Dynamic Focus on Tumor Boundaries: A Lightweight U-Net for MRI Brain Tumor Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Accurate brain tumor segmentation in MRI is crucial for diagnosis and treatment planning.
• Traditional models like U-Net face challenges with complex tumor boundaries and low contrast.
• Limitations impact clinical outcome accuracy due to inconsistent identification of critical regions.

Purpose of the Study:

• To propose a novel U-Net architecture modification integrating a spatial attention mechanism.
• To enhance the model's ability to delineate fine tumor boundaries and improve segmentation precision.
• To develop a robust tool for automated brain tumor segmentation suitable for real-time clinical applications.

Main Methods:

• Modification of the U-Net architecture with an integrated spatial attention mechanism.
• Evaluation on the Figshare dataset containing annotated MRI images of meningioma, glioma, and pituitary tumors.
• Performance comparison against established models like V-Net, DeepLab V3+, and nnU-Net.

Main Results:

• Achieved a Dice similarity coefficient (DSC) of 0.93, recall of 0.95, and AUC of 0.94.
• Outperformed V-Net, DeepLab V3+, and nnU-Net in segmentation accuracy.
• Demonstrated effectiveness in handling low-contrast boundaries, small tumor regions, and overlapping tumors.

Conclusions:

• The proposed spatial attention mechanism significantly enhances U-Net for brain tumor segmentation.
• The model offers improved precision and robustness for clinical applications.
• This approach shows potential for advancing automated diagnostic tools in medical imaging.