Deep Learning-Driven Transmission Electron Microscopy Analysis of Murine Optic Nerve Myelinated Axons.

Objective: To develop and validate a deep learning-based method for automated quantification of retinal ganglion cell axons in transmission electron microscopy (TEM) images, addressing the time-consuming and subjective nature of manual segmentation and quantification. Design: Development and validation of a deep learning-based segmentation and quantification pipeline for TEM images of murine optic nerves. Subjects: Three hundred sixty-eight optic nerve TEM images from 23 C57BL/6J and DBA/1J mice (4-9 months old) under different experimental conditions were used to develop and validate this algorithm. Methods: Murine optic nerves were dissected and imaged using TEM at ×3000 magnification. A deep learning model based on the nnU-Net architecture was trained to segment the circumferences of inner axons and outer myelinated fibers. Postprocessing operations, including morphological gap closing and removal of incomplete axons, were then performed. Quantitative measures, such as axon count, diameter, area, G-ratio, and myelin thickness, were derived from the segmentation masks. Main Outcome Measures: Segmentation performance metrics (precision, recall, F1-score) and morphometric measures (axon count, axon diameter, G-ratio, myelinated axon area, and myelin thickness). Results: The nnU-Net model achieved an inner axon F1-score of 0.771 and an outer myelinated fiber F1-score of 0.697. Quantitative measures derived from model predictions showed high concordance (R Conclusions: Our deep learning-based pipeline provides reliable and robust quantification of myelinated axons in murine TEM images, significantly reducing manual effort, processing time, and subjectivity. Financial Disclosures: The author has no/the authors have no proprietary or commercial interest in any materials discussed in this article.