VcaNet: Vision Transformer with fusion channel and spatial attention module for 3D brain tumor segmentation

Area of Science:

• Medical Image Analysis
• Artificial Intelligence in Medicine
• Neuroimaging

Background:

• Accurate brain tumor segmentation from MRI is vital but challenging due to tumor variability.
• Traditional Convolutional Neural Networks (CNNs) struggle with long-range dependencies in 3D medical data.

Purpose of the Study:

• To introduce VcaNet, a novel architecture for enhanced 3D brain tumor segmentation.
• To improve the capture of both local and global features for more accurate segmentation.

Main Methods:

• VcaNet integrates a Vision Transformer (ViT) with a Convolutional Block Attention Module (CBAM).
• The architecture features a 3D enhanced convolution (ENCO) encoder, a ViT and multi-scale feature fusion bottleneck, and a CBAM-enhanced decoder.
• Experiments were conducted on public BraTS Datasets.

Main Results:

• VcaNet demonstrated superior performance compared to existing models in 3D brain tumor segmentation.
• The model effectively handled complex spatial structures of brain tumors.
• VcaNet's 3D performance surpassed that of 2D models.

Conclusions:

• VcaNet offers a promising approach for improving 3D brain tumor segmentation accuracy.
• The integration of ViT and CBAM effectively captures essential local and global features.
• This work advances medical imaging analysis for neuro-oncology applications.