MamNet-PT: A Mamba-enhanced hybrid architecture with selective state-space modeling for uncertainty-aware brain tumor...

Area of Science:

• Medical Image Analysis
• Artificial Intelligence
• Neuro-oncology

Background:

• Brain tumor segmentation from MRI is challenging due to variable tumor size, diffuse boundaries, and data imbalance.
• Accurate segmentation is crucial for diagnosis, treatment planning, and outcome prediction.

Purpose of the Study:

• To develop MamNet-PT, a hybrid segmentation architecture for precise and efficient brain tumor segmentation.
• To address challenges of long-range dependency modeling, multi-resolution feature aggregation, and uncertainty estimation.

Main Methods:

• Integrated a selective state-space model (Mamba) into a U-Net for efficient long-range dependency modeling.
• Employed a pre-trained ResNet-50 encoder for improved feature robustness with limited data.
• Utilized a gated feature interaction mechanism for adaptive fusion of global and local features.
• Incorporated a multi-resolution pyramid fusion module and Monte Carlo Dropout for uncertainty estimation.

Main Results:

• Achieved a Dice score of 96.7% and Intersection over Union of 95.4% on the BraTS2020 benchmark.
• Outperformed existing CNN-Transformer and Mamba-based segmentation baselines.
• Ablation studies confirmed the effectiveness of individual components like selective state-space modeling and gated fusion.

Conclusions:

• MamNet-PT demonstrates a promising framework for accurate and efficient brain tumor segmentation.
• The hybrid architecture effectively handles challenges in tumor variability and boundary definition.
• Uncertainty estimation provides valuable confidence maps for clinical application and analysis.