Attention-guided multi-scale context aggregation network for multi-modal brain glioma segmentation.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Neuro-oncology

Background:

• Accurate brain glioma segmentation is crucial for diagnosis, surgical planning, and treatment evaluation.
• Manual slice-by-slice segmentation is time-consuming and prone to variability.
• Existing convolutional neural network (CNN) models show potential but require further performance improvements.

Purpose of the Study:

• To address limitations in current brain glioma segmentation methods.
• To propose an attention-guided multi-scale context aggregation network (AMCA-Net).
• To achieve accurate segmentation of brain gliomas in multi-modal MRI images.

Main Methods:

• AMCA-Net extracts and fuses multi-scale features using a self-attention mechanism.
• Global Context Information Guidance (GCIG) and Multi-Scale Fusion (MSF) modules aggregate contextual features.
• Channel Attention (CA) and Multi-Resolution Adaptation (MRA) modules enhance feature relevance and combine predictions.

Main Results:

• Evaluated on BraTS2018 and BraTS2019 datasets.
• AMCA-Net demonstrated comparable or superior performance to state-of-the-art models.
• Achieved high Dice scores and low Hausdorff 95 values for Whole Tumor (WT), Tumor Core (TC), and Enhancing Tumor (ET) regions.

Conclusions:

• AMCA-Net shows strong performance in segmenting brain glioma subregions.
• The model holds significant potential for clinical application in neuro-oncology.
• Future work will explore AMCA-Net for other segmentation tasks.