Ultrahigh resolution OCT, volume merging, and advanced signal reconstruction improve visualization of the RPE-Bruch’s complex

Abstract

Purpose : Subtle structural changes in the outer retina may be associated with physiological aging and age-related macular degeneration (AMD) pathology. We previously reported that basal laminar deposit (BLamD), an established histological marker for AMD, can be visualized clinically in early AMD eyes using ultrahigh resolution OCT (UHR OCT).
However, OCT images are degraded by speckle noise and sharpness is reduced by the acquisition process, making fine features including outer retinal layers and BLamD difficult to visualize directly. Merging multiple volumes can reduce noise and improve contrast, but requires motion correction accuracy below the scanner resolution, which is challenging due to fixational eye motion.
Using our scanner-independent framework for advanced computational motion correction and signal reconstruction (MoReOCT), we correct for involuntary saccadic motion and blinks, and improve signal-to-noise ratio and sharpness to enhance the visualization of thin outer retinal features.

Methods : Retrospective image processing was performed on OCT scans of a prospectively enrolled cohort of AMD patients (53 eyes / 39 subjects), as well as a cohort of healthy adult subjects (63 eyes / 39 s.) using a prototype UHR OCT instrument (2.7 µm axial resolution). For each eye, we acquired 6 volumetric scans over a 6 x 6 mm field centered at the fovea, with alternating horizontal and vertical B-scan direction. The eye motion trajectory of each scan was estimated from the data in a joint optimization such that the consistency of orthogonally scanned volumes was maximized.
To compensate image blur inherent in the acquisition process, we applied deconvolution at the scale of the measured PSF in the axial direction. While deconvolution increases speckle noise when applied to individual OCT acquisitions, this effect can be mitigated by jointly using all volume data in a 3-D iterative reconstruction with total variation denoising. The visualization setup uses moderate denoising, whereas the processing setup, intended for algorithmic analysis, uses more.

Results : The figures compare unprocessed B-scans to virtual B-scans resampled at identical A-scan locations from the MoReOCT volumes.

Conclusions : Volume merging and signal reconstruction can improve visibility of fine retinal structures and could aid in understanding AMD pathogenesis.

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

 

26 y/o healthy.

 

71 y/o with early dry AMD.