Abstract
Purpose: :
The number of A-scans that is feasible to acquire in a single OCT dataset is determined by the speed of the instrument and the length of the scan acquisition time. In commercially available SD-OCT instruments these factors limits the extension of the retinal area that can be covered by a reasonable uniform, reasonably dense scan pattern. Typically the field of view (FOV) is about 20x20 degrees. In order to expand the FOV, it is possible to compose a montage of several overlapping SD-OCT datasets, using the en-face features of the OCT fundus images (OFIs) to piece the scans together. The purpose of this study is to propose an automatic algorithm for montage of OFIs.
Methods: :
8 partially overlapping 200x200 OCT datasets of a given eye were acquired using a Cirrus HD-OCT (Carl Zeiss). By using blood vessel ridges as features, individual OFI pairs with sufficient overlap are registered. For each image pair, all corresponding pixels that are close enough are classified as matched pixel pairs. Global montage parameters are calculated by minimizing the sum of the square distances between all these matched pixel pairs. The procedure is then iterated, recomputing the set of matched pixel pairs and a new set of global montage parameters. Once the OFI montage is obtained, it is fairly straightforward to combine the full 3D datasets.
Results: :
The montage algorithm was tested on 6 sets (from 6 healthy eyes) and 4 sets (from 4 eyes with retinal degenerations) of 8 partially overlapping SD-OCT images with an expected resulting FOV of 35x50 degrees. For 6 healthy subjects, the montages were successfully constructed, and qualitative evaluation showed that the results of the automatic algorithm were better than manually performed montages. For 4 abnormal subjects, the central 4-6 OFIs (close to the optic disc or the fovea) have a good registration; the peripheral 2-4 OFIs have a bad registration due to lower blood vessel visibility. In order to improve montage performance for the abnormal subjects, we preprocessed OFIs to enhance blood vessel visibility, and included color fundus photographs (CFPs) in the montage.
Conclusions: :
Using our automatic montage algorithm, we can construct montages of several OFIs. These montages allow us to construct 3D OCT images over a large FOV out of separate OCT datasets acquired with a commercially available Cirrus instrument.
Keywords: image processing • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • retina