Abstract
Purpose :
The clinical perception of corneal subbasal nerve tortuosity (CNT) was recently shown to have two distinct forms, namely a short-range tortuosity (SRT) and a long-range tortuosity (LRT). We developed a method for the automatic grading of confocal corneal nerve images according to these two forms of CNT.
Methods :
CNT as estimated from confocal microscopy is becoming a widely used clinical measure. The clinical perception of CNT was however recently shown to have two distinct forms, characterized by either short-range (high frequency, low amplitude) or long-range (low frequency, high amplitude) directional changes (Lagali et al., IOVS, 2015).
Seven expert graders from different institutions visually assessed 30 confocal microscopy images and ordered them by increasing nerve tortuosity, according to either tortuosity definition (Fig.1). Their average ordered rankings were assumed as GT.
Using the same 30 images, we traced corneal nerves with a custom computerized procedure and computed for each image both SRT and LRT. SRT was obtained by a custom combination of angle-, curve-, and twist-based tortuosity measurements at different scales. LRT was obtained by a custom combination of twist- and curve-based tortuosity measurements only, still at different scales. Measurements and scales used for SRT were in general different from the ones used for LRT, since the former have to detect high frequency and low amplitude directional changes, while the latter have to detect low frequency and high amplitude directional changes.
The correlation with respect to GT of the ordered ranking estimated using either SRT or LRT was then assessed using the Spearman Rank Correlation (SRC) coefficient.
Results :
The proposed method for the automatic grading of CNT using either SRT or LRT was able to achieve SRC coefficients of 0.98 and 0.96 with GT, respectively.
Conclusions :
The proposed SRT and LRT automatic grades highly correlate to clinical perception of CNT. We now need to investigate to what degree different cornea pathologies exhibit one or the other form in their subbasal nerve patterns.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.