Abstract
Purpose :
Optical coherence tomography (OCT) image quality in a remote care setting is important for reliable retinal segmentation and macular thickness analysis. OCT quality and ILM/RPE segmentation confidence maps are generated in the low-cost OCT processing pipeline, which are associated with the quality of the OCT cube scan and the segmentation confidence. In this study, we present the agreement between those two metrics, which may support the idea to use OCT signal quality as a segmentability measure of OCT data.
Methods :
The OCT signal quality map and the segmentation confidence map were generated by the methods presented at ARVO (Zacks et al., Invest. Ophthalmol. Vis. Sci. 2021; 62(8):2135; Kho et al., Invest. Ophthalmol. Vis. Sci.2022; 63(7):3313).
474 scans from 28 subjects (one or both eyes) were captured using a low-cost OCT prototype (ZEISS, Dublin, CA). The OCT scan quality and segmentation confidence were assessed using the ETDRS grid. Results from the regression and Bland-Altman analysis are reported.
Results :
The regression analysis parameters, mean difference and p-value for each subfield are shown in Figure 1. Two exemplary quality maps are visually compared to their corresponding segmentation confidence maps (Figure 2). An average correlation coefficient R of 0.82 and a mean difference of -0.021 indicate small differences between the two maps in most of the ETDRS grid sectors.
Conclusions :
This comparison of two methods with varying computational effort shows correlation that supports the idea of using the OCT quality map as a measure of segmentability to filter out low quality scans, which improves the acquisition workflow and patient experience. The robustness and accuracy of such a method could be improved by analyzing more scans and/or picking sectors with highest correlation.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.