Lightweight Vision Transformer-Based U-Net for Brain Tumor Segmentation from MRI

TL;DR

This paper proposes a lightweight Vision Transformer-based U-Net for brain tumor segmentation from MRI, achieving improved performance over the baseline U-Net. The model combines hierarchical feature extraction with global context modeling. It outperforms the baseline U-Net by 3.75% and 3.15% in mean Intersection over Union and Dice score, respectively.

Why It Matters

Accurate brain tumor segmentation is crucial for diagnosis, treatment planning, and surgical guidance. The proposed method's improved performance can lead to better patient outcomes. Its computational efficiency also makes it more accessible for clinical applications.

Key Findings

• Improved brain tumor segmentation performance
• Effective learning of local and global features
• Computational efficiency with 2.6 million trainable parameters