Deep Feature Fusion via Graph Convolutional Network for Intracranial Artery Labeling

Area of Science:

• Medical imaging analysis
• Artificial intelligence in medicine
• Neuroscience and neuroimaging

Background:

• Intracranial arteries are vital for brain oxygenation.
• Accurate labeling of cerebral arteries is crucial for clinical applications and disease diagnosis.
• Current machine learning methods face challenges due to the complexity and variability of intracranial arteries.

Purpose of the Study:

• To develop a novel graph convolutional neural network (GCNN) model for automated cerebral artery labeling.
• To improve the accuracy and efficiency of intracranial artery anatomical labeling.
• To enhance the representation capability and performance of artery labeling models through deep feature fusion.

Main Methods:

• Implementation of stacked graph convolutions within an encoder-core-decoder architecture.
• Utilizing deep feature fusion to aggregate intermediate features from different hierarchical levels.
• Developing a GCNN model for extracting high-level representations from graph nodes and their neighbors.

Main Results:

• The proposed GCNN model with deep feature fusion demonstrated superior performance compared to existing baseline methods.
• Extensive experiments on public datasets validated the effectiveness of the novel approach.
• The model achieved a clear margin of improvement in cerebral artery labeling accuracy.

Conclusions:

• The novel GCNN approach with deep feature fusion offers a significant advancement in intracranial artery labeling.
• Accurate artery labeling facilitates medical research and aids in the diagnosis of neurological diseases.
• The method's ability to effectively extract graph information enhances prediction accuracy over current techniques.