State-of-the-Art in Retinal Optical Coherence

Tomography Image Analysis

Ahmadreza Baghaie, Roshan M. D’souza, Zeyun YuUniversity of Wisconsin-Milwaukee, WI, USA

November 18, 2014

Abstract

Optical Coherence Tomography (OCT) is one of the most emergingimaging modalities that has been used widely in the field of biomedicalimaging. From its emergence in 1990’s, plenty of hardware and softwareimprovements have been made. Its applications range from ophthalmologyto dermatology to coronary imaging etc. Here, the focus is on applicationsof OCT in ophthalmology and retinal imaging. OCT is able to non-invasively produce cross-sectional volume images of the tissues which arefurther used for analysis of the tissue structure and its properties. Dueto the underlying physics, OCT images usually suffer from a granularpattern, called speckle noise, which restricts the process of interpretation,hence requiring specialized noise reduction techniques to remove the noisewhile preserving image details. Also, given the fact that OCT images arein the µm -level, further analysis in needed to distinguish between thedifferent structures in the imaged volume. Therefore the use of differentsegmentation techniques are of high importance. The movement of thetissue under imaging or the progression of disease in the tissue also imposesfurther implications both on the quality and the proper interpretation ofthe acquired images. Thus, use of image registration techniques can bevery helpful. In this work, an overview of such image analysis techniqueswill be given.

1 Introduction

Optical Coherence Tomography (OCT) is a powerful imaging system for ac-quiring 3D volume images of tissues non-invasively. In simple terms, OCT canbe considered as echography with light [1]. Unlike the echography which isdone by sound waves, OCT imaging is not time-of-flight based and instead pro-duces the image based on the interference patterns. Fig. 1 shows a typicalretinal OCT image with false color. Throughout the past 2 decades, many im-provements have been achieved regarding the OCT imaging system which notonly improved the acquisition time but also the quality of the acquired images.Nowadays taking µm-level volume images of the tissues is very common espe-cially in ophthalmology and retinal imaging. Therefore the need for specializedOCT image analysis techniques is of high interest, making this by far the mostattractive area in biomedical imaging [2].

1

arX

iv:1

411.

0740

v2 [

cs.C

V]

17

Nov

201

4

Figure 1: Optical coherence tomograph of human retina and optic nerve [3]

From a practical point of view, several aspects of image processing techniquesare of high demand in analyzing OCT images which will be discussed in moredetail in this article. As a pre-processing step, noise reduction techniques areusually applied to OCT images. Due to the physical implications of coherentimaging, OCT images usually suffer from granular patterns which block finedetails in the tissue images. Plenty of techniques are proposed to remedy thisissue some of which will be mentioned in the following sections.

Use of image segmentation methods is another area which is widely exploredin OCT image analysis, especially in ophthalmology. For example, layered-structure of retina has been the subject of plenty of papers in the past few years.The methods that are used are usually specialized versions of the correspondingmethods in general image segmentation, ranging from graph-based techniques toactive contour methods etc. They will be further discussed later in this article.

Image registration also has its own place among the many image processingtechniques used in OCT image analysis. This becomes more obvious when onerealizes that OCT imaging is a µm-level imaging system which means evena small movement in the subject can have a big impact on the result. Eventhough the OCT imaging systems are very fast, still, the volume to be imagedis very dense so having an informative and comprehensive volume image requireshundreds of cross-sections which makes the imaging modality vulnerable to smallmovements. Also with the progress of degeneration in the tissue due to illness,image registration can be very useful in tracking the changes over time.

Based on the above-mentioned notes, the article is organized as follows: InSection 2 an overview of OCT imaging and different techniques that are used toacquire and reconstruct the images is given. Section 3 focuses on noise reductiontechniques. In Section 4 a few techniques for OCT image segmentation, with afocus on retinal layer segmentation are reviewed. Section 5 contains an overviewof the use of image registration techniques in OCT image analysis. Section 6concludes the article with pointers on some of the future paths that can be takenfor further investigations.

2

Figure 2: A typical OCT system [2]

2 Optical Coherence Tomography (OCT) Imag-ing

OCT imaging works based on the interference between a split and later re-combined broadband optical field [2]. Fig. 1 displays a typical OCT imagingsystem. The light beam from the source is exposed to a beam splitter and travelsin two paths: one toward a moving reference mirror and the other to the sampleto be imaged. The reflected light from both the reference mirror and the samplewill be fed to a photo detector in order to observe the interference pattern. Thesample usually contains particles (or layers) with different refractive indexesand the variation between neighbor refractive indexes causes intensity peaksin the interference pattern detected by the photo detector. By translating thereference mirror, a time domain interference pattern can be obtained. Also, byfrequency domain measurements of the output spectrum, depth information canbe derived. The result will be a line scan (aka A-scan) of depth of the 3D volumeto be imaged. Using multiple A-scans, 2D cross-sections and 3D volumes canbe constructed.

The most popular OCT imaging system is called time-domain OCT (TD-OCT) in which a reference mirror is translated to match the optical path fromreflections within the sample [3]. Unlike the TD-OCT, in Fourier-domain OCT(FD-OCT) there is no need for moving parts in the design of the imaging systemto obtain the axial scans [4]. In FD-OCT, the reference path length is fixed andthe detection system is replaced with a spectrometer. Because of not havingany moving parts, the speed of FD-OCT systems in acquiring images is veryhigh in comparison to TD-OCT.

Moving away from the classical behavior of light and going toward the quan-tum nature of light, Quantum OCT (Q-OCT) is a new imaging modality [5].Full field OCT (FF-OCT) is another optical coherence imaging technique inwhich a CCD camera is placed at the output instead of the single detector of

3

the TD-OCT to capture 2D en face image in a single exposure [6]. Takinginto account the polarization state of the light, polarization-sensitive OCT (PS-OCT) is another technique for imaging the birefringence within a biologicalsample [7]. Doppler OCT (D-OCT) which is also called optical Doppler to-mography (ODT) is a combination of OCT imaging system with laser Dopplerflowmetry. This system allows for the quantitative imaging of fluid flow in ahighly scattering medium; such as monitoring in vivo blood flow beneath theskin [8].

In the following sections more focus will be given to the software based imageanalysis techniques rather than hardware of OCT systems.

3 Noise Reduction

Speckle is a fundamental property of the signals and images acquired by narrow-band detection systems like SAR, ultrasound and OCT. In OCT, not only theoptical properties of the system, but also the motion of the subject to be imaged,size and temporal coherence of the light source, multiple scattering, phase devi-ation of the beam and aperture of the detector can affect the speckle [9]. Twomain processes affect the spatial coherence of the returning light beam which isused for image reconstruction: 1) multiple back-scattering of the beam, and 2)random delays for the forward-propagating and returning beam caused by mul-tiple forward scattering. In the case of tissue imaging, since the tissue is packedwith sub-wavelength diameter particles which act as scatterers, both of thesephenomena contribute to the creation of speckle. As stated in [9], two typesof speckle are present in OCT images: signal-carrying speckle which originatesfrom the sample volume in the focal zone; and signal-degrading speckle which iscreated by multiple-scattered out-of-focus light. The latter kind is what that isconsidered as speckle noise. Fig. 3 displays the common scene in retinal OCTimaging: a highly noisy image.

The distribution of the speckle can be represented with a Rayleigh distri-bution. Speckle is considered as a multiplicative noise, in contrast to Gaussianadditive noise. Due to the limited dynamic range of displays, OCT signals areusually compressed by a logarithmic operator applied to the intensity informa-tion. After this, the multiplicative speckle noise is transformed to additive noiseand can be further treated [22].

OCT noise reduction techniques can be divided into two major classes: 1)methods of noise reduction during the acquisition time and 2) post-processingtechniques. In the first class, aka compounding techniques, multiple uncorre-lated recordings are averaged. Among them, spatial compounding [11], angularcompounding [12], polarization compounding [13] and frequency compounding[14] techniques can be mentioned.

Usually the methods from the first class are not preferred since they requiremultiple scanning of the same data which extends the acquisition time. Plus,these techniques can be very restrictive in terms of different OCT imaging sys-tems in use. Therefore the use of more general post-processing techniques ismore favorable.

In the literature, plenty of post-processing methods are proposed for specklenoise reduction of OCT images. There are a few classic de-speckling techniquesfor noise reduction that are successfully used in OCT images too: Lee [15], Kuan

4

Figure 3: Typical retinal OCT image degraded by speckle noise

[16] and Frost [17] filters.One of the very interesting groups of methods for speckle reduction use the

well-known anisotropic diffusion method [18]. Due to its poor performance inthe case of very noisy images, there are different variations of this method inthe literature [19, 20, 21]. In [22] a complete comparison is provided for regularanisotropic diffusion and complex anisotropic diffusion approaches for denoisingof OCT images. Another example of using anisotropic diffusion is in [23] wherean Interval type-II fuzzy approach is used for determining the diffusivity functionof the anisotropic diffusion.

Another important group of widely used methods for de-speckling of OCTimages take advantage of multiscale/multiresolution geometric representationtechniques [1, 24]. The main reason for this is that these representations ofthe images compress the essential information of the image into a few largecoefficients while noise is spread among all of the coefficients. Generally, forsuch methods a transform domain that provides a sparse representation of im-ages and an optimal threshold is needed. Using the optimal threshold, hard-or soft-thresholding can be done for coefficient shrinkage in order to reduce thenoise. This threshold can be a constant or even better, sub-band adaptive orspatially adaptive. For example, in OCT images, since it is known that mostof the features are horizontal, higher thresholds can be applied to the verti-cal coefficients. Wavelet transform is one example of such methods which hasbeen widely used in this area [25]. Even though wavelet transform has shownpromising results, its lack of directionality imposes some limitations in properlydenoising OCT images. This can be further improved by use of dual-tree com-plex wavelet transform (DT-CWT) which doubles the directional informationof wavelet transform [26]. Still, there are better transform domains which of-fer more proper representations for OCT images. In the past few years use ofCurvelet transform [27, 28], Contourlet transform [29, 30] and Ripplet transform

5

Figure 4: FD-OCT image of optical nerve head, before (a) and after (b) curveletcoefficients’ shrinkage-based speckle noise reduction [27]

[31] showed promising results in the denoising of speckle noise. Fig. 4 showsthe result of curvelet coefficients’ shrinkage for speckle noise reduction.

Compressive sensing and sparse representation are new directions that aretaken for image acquisition and noise reduction. Very recent examples are[32, 33] which use dictionary learning and sparse representation for noise reduc-tion and image reconstruction of OCT. The process for these methods usuallyinvolves dividing the images into overlapping patches and training dictionaries.This is usually done by taking patches from both high-SNR-high-resolution (HH)and low-SNR-low-resolution (LL) samples from a volume data. The trained dic-tionary is then used in the process of denoising or reconstruction of images whichacquired using random and incomplete patterns of data reading.

4 Image Segmentation

Image segmentation is a very active area in the field of medical image analysis.The most difficult step in any medical image analysis system is the automatedlocalization and extraction of the structures of interest [34]. A critical rolefor medical image segmentation in OCT images involves the segmentation andextraction of regions of the image in order to quantify areas and volumes ofinterest in imaged tissues for further analysis and diagnosis.

The signal strength in OCT images raises from the intrinsic differences intissue optical properties. In retina, multiple layers of different types of cells can

6

Figure 5: Different views of the retina (a) Schematic illustration of cellularlayers of retina, (b) Light micrograph of a vertical scan through central humanretina, (c) Cross section of the eye with illustration of the retina, (d) OCT viewof macular retina [35]

be seen in OCT images. Fig. 2 displays different views of retina [35].During the last two decades plenty of methods are proposed for segmentation

of OCT images. For a more complete review on the methods the reader isreferred to [34] and [36]. Here an overview of the different classes of methodswill be given.

Four major problems are encountered in the segmentation of OCT images:

1. The presence of speckle noise complicates the process of the precise identi-fication of the retinal layers, therefore most of the segmentation methodsrequire pre-processing steps to reduce the noise.

2. Intensity of homogeneous areas decreases with the increase of the depth ofimaging. This is due to the fact that the intensity pattern in OCT imagesis the result of the absorption and scattering of light in the tissue.

3. Optical shadows imposed by blood vessels also can affect the performanceof segmentation methods.

4. Quality of OCT images degrades as a result of motion artifacts or sub-optimal imaging conditions.

Methods of retinal segmentation can be categorized in five classes: 1) Meth-ods applicable to A-scans, 2) intensity-based B-scans analysis, 3) active contourapproaches, 4) analysis methods using artificial intelligence and pattern recogni-tion techniques and 5) segmentation methods using 2D/3D graphs constructedfrom the 2D/3D OCT images [36, 37].

A-scan based methods consider the difference in the intensity levels to extractthe edge information. Due to the effects of speckle noise, these methods are

7

Figure 6: Results of segmentation for three different retinal OCT images withactive contour method proposed in [47]

usually used for determining the most significant edges on the retina. Examplesof such methods are presented in [3, 38, 39, 40, 41]. A-scan based methods lackthe contribution from 2D/3D data while the computational time is hight andthe accuracy is low. Taking advantage of better denoising methods, intensity-based B-scan analysis approaches work based on the intensity variations andgradients which are still too sensitive and make these methods case-dependent[42, 43, 44].

Active contour methods also provide promising results for segmentation ofretinal layers in OCT images. Methods presented in [45], [46] and [47] are threewell-known examples. Fig. 6 shows the results of layer segmentation using theactive contour-based method proposed in [47] for three OCT images.

Use of pattern recognition based methods is a new direction that attractedresearchers for more explorations. To name a few, in [48] a support vectormachine (SVM) is used to perfom the semi-automatic segmentation of retinallayers. In [49] Fuzzy C-means clustering method is used for layer segmentation.The use of random forest classifier for segmentation is also reported in [50].

2D/3D graph based methods are probably the best among all of the seg-mentation methods. During the past few years several variations of this classare introduced in the literature. Examples of such methods can be found in[37, 51, 52, 53, 54]. The method in [37] is based on diffusion maps [55] andtakes advantage of regional image texture rather than edge information. There-fore it is more robust in low contrast and poor layer-to-layer image gradients.On the other hand methods presented in [52, 53, 54] use the graph-cuts andshortest path method for segmentation of retinal layers. Fig. 7 shows the fi-nal 3D visualization of the segmented layers using the diffusion -based method

8

Figure 7: 3D visualization of the final result of segmentation [37]

proposed in [37].Despite the plethora of different methods in this area, retinal layer segmen-

tation is still an open problem with lots of room for improvement. Here, only afew of the well-known methods are shortly introduced and the reader is referredto the mentioned articles and references therein.

5 Image Registration

Having two input images, template image and reference image, image registra-tion tries to find a valid optimal spatial transformation to be applied to thetemplate image to make it more similar to the reference image [56]. Thereforethe process of image registration consists of an optimization process fulfillingsome imposed constraints. The applications of medical image registration rangefrom mosaicing of retinal images [57] to slice interpolation [58] etc.

There are two different classes of image registration techniques: 1) Para-metric approaches and 2) Non-parametric approaches [59]. In general, the pro-cess of finding the valid optimal spatial transformation can be formulated asE[u] = D[R, T ◦ u] + αS in which R and T are reference and template im-ages respectively, u is the displacement field, D is the similarity/dissimilaritymeasure (Mutual Information (MI), Normalized MI (NMI), Cross-Correlation(CC), Normalized CC, Sum of Squared Differences (SSD) and etc), S is theregularization term (which imposes additional constraints on the deformation),α is the coefficient that determines the amount of regularization and E is theenergy functional which depending on the problem either should be minimizedor maximized [60].

There are several different applications for using image registration ap-proaches in OCT image analysis. One of the basic applications is in specklenoise reduction. As mentioned in Section 3, hardware-based noise reductiontechniques take the average of several uncorrelated scans in order to reduce thenoise. The same thing can happen after finishing image acquisition too. The

9

Figure 8: Result of averaging affine registered cross-sections from close locationin retina

main idea is to gather several images of the same cross-section in retina, registerthem together and take the average. Having a small movement in the eye pro-vides us with an uncorrelated pattern of speckle. Having N B-scans, the SNRcan be improved by a factor of

√N . In [61] a dynamic programming based

method is used for compensation of the movements between several B-scansand reducing the speckle noise. A hierarchical model-based motion estimationscheme based on an affine-motion model is used in [62] for registering multipleB-scans to be used for speckle reduction. Fig. 6 shows the result of affine regis-tering the images from close locations in retina and averaging. Recently, the useof low-rank/sparse decomposition based batch alignment has been investigatedfor speckle noise reduction in OCT images too. Taking advantage of RobustPrincipal Component Analysis (RPCA) [63] and simultaneous decompositionand alignment of a stack of OCT images via linearized convex optimization,better performance is achieved in comparison to previous registration based de-noising techniques. For more details on the method and results, the reader isreferred to [64, 65]

Another application is in the registration of fundus images with en face OCTimages. This is very helpful to better correlate retinal features across differentimaging modalities. In [66] an algorithm is proposed for registering OCT fundusimages with color fundus photographs. In this paper, blood vessel ridges aretaken as features for registration. A similar approach is taken in [67] too. Usingthe curvelet transform, the blood vessels are extracted for the two modalitiesand then image registration is done in two steps: 1) registration based on onlyscaling and translation and 2) registration based on a quadratic transformation.Fig. 9 shows different stages of the results of color fundus and en face OCTimage registration of the method proposed in [66].

Motion correction of OCT images is another area in which image registration

10

Figure 9: Color fundus reference image superimposed by the correspondingblood vessel ridge image (a), en face OCT image superimposed by the corre-sponding blood vessel ridge image (b), the registration result of blood vesselridge image (c) and intensity images (d) [66]

techniques are of high interest. There are several different involuntary eye move-ments that can happen during the fixation: tremor, drifts and micro-saccades[68]. One way to deal with this issue is a hardware solution which involveshaving eye tracking equipment to compensate the eye movement during im-age acquisition time. Usually a Scanning Laser Ophthalmoscopy (SLO) deviceis merged with the OCT for tracking the eye movements during imaging [69].Taking a software approach can be more general and applicable without theneed for additional imaging equipment. In [70] at first the vessels are detectedand then using an elastic registration technique the tremor and drift motionsare corrected. Micro-saccades are corrected in the next step by finding the hor-izontal shift at each pixel in the scan that best aligns the result of tremor anddrift correction with the SLO image. In [71] a 3D aligning method for motioncorrection is proposed based on particle filtering. Fig. 10 shows one example ofthe use of the algorithm in [71] for correcting the motion in OCT images. Othertechniques can be found in [72, 73] and references therein.

Image mosaicing is another example of using image registration methods inOCT image processing. OCT systems are capable of acquiring a huge amountof data in a very short period of time; but still the field of view is very limited.Stitching several volume data from a patient can improve the interpretation ofdata significantly. Usually in such methods, a set of overlapping OCT data isacquired and then stitched together using global/local image registration tech-niques. In [74] at first a set of 8 overlapping 3D OCT volume data over a widearea of retina is acquired. Then the OCT en face fundus images are registeredtogether using blood vessel ridges as features of interest. Finally the 3D OCTdata sets are merged together using cross-correlation as a metric. A relativelysimilar approach is taken in [75] too for motion correction and 3D volume mo-saicing of OCT images. Fig. 11 shows a color encoded depth image of retina

11

Figure 10: OCT image with blood vessel discontinuity caused by eye movement(left) and corrected image (right) [71]

with three main vessel layers. One of the very recent works in this area can befound in [76] for bladder OCT images which is used with an integrated WhiteLight Cystoscopy (WLC) system.

The use of image registration techniques in analysis and processing of OCTimages has grown significantly in the past few years. Here only a few areas arementioned and there are more to explore.

6 Conclusion

Even though the techniques introduced here don not represent a comprehen-sive list of all of the approaches that are used for analysis of OCT images,the variety of the techniques gives us some pointers on the possible directionsfor further explorations. Especially with the emergence of parallel computingand GPU programming in the past few years, it is possible to have real timepre-processing and analysis systems to help with the increasing amount of dataproduced by OCT imaging systems. In this article, an overview of 3 major prob-lems in OCT image analysis is provided: noise reduction, image segmentationand image registration, and several different techniques in each category areintroduced. Starting from noise reduction, both hardware and software tech-niques are discussed to some extent. Moving to image segmentation, differentsegmentation techniques with a focus on retinal layer segmentation are intro-duced. Finally some of the image registration methods that are used for noisereduction, motion correction and image mosaicing of OCT images are reviewed.Overall, the use of image analysis techniques for OCT is a rapidly growing fieldand there remain many areas for investigation.

References

[1] Pizurica, Aleksandra, et al. ”Multiresolution denoising for optical coherencetomography: a review and evaluation.” Current Medical Imaging Reviews4.4 (2008): 270-284.

12

Figure 11: Widefield mosaic of retinal layers displaying three main vessel layers.The initial images are on the left and the final result is on the right[75]

[2] Tomlins, Peter H., and R. K. Wang. ”Theory, developments and applicationsof optical coherence tomography.” Journal of Physics D: Applied Physics38.15 (2005): 2519.

[3] Huang, David, et al. ”Optical coherence tomography.” Science 254.5035(1991): 1178-1181.

[4] Fercher, Adolph F., et al. ”Measurement of intraocular distances bybackscattering spectral interferometry.” Optics Communications 117.1(1995): 43-48.

[5] Nasr, Magued B., et al. ”Demonstration of dispersion-canceled quantum-optical coherence tomography.” Physical review letters 91.8 (2003): 083601.

[6] Beaurepaire, E., et al. ”Full-field optical coherence microscopy.” Optics let-ters 23.4 (1998): 244-246.

[7] de Boer J F, Srinivas S M, Nelson J S, Milner T E and Ducros M G et al2002 Handbook of Optical Coherence Tomography ed B E Bouma and G JTearney (New York: Dekker) pp 237-74

[8] Chen, Zhongping, et al. ”Optical Doppler tomography: imaging in vivoblood flow dynamics following pharmacological intervention and photody-namic therapy.” Photochemistry and photobiology 67.1 (1998): 56-60.

[9] Schmitt, Joseph M., S. H. Xiang, and Kin M. Yung. ”Speckle in opticalcoherence tomography.” Journal of biomedical optics 4.1 (1999): 95-105.

[10] Salinas, Harry M., and D. Cabrera Fernandez. ”Comparison of PDE-basednonlinear diffusion approaches for image enhancement and denoising in op-tical coherence tomography.” IEEE transactions on medical imaging 26.6(2007): 761-771.

13

[11] Avanaki, Mohammad RN, et al. ”Spatial compounding algorithm forspeckle reduction of dynamic focus OCT images.” Photonics TechnologyLetters, IEEE 25.15 (2013): 1439-1442.

[12] Schmitt, J. M. ”Array detection for speckle reduction in optical coherencemicroscopy.” Physics in Medicine and Biology 42.7 (1997): 1427.

[13] Kobayashi, Masaru, et al. ”Polarization-independent interferometricoptical-time-domain reflectometer.” Lightwave Technology, Journal of 9.5(1991): 623-628.

[14] Pircher, Michael, et al. ”Speckle reduction in optical coherence tomographyby frequency compounding.” Journal of Biomedical Optics 8.3 (2003): 565-569.

[15] Lee, Jong-Sen. ”Digital image enhancement and noise filtering by use oflocal statistics.” Pattern Analysis and Machine Intelligence, IEEE Transac-tions on 2 (1980): 165-168.

[16] Kuan, Darwin T., et al. ”Adaptive noise smoothing filter for images withsignal-dependent noise.” Pattern Analysis and Machine Intelligence, IEEETransactions on 2 (1985): 165-177.

[17] Frost, Victor S., et al. ”A model for radar images and its application toadaptive digital filtering of multiplicative noise.” Pattern Analysis and Ma-chine Intelligence, IEEE Transactions on 2 (1982): 157-166.

[18] Perona, Pietro, and Jitendra Malik. ”Scale-space and edge detection us-ing anisotropic diffusion.” Pattern Analysis and Machine Intelligence, IEEETransactions on 12.7 (1990): 629-639.

[19] Yu, Yongjian, and Scott T. Acton. ”Speckle reducing anisotropic diffusion.”Image Processing, IEEE Transactions on 11.11 (2002): 1260-1270.

[20] Abd-Elmoniem, Khaled Z., A. Youssef, and Yasser M. Kadah. ”Real-timespeckle reduction and coherence enhancement in ultrasound imaging vianonlinear anisotropic diffusion.” Biomedical Engineering, IEEE Transactionson 49.9 (2002): 997-1014.

[21] Krissian, Karl, et al. ”Oriented speckle reducing anisotropic diffusion.” Im-age Processing, IEEE Transactions on 16.5 (2007): 1412-1424.

[22] Salinas, Harry M., and D. Cabrera Fernandez. ”Comparison of PDE-basednonlinear diffusion approaches for image enhancement and denoising in op-tical coherence tomography.” IEEE transactions on medical imaging 26.6(2007): 761-771.

[23] Puvanathasan, Prabakar, and Kostadinka Bizheva. ”Interval type-II fuzzyanisotropic diffusion algorithm for speckle noise reduction in optical coher-ence tomography images.” Optics express 17.2 (2009): 733-746.

[24] Jacques, Laurent, et al. ”A panorama on multiscale geometric represen-tations, intertwining spatial, directional and frequency selectivity.” SignalProcessing 91.12 (2011): 2699-2730.

14

[25] Adler, Desmond C., Tony H. Ko, and James G. Fujimoto. ”Speckle reduc-tion in optical coherence tomography images by use of a spatially adaptivewavelet filter.” Optics letters 29.24 (2004): 2878-2880.

[26] Selesnick, Ivan W., Richard G. Baraniuk, and Nick C. Kingsbury. ”Thedual-tree complex wavelet transform.” Signal Processing Magazine, IEEE22.6 (2005): 123-151.

[27] Jian, Zhongping, et al. ”Speckle attenuation in optical coherence tomogra-phy by curvelet shrinkage.” Optics letters 34.10 (2009): 1516-1518.

[28] Jian, Zhongping, et al. ”Three-dimensional speckle suppression in opticalcoherence tomography based on the curvelet transform.” Optics express 18.2(2010): 1024-1032.

[29] Guo, Qing, et al. ”Image denoising algorithm based on contourlet transformfor optical coherence tomography heart tube image.” Image Processing, IET7.5 (2013): 442-450.

[30] Xu, Jianbing, et al. ”Speckle reduction of retinal optical coherence tomogra-phy based on contourlet shrinkage.” Optics letters 38.15 (2013): 2900-2903.

[31] Gupta, Deep, R. S. Anand, and Barjeev Tyagi. ”Ripplet domain non-linearfiltering for speckle reduction in ultrasound medical images.” BiomedicalSignal Processing and Control 10 (2014): 79-91.

[32] Fang, Leyuan, et al. ”Sparsity based denoising of spectral domain opticalcoherence tomography images.” Biomedical optics express 3.5 (2012): 927-942.

[33] L. Fang, S. Li, R. McNabb, Q. Nie, A. Kuo, C. Toth, J. A. Izatt, and S.Farsiu, “Fast acquisition and reconstruction of optical coherence tomographyimages via sparse representation,” IEEE Trans. Med. Imag., vol. 32, no. 11,pp. 2034–2049, Nov. 2013

[34] DeBuc, D. Cabrera. ”A review of algorithms for segmentation of retinal im-age data using optical coherence tomography.” Image Segmentation (2011):15-54.

[35] Garvin, M., ”Automated 3D segmentation and analysis of retinal OpticalCoherence Tomography images”, PhD thesis, University of Iowa, Iowa, USA,2008

[36] Kafieh, Raheleh, Hossein Rabbani, and Saeed Kermani. ”A Review of Al-gorithms for Segmentation of Optical Coherence Tomography from Retina.”Journal of medical signals and sensors 3.1 (2013): 45.

[37] Kafieh, Raheleh, et al. ”Intra-retinal layer segmentation of 3D optical co-herence tomography using coarse grained diffusion map.” Medical imageanalysis 17.8 (2013): 907-928.

[38] Hee, Michael R., et al. ”Optical coherence tomography of the humanretina.” Archives of ophthalmology 113.3 (1995): 325-332.

15

[39] Dillenseger, George A., Weber A, Pechereau A. Optical coherence tomog-raphy image processing. Invest Ophthalmol Vis Sci. (2000);41:165–73

[40] Koozekanani, Dara, Kim Boyer, and Cynthia Roberts. ”Retinal thicknessmeasurements from optical coherence tomography using a Markov boundarymodel.” Medical Imaging, IEEE Transactions on 20.9 (2001): 900-916.

[41] Shahidi, Mahnaz, Zhangwei Wang, and Ruth Zelkha. ”Quantitative thick-ness measurement of retinal layers imaged by optical coherence tomography.”American journal of ophthalmology 139.6 (2005): 1056-1061.

[42] Baroni, Maurizio, Pina Fortunato, and Agostino La Torre. ”Towards quan-titative analysis of retinal features in optical coherence tomography.” Medi-cal engineering and physics 29.4 (2007): 432-441.

[43] Tan, Ou, et al. ”Mapping of macular substructures with optical coherencetomography for glaucoma diagnosis.” Ophthalmology 115.6 (2008): 949-956.

[44] Kajic, Vedran, et al. ”Robust segmentation of intraretinal layers in thenormal human fovea using a novel statistical model based on texture andshape analysis.” Optics express 18.14 (2010): 14730-14744.

[45] Fernandez DC, Villate N, Puliafito CA, Rosenfeld PJ. Comparing totalmacular volume changes measured by optical coherence tomography withretinal lesion volume estimated by active contours. Invest Ophthalmol VisSci. (2004);45 E-Abstract 3072

[46] Mishra, Akshaya, et al. ”Intra-retinal layer segmentation in optical coher-ence tomography images.” Optics express 17.26 (2009): 23719-23728.

[47] Yazdanpanah, Azadeh, et al. ”Segmentation of intra-retinal layers fromoptical coherence tomography images using an active contour approach.”Medical Imaging, IEEE Transactions on 30.2 (2011): 484-496.

[48] Fuller, Alfred R., et al. ”Segmentation of three-dimensional retinal imagedata.” Visualization and Computer Graphics, IEEE Transactions on 13.6(2007): 1719-1726.

[49] Mayer, M., Tornow, R., Bock, R., Hornegger, J., Kruse, F., 2008. Auto-matic nerve fiber layer segmentation and geometry correction on spectraldomain OCT images using fuzzy c-means clustering. Investigative Ophthal-mology and Visual Science 49, E-Abstract 1880.

[50] Lang, Andrew, et al. ”Retinal layer segmentation of macular OCT imagesusing boundary classification.” Biomedical optics express 4.7 (2013): 1133-1152.

[51] Garvin, Mona Kathryn, et al. ”Intraretinal layer segmentation of macu-lar optical coherence tomography images using optimal 3-D graph search.”Medical Imaging, IEEE Transactions on 27.10 (2008): 1495-1505.

[52] Chiu, Stephanie J., et al. ”Automatic segmentation of seven retinal lay-ers in SDOCT images congruent with expert manual segmentation.” Opticsexpress 18.18 (2010): 19413-19428.

16

[53] Chiu, Stephanie J., et al. ”Validated automatic segmentation of AMDpathology including drusen and geographic atrophy in SD-OCT images.”Investigative ophthalmology and visual science 53.1 (2012): 53-61.

[54] Srinivasan, Pratul P., et al. ”Automatic segmentation of up to ten layerboundaries in SD-OCT images of the mouse retina with and without missinglayers due to pathology.” Biomedical optics express 5.2 (2014): 348-365.

[55] Coifman, Ronald R., and Stephane Lafon. ”Diffusion maps.” Applied andcomputational harmonic analysis 21.1 (2006): 5-30.

[56] Zitova, Barbara, and Jan Flusser. ”Image registration methods: a survey.”Image and vision computing 21.11 (2003): 977-1000.

[57] Patton, Niall, et al. ”Retinal image analysis: concepts, applications andpotential.” Progress in retinal and eye research 25.1 (2006): 99-127.

[58] Baghaie, A., Yu, Z.: Curvature-Based Registration for Slice Interpolationof Medical Images. In: Y.J. Zhang, J.M.R.S. Tavares (Eds.): CompIMAGE2014, LNCS 8641, pp. 69-80, Springer (2014)

[59] Modersitzki, Jan.: Numerical methods for image registration. OUP Oxford,(2003)

[60] Sotiras, Aristeidis, Christos Davatzikos, and Nikos Paragios. ”Deformablemedical image registration: A survey.” Medical Imaging, IEEE Transactionson 32.7 (2013): 1153-1190.

[61] Jorgensen, Thomas Martini, et al. ”Enhancing the signal-to-noise ratio inophthalmic optical coherence tomography by image registration—methodand clinical examples.” Journal of biomedical optics 12.4 (2007): 041208-041208.

[62] Alonso-Caneiro, David, Scott A. Read, and Michael J. Collins. ”Specklereduction in optical coherence tomography imaging by affine-motion imageregistration.” Journal of biomedical optics 16.11 (2011): 116027-1160275.

[63] Candes, Emmanuel J., et al. ”Robust principal component analysis?.” Jour-nal of the ACM, 58.3 (2011): 11.

[64] Peng, Yigang, et al. ”RASL: Robust alignment by sparse and low-rank de-composition for linearly correlated images.”, IEEE Transactions on PatternAnalysis and Machine Intelligence, 34.11 (2012): 2233-2246.

[65] Baghaie, Ahmadreza, Roshan M. D’souza, and Zeyun Yu. ”Sparse And LowRank Decomposition Based Batch Image Alignment for Speckle Reductionof retinal OCT Images.” arXiv preprint arXiv:1411.4033 (2014).

[66] Li, Ying, et al. ”Registration of OCT fundus images with color fundusphotographs based on blood vessel ridges.” Optics express 19.1 (2011): 7-16.

[67] Golabbakhsh, Marzieh, and Hossein Rabbani. ”Vessel-based registration offundus and optical coherence tomography projection images of retina usinga quadratic registration model.” IET Image Processing 7.8 (2013): 768-776.

17

[68] Hart Jr., W.H.: Adler’s physiology of the eye: clinical application, 9th edn.,Mosby-Year Book Inc., 11830 Westline Industrial Drive, St. Louis, Missouri63146 (1992)

[69] Pircher, Michael, et al. ”Simultaneous SLO/OCT imaging of the humanretina with axial eye motion correction.” Optics express 15.25 (2007): 16922-16932.

[70] Ricco, Susanna, et al. ”Correcting motion artifacts in retinal spectral do-main optical coherence tomography via image registration.” Medical ImageComputing and Computer-Assisted Intervention–MICCAI 2009. SpringerBerlin Heidelberg, 2009. 100-107.

[71] Xu, Juan, et al. ”Alignment of 3-D optical coherence tomography scansto correct eye movement using a particle filtering.” Medical Imaging, IEEETransactions on 31.7 (2012): 1337-1345.

[72] Kraus, Martin F., et al. ”Motion correction in optical coherence tomogra-phy volumes on a per A-scan basis using orthogonal scan patterns.” Biomed-ical optics express 3.6 (2012): 1182-1199.

[73] Kraus, Martin F., et al. ”Quantitative 3D-OCT motion correction with tiltand illumination correction, robust similarity measure and regularization.”Biomedical Optics Express 5.8 (2014): 2591-2613.

[74] Li, Ying, et al. ”Automatic montage of SD-OCT data sets.” Optics express19.27 (2011): 26239-26248.

[75] Hendargo, Hansford C., et al. ”Automated non-rigid registration and mo-saicing for robust imaging of distinct retinal capillary beds using speckle vari-ance optical coherence tomography.” Biomedical optics express 4.6 (2013):803-821.

[76] Lurie, Kristen L., Roland Angst, and Audrey K. Ellerbee. ”Automated mo-saicing of feature-poor optical coherence tomography volumes with an inte-grated white light imaging system.”, Biomedical Engineering, IEEE Trans-actions on, 61(7):2141-53 (2014)

18