Using a variation of a previously reported algorithm,
15 we developed an algorithm in MatLab (version 6; The Math Works, Natick, MA) to determine the position and variability of the choroidal/scleral peak (P4) in PCI axial length tracings. Briefly, the algorithm analyzed the raw data consisting of up to 80 PCI tracings from a given subject. Each waveform was examined to determine whether it contained eye movements (characterized by large amplitude signal transients), whether reflections were in a consistent location, and whether there was a high signal-to-noise ratio. Based on these criteria, the algorithm accepted tracings that contained potential peaks and excluded tracings containing only noise or artifact. The accepted tracings with potential peaks (i.e.,
n tracings) were signal averaged and filtered to generate one posterior eye wall waveform that was curve-fit with three Gaussian curves
(Fig. 3) . The resultant curve-fit waveform was then analyzed for the presence of a P4 signal. In summary, a P4 was present if the P4 Gaussian peak was above the level of the noise and was within 550 μm from P3 (i.e., the RPE). If a subject’s curve-fit waveform contained a P4 signal, the variability of the P4 determination was estimated from the original dataset using a bootstrap technique.
17 Novel bootstrap datasets were generated by random sampling with replacement
n times from the original accepted dataset of
n tracings.
17 The bootstrap datasets were signal averaged and filtered to generate one bootstrap waveform that was curve-fit and analyzed for the presence of a P4 signal. The process was repeated until 200 bootstrap P4s were extracted, from which the variability was estimated. A more detailed description of the algorithm is available in the
Appendix.
Given the limited amount of data available in a single session, the assumptions for the algorithm were conservative because it was automated and user-independent when used to analyze a single measurement session for the presence and variability of a P4 signal. In analyzing the first two datasets containing subjects with only one measurement session, the algorithm sequentially ran through the signal-averaged data from each subject and determined whether there was a P4. If the subject’s readings contained a P4, the bootstrap statistical method was applied to calculate the variability for P4.
In contrast, for subjects in the diurnal dataset, there were 12 measurement sessions for each eye (six on each day). The additional measurement sessions in essence increased the signal-to-noise ratio of the P4 signal and the number of subjects whose data had an identifiable P4.
Figure 4illustrates how the availability of multiple measurement sessions at different time points for a single eye improved the identification of P4. It provided face validity to the approach used for the diurnal dataset, whereby the composite data from each eye were examined by one of the authors (JSB), and the algorithm was applied to each time point when there was visible evidence of a P4.