Deep Learning Facilitated Study of the Relationship between Visual Field Sensitivity (VFS) and Photoreceptor Outer Segment (OS) Metrics in Retinitis Pigmentosa (RP)

Abstract

Purpose : It is known that VFS correlates with ellipsoid zone (EZ) width or area in RP. However, fewer studies reported the relationship between VFS and OS volume. Here we examined the association between VFS and 3D OS metrics with the assistance of a deep learning model (DLM) for automatic segmentation of retinal layers in OCT images.

Methods : Twenty-five patients with RP participated the study. Octopus spot size III VFS were obtained from both eyes of each patient (10-2 for EZ limited in the macula and 30-2 for EZ extended beyond macula). Spectralis high-resolution 9mm 121-line macular volume scans were obtained for all eyes. EZ and apex RPE were automatically segmented by a DLM (Wang et al., TVST 2021). Two graders performed manual correction of DLM segmentation to serve as a reference. The 3D OS map from a volume scan was reconstructed by interpolating the discrete 2D B-scan OS layers defined by EZ and apex RPE over the scan area. OS metrics, including mean OS length, EZ area and OS volume, were computed from the 3D OS maps generated from both DLM (fully automatic) and the reference (manual grading). A regression analysis was conducted to examine the relationship between the mean VFS in dB over the scan area and log OS metrics. Bland-Altman and correlation analyses were employed to compare OS metrics determined by DLM to the reference.

Results : Mean VFS, average OS length, EZ area, and OS volume ranged from 3.1 to 30.2 dB, 7.4 to 31.4 mm, 0.26 to 66.7 mm², and 0.0019 to 1.98 mm³, respectively. Mean VFS was significantly correlated with log average OS length (r = 0.45), EZ area (r = 0.84) and OS volume (r = 0.84) determined by DLM. The model performed similarly to the reference as shown in the Table. Bland-Altman analysis showed a close agreement between the OS metrics determined by DLM and the reference. EZ area and OS volume measured by the model was highly correlated with the reference (r=0.992 and r=0.996, respectively).

Conclusions : Resembling to EZ area, OS volume significantly correlates with retinal sensitivity. The large dynamic range of OS volume may render it being an effective biomarker to assess RP progression. The close agreement between OS metrics determined by DLM and the reference suggests that deep learning may provide efficient tools to facilitate the study on the structure and function relationship in RP.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.