A reliable, fully-automatic pipeline for 3D motion correction and volume fusion enables investigation of smaller and lower-contrast OCT features

Abstract

Purpose : To reliably improve visualization and quantification of focal and low-contrast features by volumetrically fusing multiple OCT raster scans. Our pipeline inverts scanning effects including eye motion and OCT signal change in the retinal volumes. Applications include shape analysis of hyperreflective foci (HRF), improved retinal sub-band visualization, and thickness mapping of the micron-scale hyporeflective band between retinal pigment epithelium (RPE) and Bruch's membrane (BrM) in age-related macular degeneration (AMD).

Methods : 6 x 6 mm fovea-centered scans were acquired using a high-resolution OCT prototype developed at MIT (500 x 500 A-scans, 2.7 µm axial resolution, 128 kHz A-scan rate, no eye tracking). 6 volumes were acquired sequentially with orthogonally alternating fast scan direction (15 s acquisition) and fused with the pipeline detailed in Fig. 1.

Results : We analyzed 129 datasets from 98 eyes (19 – 93 y/o) with various pathologies and artifacts (repeated saccadic motions & blinks, low signal, …). Improved image quality and artifact reduction were achieved in a majority of datasets (Fig. 2). 5 datasets were excluded (4%, retina exceeded axial imaging range). 2 fusions failed (1.6%), 3 had residual artifacts limiting usage (2%), 12 had minor residual artifacts (9%).

Conclusions : Our volume fusion pipeline can achieve reliability comparable to eye tracking and correct eye motion more accurately. Volumetric, quantitative imaging biomarker studies are enabled which, with the additional image quality improvement, can investigate smaller and lower-contrast features, which will facilitate design of new clinical studies for disease pathogenesis and progression.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.

 

Flow chart of the fully-automatic volume fusion pipeline (example 71 y/o female). The foreground mask is needed for OCT signal/illumination correction. Blinks are removed. Retina cropping reduces computational load without loss of features. Denoising, prealignment and multi-resolution improve convergence of motion- and illumination correction, which are based on crossed optimization of inverse models (Ploner, MICCAI, 2022 & ISBI, 2023). Inconsistent data is detected by comparison to co-located scans. Corrected remaining data is merged.

 

Improved visualization of a) thin hyporeflective band between RPE and BrM, b) subband below external limiting membrane, c) HRF.