Axon and Myelin Sheath Segmentation in Electron Microscopy Images using Meta Learning.

Area of Science:

• Neuroscience
• Biomedical Imaging
• Computational Biology

Background:

• Neurological diseases alter myelinated axon morphology.
• Accurate quantification of these changes is vital for disease characterization and treatment monitoring.
• Electron microscopy (EM) images are essential for detailed structural analysis.

Purpose of the Study:

• To develop a robust pipeline for segmenting axons and myelin sheaths in EM images.
• To enable the computation of novel biomarkers for hypoglossal nerve degeneration and regeneration.
• To address challenges posed by morphological variations and limited annotated data.

Main Methods:

• A meta-learning based training strategy was employed.
• A U-net like encoder-decoder deep neural network architecture was utilized.
• The pipeline was trained on EM images at 500X and 1200X magnifications.

Main Results:

• The proposed pipeline demonstrated improved segmentation performance.
• Segmentation accuracy increased by 5% to 7% compared to standard deep learning methods.
• The model was successfully tested on images with varying magnifications (250X and 2500X).

Conclusions:

• The meta-learning pipeline offers a robust solution for axon and myelin sheath segmentation in EM images.
• This method enhances the analysis of neurodegenerative and neuroregenerative processes.
• The approach shows promise for developing new EM-based biomarkers for neurological conditions.