HC-Net: A hybrid convolutional network for non-human primate brain extraction.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Primate Neuroscience

Background:

• Brain extraction (skull stripping) is crucial for analyzing brain magnetic resonance imaging (MRI) data.
• Existing methods struggle with non-human primate brains, particularly macaque MRI data, due to small sample sizes and thick-slice scanning.
• Traditional deep convolutional neural networks (DCNNs) often yield suboptimal results for macaque brain extraction.

Purpose of the Study:

• To develop an advanced deep learning model for accurate and efficient brain extraction in macaque MRI data.
• To address the limitations of current methods when applied to non-human primate neuroimaging.
• To propose a novel hybrid convolutional neural network (HC-Net) for improved macaque brain MRI analysis.

Main Methods:

• Proposed a symmetrical, end-to-end trainable hybrid convolutional neural network (HC-Net).
• Integrated 3D convolutions using three consecutive slices from three axes to leverage spatial information between adjacent slices.
• Employed a hybrid architecture combining 3D and 2D convolutions to balance spatial feature capture and prevent overfitting in small datasets.

Main Results:

• HC-Net demonstrated superior performance in inference time, averaging approximately 13 seconds per volume.
• Achieved high accuracy with a mean Dice coefficient of 95.46% on macaque brain data from diverse sources.
• The model exhibited good generalization and stability across different brain extraction scenarios.

Conclusions:

• HC-Net effectively overcomes the challenges of brain extraction in macaque MRI data.
• The hybrid approach balances computational efficiency and accuracy, outperforming traditional DCNNs.
• HC-Net offers a robust and reliable tool for non-human primate neuroimaging analysis.