CAGs-Net: A Novel Adjacent-Context Network With Channel Attention Gate for 3D Brain Tumor Image Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Accurate brain tumor segmentation is critical for clinical decisions but challenging due to lesion characteristics like small volume, diverse morphology, and unclear MRI boundaries.
• Traditional methods often overlook inter-layer semantic correlations and lack explicit feature refinement control in attention mechanisms.

Purpose of the Study:

• To introduce CAGs-Net, a novel network designed to overcome limitations in automated brain tumor segmentation.
• To enhance feature refinement by integrating adjacent-layer context and channel attention mechanisms.

Main Methods:

• CAGs-Net progressively builds semantic dependencies between UNet hierarchy layers for integrated local and global context.
• Channel attention gates fuse shallow appearance and deep semantic features, refining voxel responses via channel-wise relationships.
• A hybrid loss function (generalized dice loss + binary cross-entropy) addresses class imbalance in lesion segmentation.

Main Results:

• CAGs-Net demonstrates superior performance compared to traditional UNet-based methods.
• Experimental results confirm enhanced segmentation accuracy for brain tumors using the proposed CAGs-Net architecture.
• The integration of adjacent-context modeling and channel attention gates proves effective for feature refinement.

Conclusions:

• CAGs-Net offers a significant advancement in automated brain tumor segmentation.
• The network's unique combination of adjacent-context modeling and channel attention gates improves segmentation accuracy.
• This approach holds promise for improving clinical decision-making in neuro-oncology.