Related Experiment Video
Updated: Sep 23, 2025
14:15
Multicolor 3D Printing of Complex Intracranial Tumors in Neurosurgery
Published on: January 11, 2020
7.2K
D2-Net: Dual Disentanglement Network for Brain Tumor Segmentation With Missing Modalities
IEEE Transactions on Medical Imaging
|May 16, 2022
Summary
This study introduces a Dual Disentanglement Network (D²-Net) to improve brain tumor segmentation using multi-modal Magnetic Resonance Imaging (MRI), even when some data is missing. The novel approach effectively handles missing modalities by explicitly learning correlations between imaging data and tumor regions.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Multi-modal Magnetic Resonance Imaging (MRI) is vital for brain tumor segmentation, aiding diagnosis and prognosis.
- Missing MRI data is a common challenge in clinical settings, hindering the performance of existing segmentation methods.
- Current methods often fail to explicitly capture correlations between modalities and tumor regions when data is incomplete.
Purpose of the Study:
- To develop a novel deep learning framework for robust brain tumor segmentation with missing multi-modal MRI data.
- To explicitly exploit correlations among modality-specific information and tumor-specific knowledge for improved segmentation accuracy.
- To address the limitations of current methods that rely on complete data and ignore inter-modal and inter-region relationships.
Main Methods:
- Propose a Dual Disentanglement Network (D²-Net) comprising a Modality Disentanglement Stage (MD-Stage) and a Tumor-Region Disentanglement Stage (TD-Stage).
- Implement a spatial-frequency joint contrastive learning scheme in MD-Stage to decouple modality-specific information.
- Utilize an affinity-guided dense tumor-region knowledge distillation mechanism in TD-Stage to align teacher and student network features for decomposing tumor representations.
Main Results:
- The D²-Net demonstrates superior performance in brain tumor segmentation under missing modality conditions compared to state-of-the-art methods.
- The framework effectively learns sufficient information for segmentation despite the absence of certain MRI modalities.
- Experiments on the BraTS-2018 database validate the proposed method's effectiveness and robustness.
Conclusions:
- The proposed D²-Net offers a significant advancement in brain tumor segmentation for scenarios with missing multi-modal MRI data.
- Explicitly modeling correlations between modalities and tumor regions is crucial for handling incomplete data.
- The method provides a promising solution for improving clinical diagnosis and prognosis through accurate automated segmentation.
