Automated Brain Tumor MRI Segmentation Using ARU-Net with Residual-Attention Modules

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Accurate brain tumor segmentation in MRI is crucial for diagnosis and treatment planning.
• Existing methods struggle with the heterogeneity and complexity of tumor regions.

Purpose of the Study:

• To develop a robust, automated deep learning method for precise brain tumor segmentation.
• To enhance segmentation accuracy and generalization capabilities.

Main Methods:

• Proposed Attention Res-UNet (ARU-Net) architecture integrating residual connections, Adaptive Channel Attention (ACA), and Dimensional-space Triplet Attention (DTA).
• Pre-processed MRI images using CLAHE, denoising, and Linear Kuwahara filtering.
• Trained ARU-Net on the BTMRII dataset and compared against baseline U-Net, DenseNet121, and Xception models.

Main Results:

• ARU-Net achieved 98.3% accuracy, 98.1% DSC, 96.3% IoU, and a superior F1-score.
• Demonstrated significant improvements over baseline U-Net, particularly in segmenting heterogeneous tumor structures.
• Visualizations confirmed smoother boundaries and more precise tumor contours across all tumor classes.

Conclusions:

• ARU-Net offers a highly reliable and precise solution for automated brain tumor segmentation.
• The model's superior performance highlights its potential for clinical application.
• Contributes novel insights to medical image analysis and deep learning in healthcare.