A joint parcellation and boundary network with multi-rate-shared dilated graph attention for cortical surface...

Area of Science:

• Neuroimaging and Computational Neuroscience
• Medical Image Analysis
• Machine Learning for Healthcare

Background:

• Cortical surface parcellation is vital for understanding neurological diseases, but current methods struggle with learning boundary patterns.
• Accurate segmentation into anatomically and functionally significant regions is essential for clinical diagnosis and treatment.

Purpose of the Study:

• To develop an improved method for cortical surface parcellation that addresses limitations in boundary learning.
• To enhance the effectiveness and accuracy of segmenting the human brain's cortical surface.

Main Methods:

• Proposed a joint parcellation and boundary network (JPBNet) integrating boundary learning into the parcellation process.
• Developed a multi-rate-shared dilated graph attention (MDGA) module, extending dilated convolution to graph data.
• Incorporated boundary and parcellation enhancement modules within each layer, supervised by graph attention mechanisms.

Main Results:

• The proposed JPBNet demonstrated superior performance compared to existing methods on a public dataset.
• The method effectively captures informative features for both boundary detection and parcellation tasks.
• Experimental results show strong performance in identifying boundaries crucial for cortical surface parcellation.

Conclusions:

• The JPBNet significantly advances cortical surface parcellation by effectively learning and utilizing boundary information.
• This approach offers a more robust and accurate method for segmenting brain regions, aiding neurological disease research and diagnosis.