Enhancing brain tumor segmentation in MRI images using the IC-net algorithm framework.

Area of Science:

• Medical Image Analysis
• Artificial Intelligence in Healthcare
• Neuro-oncology Imaging

Background:

• Accurate segmentation of brain tumors in MRI is critical for diagnosis and treatment planning.
• Existing methods face challenges due to the diverse nature and complexity of brain tumor images.
• Distinguishing tumorous regions from healthy brain tissue remains a significant hurdle in medical image analysis.

Purpose of the Study:

• To develop a novel semantic segmentation architecture, IC-Net (Inverted-C), for enhanced accuracy in brain tumor segmentation from MRI images.
• To improve the precision and robustness of separating tumorous regions from healthy tissues.
• To address the limitations of current segmentation techniques in handling diverse brain tumor characteristics.

Main Methods:

• Proposed IC-Net (Inverted-C) architecture, integrating Multi-Attention (MA) blocks, Feature Concatenation Networks (FCN), and Attention-blocks.
• MA-blocks aggregate multi-attention features for adaptability to various tumor sizes and shapes.
• Attention-blocks focus on critical image regions, while FCN-blocks capture diverse features for robustness.

Main Results:

• IC-Net demonstrated superior performance over U-Net and other contemporary segmentation techniques on the BraTS 2020 dataset.
• Achieved high metrics: Accuracy (99.65%), Loss (0.0159), Specificity (99.44%), Sensitivity (99.86%).
• Dice Similarity Coefficient (DSC) scores for core, whole, and enhancing tumors were 0.998717, 0.888930, and 0.866183, respectively.

Conclusions:

• IC-Net offers a significant advancement in brain tumor segmentation accuracy and efficiency.
• The architecture effectively handles the complexities and diversity of brain tumor imaging.
• IC-Net shows potential for improved clinical applications in neuro-oncology and medical image analysis.