Automated pipeline for nerve fiber selection and g-ratio calculation in optical microscopy: exploring staining protocol variations.

G-ratio is crucial for understanding the nervous system's health and function as it measures the relative myelin thickness around an axon. However, manual measurement is biased and variable, emphasizing the need for an automated and standardized technique. Although deep learning holds promise, current implementations lack clinical relevance and generalizability. This study aimed to develop an automated pipeline for selecting nerve fibers and calculating relevant g-ratio using quality parameters in optical microscopy. Histological sections from the sciatic nerves of 16 female mice were prepared and stained with either p-phenylenediamine (PPD) or toluidine blue (TB). A custom UNet model was trained on a mix of both types of staining to segment the sections based on 7,694 manually delineated nerve fibers. Post-processing excluded non-relevant nerves. Axon diameter, myelin thickness, and g-ratio were computed from the segmentation results and its reliability was assessed using the intraclass correlation coefficient (ICC). Validation was performed on adjacent cuts of the same nerve. Then, morphometrical analyses of both staining techniques were performed. High agreement with the ground truth was shown by the model, with dice scores of 0.86 (axon) and 0.80 (myelin) and pixel-wise accuracy of 0.98 (axon) and 0.94 (myelin). Good inter-device reliability was observed with ICC at 0.87 (g-ratio) and 0.83 (myelin thickness), and an excellent ICC of 0.99 for axon diameter. Although axon diameter significantly differed from the ground truth (