A Macaque Brain Extraction Model Based on U-Net Combined with Residual Structure.

Area of Science:

• Neuroimaging
• Computer Vision
• Primate Neuroscience

Background:

• Accurate brain tissue extraction is crucial for neuroimaging research.
• Existing tools struggle with macaque brain MRI due to anatomical differences.
• Limited macaque MRI data hinders deep learning model training and generalization.

Purpose of the Study:

• To develop an improved method for macaque brain MRI extraction.
• To address limitations of insufficient training data and poor model generalization.
• To enhance the accuracy and efficiency of brain tissue segmentation in nonhuman primates.

Main Methods:

• Utilized a transfer learning strategy combining human brain MRI data for pre-training.
• Introduced a residual network structure within a U-Net model, creating the ResTLU-Net.
• Trained and validated the model on diverse macaque brain MRI datasets from multiple research sites.

Main Results:

• The ResTLU-Net model achieved high accuracy in macaque brain MRI extraction, with a mean Dice score of 95.81%.
• The method demonstrated robust generalization across different medical centers' data without requiring denoising or correction.
• Extraction time was efficient, approximately 30-60 seconds per task on an NVIDIA 1660S GPU.

Conclusions:

• The proposed transfer learning strategy and ResTLU-Net model significantly improve macaque brain MRI extraction.
• This approach offers a reliable and efficient solution for nonhuman primate neuroimaging research.
• The method enhances the ability to analyze diverse macaque brain MRI datasets.