MSFR-Net: Multi-modality and single-modality feature recalibration network for brain tumor segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Accurate brain tumor segmentation from multi-modal MRI is crucial for treatment planning.
• Existing methods often overlook the distinct correlations between single MRI modalities and specific tumor sub-regions.
• Physicians utilize modality-specific characteristics (e.g., T2 for edema, T1-contrast for enhancing tumor) for manual labeling.

Purpose of the Study:

• To develop an automated brain tumor segmentation method that integrates both multi-modality fusion and single-modality feature characteristics.
• To improve segmentation performance by explicitly modeling the relationship between individual MRI sequences and tumor sub-components.

Main Methods:

• Proposed a Multi-Modality and Single-Modality Feature Recalibration Network (MSFR-Net).
• Employed independent pathways for multi-modality and single-modality information processing.
• Introduced a Dual Recalibration Module (DRM) to unify features from both pathways into a common space.

Main Results:

• Achieved competitive and superior performance compared to state-of-the-art methods on BraTS 2015 and BraTS 2018 datasets.
• On BraTS 2018, obtained Dice coefficients of 0.86 and Hausdorff distance of 4.82.
• On BraTS 2015, achieved Dice coefficients of 0.80, positive predictive value of 0.76, and sensitivity of 0.78.

Conclusions:

• The proposed method effectively combines manual labeling insights with automated segmentation by incorporating single-modality correlations.
• MSFR-Net demonstrates improved brain tumor segmentation performance, showing potential for clinical application.
• The codebase for MSFR-Net is publicly available for research and development.