DFuse-Net: Disentangled feature fusion with uncertainty-aware learning for reliable multi-modal brain tumor...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Accurate brain tumor segmentation from multi-modal MRI is vital for clinical applications.
• Challenges include modality-specific noise, semantic inconsistency, and model uncertainty.
• Existing methods struggle to fully leverage complementary information across MRI modalities.

Purpose of the Study:

• To propose DFuse-Net, a novel network for reliable multi-modal brain tumor segmentation.
• To address challenges in fusing information from different MRI modalities.
• To improve the accuracy and reliability of brain tumor segmentation for clinical use.

Main Methods:

• DFuse-Net employs disentangled feature fusion and uncertainty-aware learning.
• It explicitly separates modality-shared and modality-specific representations.
• Specialized modules (DTFM, DSFM) integrate texture and semantic information; contrastive and consistency learning enhance features; Monte Carlo dropout estimates uncertainty.

Main Results:

• DFuse-Net demonstrated superior performance compared to state-of-the-art methods on BraTS datasets.
• The proposed fusion and uncertainty estimation strategies proved effective.
• Enhanced feature discriminability and structural coherence were achieved.

Conclusions:

• DFuse-Net offers a reliable approach for multi-modal brain tumor segmentation.
• The method shows significant potential for clinical diagnosis and treatment planning.
• Uncertainty estimation improves the robustness of the segmentation results.