Abstract
Purpose :
The robustness of macular thickness analysis (MTA) depends on the performance of the inner limiting membrane (ILM) and retinal pigment epithelium (RPE) segmentation. With a low-cost OCT, multiple scans are acquired to increase the probability of a successful MTA. This abstract proposes a method to generate a joint ILM/RPE segmentation confidence map to automatically select the most reliable scans for analysis.
Methods :
A low-cost OCT prototype system (ZEISS, Dublin, CA) and a CIRRUSTM HD-OCT 5000 (ZEISS, Dublin, CA) imaged 43 patients (70 eyes) with pathologies, including age-related macular degeneration. On the low-cost device, each eye was imaged 2-3 times using a 5.78x7 mm OCT volume capturing 512 A-scans/B-scan and 128 B-scans with 2.77 mm of depth. A 6x6 mm scan with the same density was also captured on CIRRUS.
The ILM and RPE layers were segmented in all images acquired with the low-cost device. A confidence map for each layer was generated from the processed image at each segmentation point. The algorithm was trained using maps data classified as high or low confidence, using a threshold to generate likelihood functions for Bayesian inference. The joint segmentation confidence maps (JSCM) of each set of OCT volumes were created by a posterior probability using the likelihood functions (Fig 1a). The ETDRS grid has 3 concentric circles of 0.5, 1.5, and 2.89 mm radius placed at the fovea center (Fig 1b). The confidence index for a scan was calculated by averaging the JSCM over the full grid.
From 215 scans on the low-cost device and JSCM, 70 high and 70 low-confidence scan MTAs of the same eyes (no scan overlap) were compared to 70 baseline CIRRUS MTAs of the same eyes.
Results :
Table 1 shows the correlation and Bland-Altman agreement for each ETDRS subfield between the low-cost OCT and CIRRUS MTA. There was a dramatic improvement when high confidence scans were used vs. low confidence scans.
Conclusions :
We demonstrated a joint ILM/RPE segmentation confidence method that shows the best scans for MTA. Multiple scans of the same eye can be acquired to increase the probability of acquiring a high confidence scan. Improved MTA in a low-cost device is important for detection and management of retinal diseases.
This is a 2021 ARVO Annual Meeting abstract.