Segmentation of the hippocampus by transferring algorithmic knowledge for large cohort processing.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• The hippocampus is crucial for human brain function and behavior.
• Accurate segmentation of the hippocampus is vital for studying neuropsychiatric disorders.
• Existing automated methods like FreeSurfer have limitations.

Purpose of the Study:

• To develop and validate a novel deep learning-based method for segmenting the bilateral hippocampus.
• To improve the accuracy and efficiency of hippocampal segmentation in large human cohorts.

Main Methods:

• Utilized a deep convolutional neural network (ConvNet) with a learned appearance model.
• Trained the model on a diverse dataset from multiple cohorts, incorporating FreeSurfer outputs and synthetic data.
• Employed a robust data augmentation scheme to enhance model generalizability.

Main Results:

• The proposed deep learning method achieves robust and fast hippocampal segmentation (under 30 seconds per subject).
• Demonstrated strong qualitative and quantitative performance in cohort-wide comparisons against FreeSurfer.
• The method effectively encodes and potentially improves upon existing anatomical knowledge.

Conclusions:

• Deep neural networks offer a powerful approach for anatomical segmentation in neuroimaging.
• This novel method provides an accurate, efficient, and accessible tool for hippocampal analysis in neuroscience research.
• The trained model is publicly available to facilitate further research.