Rapid and Precise Semi-Automatic Axon Quantification in Human Peripheral Nerves

Abstract

We developed a time-efficient semi-automated axon quantification method using freeware in human cranial nerve sections stained with paraphenylenediamine (PPD). It was used to analyze a total of 1238 facial and masseteric nerve biopsies. The technique was validated by comparing manual and semi-automated quantification of 129 (10.4%) randomly selected biopsies. The software-based method demonstrated a sensitivity of 94% and a specificity of 87%. Semi-automatic axon counting was significantly faster (p < 0.001) than manual counting. It took 1 hour and 47 minutes for all 129 biopsies (averaging 50 sec per biopsy, 0.04 seconds per axon). The counting process is automatic and does not need to be supervised. Manual counting took 21 hours and 6 minutes in total (average 9 minutes and 49 seconds per biopsy, 0.52 seconds per axon). Our method showed a linear correlation to the manual counts (R = 0.944 Spearman rho). Attempts have been made by several research groups to automate axonal load quantification. These methods often require specific hard- and software and are therefore only accessible to a few specialized laboratories. Our semi-automated axon quantification is precise, reliable and time-sparing using publicly available software and should be useful for an effective axon quantification in various human peripheral nerves.

Figure 1

shows a schematic illustration of the peripheral extracranial facial nerve, its relevant branches and biopsy sites (red ellipses). The yellow lines indicate topographic levelling.

Figure 2

(A) Grey scale image, cranial zygomatic branch, 200x magnified, unprocessed. (B) Fascicles extracted (Background and artifact deletion) (C) Contrast enhancement with Fijis ‘CLAHE’ Local contrast enhancement function (D) ‘Auto Local Threshold’ – binary image (E)‚ Analyze Particles‘ (F) Overlay for demonstration purposes.

Figure 3

shows a Bland-Altman plot plotting the differences between the semi-automated and manual method on the y-axis against the manual counts as goldstandard values on the x-axis. In this depiction an outlier is removed for format purposes (−3832 axon difference, manual count 1894). The mean difference of automated and manual counts is −147 (blue line). Thus, the automatic method produces higher counts than manual. The range within the 95% limits of agreement is high (red dotted lines). The difference between counting methods increases with increasing absolute axon count (black regression line).

Figure 4

shows a Bland-Altman plot of axon counts up to 2000, representing the specific margin of interest. This figure shows clearly that in lower axon counts the methods match well, dispersing with higher counts. The mean difference between automated and manual counts is −136 (blue line). 95% limits of agreement are shown with red dotted lines, while the regression line is drawn as a black line.

Figure 5

shows a scatter chart of all axon counts comparing manual and semi-automated methods. The cut off value of 900 axons is indicated as a vertical and horizontal line. Both green shaded areas show the counts of semi-automated and manual methods matching greater than 900 (top right), or lower than 900 (bottom left). Areas shaded red show specimens falsely identified as having more than 900 axons by the automated method (top left), or falsely identified as having less than 900 axons (bottom right).