Clinically Based Automated Tracing and Tortuosity Estimation of Corneal Nerve Fibers from Confocal Microscopy Images

Purpose: The purpose of this study was the development of an algorithm able to automatically trace corneal nerves and to estimate a nerve tortuosity index that is useful in clinical practice. Methods: In vivo confocal microscopy is an imaging technique that allows the clinical assessment of corneal and systemic diseases. Many studies have demonstrated a correlation between the tortuosity level of nerve fibers in the subbasal plexus layer and some pathologies. We developed an algorithm that provides fully automatic tracing of nerve fibers. It also includes a new way of dealing with bifurcations, separating the main paths from the secondary ones. Based on this automated tracing, the tortuosity was estimated as the absolute curvature, tortuosity density, and fractal dimension. These metrics were considered first individually and then as a linear combination of 2 or 3 of them. We investigated the capability of the estimated tortuosity to emulate the clinical classification into low, mid, and high tortuosity levels. Furthermore, we investigated its ability to distinguish healthy subjects from pathological subjects. Results: Excellent agreement between manual and automated grouping of tortuosity (96.6% accuracy) was obtained. Moreover, the proposed algorithm could differentiate between healthy and pathological subjects with an accuracy of 77.1% by analyzing each image individually. The accuracy improved to 86.31% by considering 3 images of the same subject simultaneously. Conclusions: The proposed framework provides completely automated analysis of corneal nerve images. The results demonstrate the ability of our method to emulate the clinical classification of tortuosity levels and its potential for identifying healthy and pathological subjects.