@MASTERSTHESIS\{IMM2006-04443,
    author       = "J. Thomadsen",
    title        = "Characterization of retinal {OCT} images with macular holes",
    year         = "2006",
    keywords     = "{OCT,} optical coherence tomography, speckle, macular hole, regularized dynamic programming, retinal layers, image registration, di usion, active contours, snakes",
    school       = "Informatics and Mathematical Modelling, Technical University of Denmark, {DTU}",
    address      = "Richard Petersens Plads, Building 321, {DK-}2800 Kgs. Lyngby",
    type         = "",
    note         = "Supervised by Assoc. Prof. Bjarne Ersb{\o}ll, and Assoc. Prof. Rasmus Larsen, {IMM}.",
    url          = "http://www2.compute.dtu.dk/pubdb/pubs/4443-full.html",
    abstract     = "An imaging technology called Optical Coherence Tomography (OCT) has among other places found its application within ophthalmology. {OCT} can produce high- resolution cross sectional images of the internal microstructure of the retina. In this thesis methods to reduce noise in {OCT} will be investigated. {OCT} images are in particular a ected by a type of noise called speckle that arise due to interference. Three different types of di usion are applied to single {OCT} images to test its ability to reduce noise. None of the di usion methods produce satisfactory results, so an iterative method is developed that averages images taken of the same retinal location. Each image is registered vertically and horizontally to a template, before averaging is done. The method is robust to parametrical changes, and the average image has significantly less noise than the originals.

Retinal {OCT} images taken of a pathology called macular hole, are investigated to estimate descriptive parameters that could be relevant in evaluating the current state of the pathology. Di erent descriptors are evaluated pre- and postoperative. These descriptors are to be used in a case study at Herlev Hospital, where di erent surgical techniques to treat macular hole are evaluated. The descriptors can be extracted once a set of transitional layers have been located. They are found automatically or semi-automatically. If these layers are determined in {OCT} images scanning the retina at different locations, the neuroretinal thickness can be represented as a surface map or {3D} surface, in this way visualizing the entire retina instead of slices of it."
}