Abstract
Purpose :
Numerous Optical Coherence Tomography (OCT) and OCT Angiography (OCTA) volumes populate clinical and research databases. Yet, their usefulness is limited by inherent imaging artifacts (i.e. speckle noise and low signal to noise). The quality of these volumes is enhanced through aligning and averaging multiple previously collected OCT and OCTA volumes.
Methods :
Data from an OCTA device (AngioPlex, Carl Zeiss Meditec) was used following IRB approval for secondary image analysis. From healthy subjects, we selected 4 or more 3 mm scan volume pairs of OCT and OCTA data. These anisotropic (245x245x1024 voxels) volumes formed two datasets. Dataset 1 one had 9 volumes from a single subject's right eye. Dataset 2 had 32 eyes from 16 subjects and included repeat scan sessions. Our pipeline, implemented in Matlab (R2022b) and Flirt (6.0.5.2), followed a series of stages. Preprocessing resized the volumes to isotropic dimensions (200x200x300 voxels). Surface alignment utilized iterative closest points to align ILM layer surfaces of OCTs. Intensity based alignment employed a rigid 6-degree-of-freedom alignment using Flirt with specific rotational search parameters. OCT and OCTA images were separately averaged after alignment. Assessment of alignment quality relied on three intensity-based metrics: root mean square error (RMSE), normalized cross correlation (NCC), and mutual information (MI). These metrics were used to evaluate the accuracy of the alignment. To assess the quality of the averaged we used signal to noise ratio (SNR).
Results :
For OCTAs in dataset 1, RMSE decreased from 32.43 au to 11.00 au, NCC increased from 0.11 au to 0.92 au, and MI increased from 0.39 bits to 1.03 bits. For OCTAs in dataset 2, RMSE decreased from 21.36 au to 13.28 au, NCC increased from 0.30 au to 0.72 au, and MI increased from 0.16 bits to 0.43 bits. All changes were significant (p<0.01). In dataset 2, the average SNR increased from 8.6 dB to 13.8 dB to 16.5 dB to 18.3 dB for the average of 1, 2, 3, and 4 OCT volumes respectiviely.
Conclusions :
Our automated pipeline aligns OCT and OCTA volumes and improves alignment accuracy over the original machine alignment as measured by NCC, MI, and RMSE. This pipeline can be used to average and improve the quality of image alignments or track longitudinal changes in retinal vasculature.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.