Neuron collinearity differentiates human hippocampal subregions: a validated deep learning approach.

Area of Science:

• Neuroscience
• Computational Biology
• Neuroanatomy

Background:

• The hippocampus is crucial for memory and navigation but vulnerable to neurodegeneration.
• Existing methods for hippocampal subregion parcellation using cytoarchitecture are limited.
• Pyramidal neuron orientation and collinearity are potential biomarkers for neurological diseases like schizophrenia.

Purpose of the Study:

• To develop a high-throughput deep learning method for automated pyramidal neuron orientation and collinearity extraction in hippocampal subregions.
• To quantitatively assess and compare pyramidal neuron collinearity across different hippocampal subregions.
• To establish neuron collinearity as a novel parameter for hippocampal subregion segmentation and disease investigation.

Main Methods:

• Utilized a deep learning approach based on the Cellpose algorithm for automated cellular segmentation.
• Quantified pyramidal neuron orientation and collinearity for over 479,000 neurons across 168 hippocampal partitions.
• Corrected orientation data for hippocampal curvature and validated deep learning results with manual assessment.

Main Results:

• Demonstrated significant differences in pyramidal neuron collinearity among hippocampal subregions (P < 0.001).
• Identified Cornu Ammonis 3 as the most collinear subregion, followed by CA2, CA1, medial/uncal regions, and subiculum.
• Validated the deep learning method's accuracy through manual orientation assessment.

Conclusions:

• Pyramidal neuron collinearity serves as a quantitative metric for hippocampal subregion segmentation, effectively differentiating CA2 and CA3.
• The novel deep learning approach enables large-scale analyses of hippocampal architecture.
• This methodology provides a foundation for investigating mental illnesses at the cellular level.