artificial intelligence in medicine
- Neuro-oncology imaging
Background:
- Magnetic resonance imaging (MRI) enables non-invasive, multi-modal brain tumor segmentation (MBTS).
- Convolutional neural networks (CNNs) show promise in MBTS, but limited data hinders full exploration of multi-modal MRI information.
- Existing MBTS models struggle to mine complementary information among different MRI modalities due to data scarcity.
Purpose of the Study:
- To propose a novel quaternion mutual learning strategy (QMLS) for enhanced multi-modal brain tumor segmentation.
- To address the challenge of limited data in MBTS by fully exploring complementary information across modalities.
- To develop a computationally efficient model with reduced parameters for clinical application.
Main Methods:
- Developed a quaternion mutual learning strategy (QMLS) comprising a voxel-wise lesion knowledge mutual learning (VLKML) mechanism and a quaternion multi-modal feature learning (QMFL) module.
- VLKML mechanism facilitates network convergence for robust data augmentation, expanding limited datasets.
- QMFL module utilizes quaternion-valued representations to capture inter-modal complementary information in the hypercomplex domain, reducing parameters by approximately 75%.
Main Results:
- QMLS demonstrated superior performance compared to existing popular methods on the BraTS 2020 and BraTS 2019 datasets.
- The proposed method achieved better segmentation results with significantly reduced computational cost.
- Experiments confirmed the effectiveness of the QMLS in leveraging multi-modal MRI data.
Conclusions:
- The proposed QMLS offers a novel and effective approach for multi-modal brain tumor segmentation.
- This strategy enhances the utilization of complementary information from multi-modal MRI, overcoming data limitations.
- The algorithm's improved performance and reduced parameter count facilitate the clinical application of automatic brain tumor segmentation.
