Basic Science and Pathogenesis.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Biology

Background:

• Neurodegenerative diseases cause brain atrophy, quantifiable by MRI but at high cost.
• Three-dimensional (3D) modeling and machine learning (ML) offer potential for cost-effective visualization and quantitative analysis of brain atrophy.
• This study explores ML-based segmentation of 3D brain models for neuroanatomical analysis.

Purpose of the Study:

• To develop and evaluate ML techniques for segmenting 3D brain models to identify neuroanatomical sub-regions.
• To assess the potential of these methods for quantitative measurement of brain atrophy.
• To provide a cost-effective alternative to traditional neuroimaging for atrophy assessment.

Main Methods:

• Utilized Python and Blender for neuroanatomical analysis and 3D model manipulation.
• Generated 3D models of postmortem brain tissue using structured-light scanning and processed them in Python.
• Manually segmented models in Blender, exported data for a Python-based ML pipeline, and evaluated performance using standard metrics.

Main Results:

• Random forest and radius neighbor classifiers in scikit-learn showed high accuracy, precision, and recall.
• A custom Tensorflow model demonstrated improved accuracy but lower precision and recall.
• ML pipelines achieved 60-90% accuracy and precision in segmenting 3D brain models using multi-label classification.

Conclusions:

• ML techniques show promise for identifying neuroanatomical sub-regions and measuring atrophy from 3D brain models.
• Further validation with other neuroimaging modalities is crucial for consistency.
• Integrating 2D images with 3D models could yield MRI-comparable atrophy measurements, benefiting neuropathology and biomedical research.