Abstract
Purpose :
A pattern electroretinogram (PERG) stimulus source was developed to probe the peripheral retina, an approach referred to as peripheral PERG (pPERG). The best analysis approach for pPERG responses is under exploration. Here we evaluate multiple approaches, including principal component analysis and cluster analysis, and compare the sensitivity of pPERG to that of traditional PERG in early and mid-stage glaucoma when using the most sensitive analysis method for each test.
Methods :
18 normally-sighted subjects and 13 glaucoma patients (HVF mean deviation 0.94 to -9.07, at least one OCT sector outside normal limits) were recruited. pPERG and PERG responses were recorded on the same day with DTL electrodes. For PERG, ISCEV-recommended standard stimulus parameters were used. The pPERG stimulus used 10 degree checks, 4.6 rps, presented to the peripheral retina (35-85 degrees of visual angle all directions), pattern ON-luminance approximately 15X the ISCEV-recommended minimum. Ability to detect responses from glaucoma patients was evaluated using traditional response features (amplitudes and implicit times of response peaks and troughs), the entire voltage vs. time response waveforms, cluster analysis, and principal components derived from the responses (PCA). Sensitivity of each approach was quantified using the area under ROC curves (AUC).
Results :
PCA performed on traditional feature values gave the best sensitivity for pPERG, yielding AUC of 79.5% and 93.2% for PERG and pPERG, respectively. Cluster analysis performed on the entire waveform gave the best sensitivity for PERG, yielding AUC of 88.0% and 72.2% for PERG and pPERG, respectively. In contrast, AUC for the best single key feature obtained in each test were 80.1% and 84.2%, for PERG (N95 amplitude) and pPERG (F1 amplitude).
Conclusions :
The best performing analysis approaches for detecting glaucomatous damage were comprehensive in their use of information contained in the response waveform, extending beyond the information contained in a single feature (e.g. N95 amplitude). This may be explained by the existence of a continuum of states of dysfunction among the different retinal cell types ultimately affected by glaucoma, resulting in a global effect on the waveform best detected by holistic analysis approaches.
This is a 2020 ARVO Annual Meeting abstract.