An Open-Source, User-Friendly Machine-Learning Method for Automated Segmentation and Analysis of Peripheral Nerve Cross-Sections.

Background: Quantitative neuromorphometric analysis of the peripheral nerve is paramount to nerve regeneration research. However, this technique relies upon accurate segmentation and determination of myelin and axonal area. Manual histologic analysis methods are time-consuming and subject to error and bias. The authors demonstrate and validate a user-friendly method relying on open-source machine-learning software and requiring no coding knowledge.

Methods: Nineteen rat facial nerve segments were fixed, osmicated, embedded in epoxy, sectioned, and stained with toluidine blue. Whole nerve cross-sections were scanned at 20×. Images were preprocessed in ImageJ to measure fascicular area and remove background. Images were imported into Ilastik, and the pixel classification module was used to segment myelin and axons. A novel CellProfiler pipeline was used with Otsu 2-threshold processing to segment and quantify individual axon and myelin objects from segmentations. Axon counts and g-ratio of nerve samples were acquired manually and using this technique to compare between methods.

Results: Analysis of an entire nerve cross-section can be completed with this method in less than 5 minutes. Bland-Altman plot and intraclass correlation coefficient demonstrated reliability between axon counts performed manually and with this protocol. There was no significant difference in accuracy in determining g-ratio compared with manual techniques ( P < 0.05).

Conclusions: The protocol presented here demonstrates a novel and accurate means of segmenting and analyzing histologic nerve cross-sections. This method decreases both user time and potential bias, enables analysis of an entire nerve versus random sampling, and uses only open-source software freely accessible to researchers.