Repeatability, Reproducibility, and Bias of the Tracking Scanning Laser Ophthalmoscope

Abstract

Purpose : To evaluate the sources of variation and accuracy of the tracking scanning laser ophthalmoscope (TSLO) as an eye-tracking measurement system

Methods : Gauge Repeatability and Reproducibility (Gauge R&R) was assessed on TSLO outputs to examine variability from apparatus, operator, signal input, and interaction between operator/signals. Twelve total pre-determined amplitude and frequency values of sinusoidal motion, representing the span of human fixation, were input into a custom-built steerable model eye. Three operators (AA,AB,AC) recorded three 10-second videos for each input on three separate days. The second frequency input at 30 Hz was discarded due to overlap with the device framerate. We completed a variance component analysis, and linearity and bias analyses to assess TSLO’s precision and accuracy. Bias was measured as the difference between the true value and observed value.

Results : Variance component analysis showed that part-to-part variation was the main contributor to variance for both frequency (99.99%) and amplitude (99.93%) inputs. The repeatability and reproducibility variance components contributed minimally to variability, at 0.007% for frequency and 0.14% for amplitude. There was a positive correlation between frequency signal input value and bias for all operators (max p=1.7e^-13), indicating the presence of linearity (i.e., higher frequencies have lower accuracy). For amplitude measurements, there was a tendency to underestimate large amplitudes, with linearity being present for operator 2 (p=0.03) and 3 (p=8.7e^-6). However, the maximum significant bias was 3.75 Hz at a frequency of 100Hz and -12 arcsecs at an amplitude of 360 arcsecs.

Conclusions : The TSLO can reliably distinguish between signals of different values, with very minimal contribution from apparatus error or operator bias. A large proportion of the TSLO variation stemmed from differences between signal inputs, 99.99% for amplitude and 99.3% for frequency. Gage R&R contribution fell well below 10%, at 0.007% for frequency and 0.14% for amplitude. Overall, TSLO showed excellent operator repeatability and apparatus reproducibility. Biases noted would be likely to be clinically insignificant.

This is a 2021 ARVO Annual Meeting abstract.

 

Linearity Plots for Frequency and Amplitude Signals by Operator. Bias is plotted against the signal input value for frequency (top) and amplitude (bottom).

 

Gage Linearity and Bias Results for Frequency and Amplitude.