Purchase this article with an account.
S. Farsiu, B. A. Bower, J. A. Izatt, C. A. Toth; Image Fusion Based Resolution Enhancement of Retinal Spectral Domain Optical Coherence Tomography Images. Invest. Ophthalmol. Vis. Sci. 2008;49(13):1845.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
We hypothesize that recent advances in image processing can be used to generate high-quality spectral domain optical coherence tomography (SDOCT) retinal images even for non-cooperative patients. SDOCT image sequences are usually captured as progressively azimuthally translated B-scans, creating a noisy 3D representation of the retinal layers (volumetric scan). Alternatively, it is possible to capture a sequence of repeated SDOCT B-scans from a unique azimuthal position, which in a post-processing step are registered and summed, creating a unique less noisy image (lateral repeated scan). While for consenting adults, the azimuthal and lateral resolutions of SDOCT scans are often acceptable, in imaging infants or severely impaired adults, the uncontrolled rapid patient motion limits the acquired resolution as defined by the number of the B-scans in the volumetric or the number of A-scans in the lateral repeated scanning scenarios.
Novel image fusion methodologies were used to create images with high azimuthal and lateral resolutions. For the volumetric scanning scenario, instead of capturing a single time-consuming high-density 3D scan, we propose to capture several volume scans each with significantly fewer B-scans. As each sparsely sampled sequence is acquired relatively fast, patient motion artifacts are minimal. By exploiting known camera synchronization principles, these artifact-free sequences are registered and interlaced, creating a dense volumetric retinal representation. Alternatively, for the lateral repeated scanning scenario, fast B-scans with fewer A-scans can be captured. These low-resolution B-scans are then registered and interpolated, creating a less noisy B-scan with higher number of A-scans, compared to each individual B-scan.
For the volumetric scenario, two sparsely sampled SDOCT sequences of a patient retina were fused creating a denser 3D representation; the experiment was successfully repeated for fusing three such sequences. For the lateral repeated scanning scenario, images of a retina phantom with 250 A-scans were fused, creating an image with 1000 A-scans with reduced noise artifacts compared to each original frame.
A proof of concept for using modern image fusion algorithms to create high quality images from fast sparsely sampled scans is provided. We believe, this pioneering work will be used as a stepping stone toward many other image fusion based ophthalmic SDOCT system designs, aimed at patients with uncontrollable motion or pediatric imaging.
This PDF is available to Subscribers Only