Related Experiment Video
Updated: Jul 23, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Joint learning-based feature reconstruction and enhanced network for incomplete multi-modal brain tumor segmentation.
Yueqin Diao1, Fan Li1, Zhiyuan Li1
1Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, China; Yunnan Key Laboratory of Artificial Intelligence, Kunming 650500, China.
Computers in Biology and Medicine
|July 14, 2023
Summary
This study introduces a novel joint learning method for brain tumor segmentation using incomplete multimodal Magnetic Resonance Imaging (MRI) data. The approach reconstructs missing modalities, enhancing segmentation accuracy even with absent data.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuro-oncology
Background:
- Multimodal Magnetic Resonance Imaging (MRI) enhances brain tumor segmentation.
- Missing MRI modalities in clinical diagnosis impair segmentation performance.
- Existing methods use modal fusion for shared feature representations.
Purpose of the Study:
- To develop a robust brain tumor segmentation method for incomplete multimodal MRI data.
- To address the challenge of missing modalities in deep learning segmentation techniques.
- To improve segmentation accuracy and model robustness under various missing modality scenarios.
Main Methods:
- A joint learning framework combining feature reconstruction and enhancement for incomplete multimodal brain tumor segmentation.
- An information learning mechanism to transfer knowledge from complete to single modalities.
- A feature reconstruction module to recover missing modality information from available data.
- A feature enhancement mechanism to leverage reconstructed missing modality information.
Main Results:
- The proposed model achieved Dice similarity scores of 86.28% (whole tumor), 77.02% (tumor core), and 59.64% (enhanced tumor) on the BraTS2018 dataset.
- Demonstrated superior performance compared to state-of-the-art methods in missing modality situations.
- The method effectively obtains comprehensive brain tumor information despite missing MRI data.
Conclusions:
- The proposed joint learning method significantly enhances brain tumor segmentation accuracy with incomplete multimodal MRI.
- The feature reconstruction and enhancement approach improves model robustness and handles missing data effectively.
- This framework offers a promising solution for real-world clinical scenarios with incomplete neuroimaging data.
