Abstract
Purpose :
To assess the sensitivity of the recently proposed fractal dimension (FD) based method to differentiate between normal and dry eye subjects using high speed videokeratoscopy (HSV).
Methods :
Retrospective data from HSV measurements of 19 normal subjects and 11 dry eye subjects were analysed. Dry eye classification was made based on a standard clinical assessment. Three measurements of right eye were taken in suppressed blinking conditions for each patient, with a 3-minute break between each measurement. For the analysis, the median of the three measurements was considered. The maximum time of recording was 30 s. Dynamic Tear Film Surface Quality (TFSQ) indicator, breaks feature indicator (BFI), and the distortions feature indicator (DFI) were computed by means of calculating the FD of the HSV images. The first second after a blink was removed from the analysis in order to avoid the formation phase of the tear film, the remaining data was used to estimate the means and the trends of TFSQ, BFI, and DFI. Additionally, a bilinear function was used to estimate the transition point that is referred here as the break-up time (BUT). The receiver operating characteristics (ROC) were calculated and the area under the curve (AUC) was computed, among other parameters that provide the discrimination performance of the method (cut-off value, Youlden’s index (γ) and discriminant power (DP)).
Results :
The parameter that provided the best performance in discriminating dry eye subjects from normal subjects was the mean BFI (AUC=0.83). Table 1 summarizes the AUC, sensitivity, specificity, γ and DP for the optimized cut-off point for the best performing parameter of each of the dynamics descriptors (TFDQ, BFI and DFI).
Conclusions :
The FD approach for HSV has shown to achieve better discrimination results than previously proposed method, bringing the performance of the HSV technology closer to that of the more sophisticated, but clinically unavailable, techniques such as the Lateral Sharing Interferometry.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.