WO2012027849A1 - Système et procédé d'acquisition rapide d'image oct à l'aide d'un échantillonnage compressif - Google Patents

Système et procédé d'acquisition rapide d'image oct à l'aide d'un échantillonnage compressif Download PDF

Info

Publication number
WO2012027849A1
WO2012027849A1 PCT/CA2011/050532 CA2011050532W WO2012027849A1 WO 2012027849 A1 WO2012027849 A1 WO 2012027849A1 CA 2011050532 W CA2011050532 W CA 2011050532W WO 2012027849 A1 WO2012027849 A1 WO 2012027849A1
Authority
WO
WIPO (PCT)
Prior art keywords
radial
optical coherence
coherence tomography
tomography image
image acquisition
Prior art date
Application number
PCT/CA2011/050532
Other languages
English (en)
Inventor
Mirza Faisal Beg
Evgeniy Lebed
Marinko V. Sarunic
Original Assignee
Simon Fraser University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from CA2737822A external-priority patent/CA2737822C/fr
Application filed by Simon Fraser University filed Critical Simon Fraser University
Priority to US13/819,713 priority Critical patent/US9014451B2/en
Priority to CA2844087A priority patent/CA2844087C/fr
Publication of WO2012027849A1 publication Critical patent/WO2012027849A1/fr

Links

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
    • A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
    • A61B3/102—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for optical coherence tomography [OCT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/003—Reconstruction from projections, e.g. tomography
    • G06T11/006—Inverse problem, transformation from projection-space into object-space, e.g. transform methods, back-projection, algebraic methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • G06T9/007—Transform coding, e.g. discrete cosine transform
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2211/00—Image generation
    • G06T2211/40—Computed tomography
    • G06T2211/424—Iterative

Definitions

  • the present invention relates generally to the field of signal acquisition and reconstruction, and more particularly, to the field of signal acquisition and reconstruction of images acquired using Optical Coherence Tomography ("OCT").
  • OCT Optical Coherence Tomography
  • OCT Optical Coherence Tomography
  • Optical coherence tomography works similarly to ultrasound tomography, however substituting use of light waves instead of sound waves to reconstruct images of tissue layers based on the reflection of light from tissue interfaces.
  • an OCT system can reconstruct a depth-profile of the sample structure.
  • Three-dimensional images can then be created by scanning the light beam laterally across the sample surface to create a 3D tomographic grid.
  • Conventional OCT scanning schemes suffer from a number of drawbacks.
  • Morphometric analysis and clinical applications typically demand high resolution OCT images, which necessitate dense sampling, leading to long scan times.
  • a person subject to the OCT scan is often required to remain still, sometimes for up to ten seconds, with a fixed gaze on a point without blinking - a challenging feat even for the most determined and those without eye health issues.
  • Long scan times may inevitably increase the likelihood of image corruption from motion artifacts, such as image blurring and ghosting resulting from the subject's eye blinking during an OCT scan session.
  • some OCT systems may implement motion tracking features in an attempt to reduce the potential effect of motion artifacts in OCT images, such motion tracking systems typically do not reduce OCT scan times, and may in fact increase scan durations.
  • Compressive sampling is a technique for signal acquisition and reconstruction utilizing the prior knowledge that the sampled signal is sparse or compressible in nature.
  • Conventional acquisition and reconstruction of images from frequency data using compressive sampling techniques typically follows the basic principle of the Nyquist density sampling theory, which states that to reconstruct an image, the number of Fourier samples that need to be acquired must match the desired resolution, and by extension, the number of pixels of the image.
  • Compressive sampling suggests the possibility of new data acquisition protocols that show that super-resolved signals and images may be reconstructed from far fewer data or measurements than that which was considered necessary under the Nyquist sampling theory.
  • An overview of compressive sampling may be found, for example, in E.J. Candes, J. Romberg, and T.
  • a method for rapid OCT image acquisition includes acquiring by OCT a plurality of compressive measurements (y) representing a set of under-sampled OCT data in a Dirac domain below a Nyquist rate by sampling an object of interest at randomly spaced vertical and horizontal lines in a Cartesian geometry using a raster scan, and recovering a 3D volumetric OCT image (f) from the compressive measurements (y) using compressive sampling.
  • a method for rapid OCT image acquisition further includes forming a sparsifying matrix (S) capable of transforming the 3D volumetric OCT image (f) into a sparse representation, and recovering the 3D volumetric OCT image (f) from the compressive measurements (y) based at least in part on the sparsifying matrix (S).
  • the sparsifying matrix (S) represents a matrix representation of the 3D volumetric OCT image (f) in a shift-invariant wavelet transform domain.
  • the compressive measurements (y) represent a range between 25 and 77 percent of the data present in the 3D volumetric OCT image (f). In other embodiments, the compressive measurements (y) represent less than 50 percent of the data present in the 3D volumetric OCT image (f).
  • the method for rapid OCT image acquisition comprises acquiring by Optical Coherence Tomography a plurality of compressive measurements (y) representing a set of under-sampled Optical Coherence Tomography data in a Dirac domain below a Nyquist rate by sampling an object of interest at randomly angularly spaced radial scan lines in a radial geometry using a radial scan.
  • the exemplary method further comprises mapping said compressive measurements (y) from said radial geometry to a Cartesian grid having radial coordinates; recovering a set of recovered radial scan images (f) from the
  • the compressive measurements (y) represent a range between about 12 and 40 percent of the data present in the set of recovered radial scan images (f).
  • mapping said compressive measurements (y) from said radial geometry to a Cartesian grid having radial coordinates additionally comprises mapping the radial coordinate ⁇ to the x Cartesian grid coordinate, and mapping the radial coordinate r to the y Cartesian grid coordinate.
  • the method for rapid Optical Coherence Tomography image acquisition may be adapted for use with either time-domain and/or Fourier domain OCT (also known as spectral domain OCT and/or swept source domain OCT) systems and imaging technologies, and also for use with other imaging systems, and in further alternative embodiments may be adapted for particular use with alternative emissive imaging means comprising one or more of: ultrasound, magnetic resonance and x-ray imaging.
  • OCT also known as spectral domain OCT and/or swept source domain OCT
  • alternative emissive imaging means comprising one or more of: ultrasound, magnetic resonance and x-ray imaging.
  • FIG. 1 illustrates a schematic view of a conventional OCT scan pattern
  • FIG. 2 illustrates a schematic view of an OCT scan pattern according to an embodiment of the present method for rapid OCT image acquisition using compressive sampling.
  • FIG. 3 illustrates a flow diagram of a method for rapid OCT image acquisition using compressive sampling according to an exemplary embodiment of the invention.
  • FIG. 4A illustrates an exemplary transversal slice, or C-scan, extracted from a full OCT volume of an optic nerve head acquired using conventional OCT raster scan geometry.
  • FIG. 4B illustrates a C-scan of a mask that discards 53% of the original samples as shown in FIG. 4A according to an embodiment of the invention.
  • FIG. 4C illustrates C-scan of a reconstructed OCT volume recovered from 47% of the original samples as shown in FIG. 4A missing according to an embodiment of the invention.
  • FIG. 4D shows contrast-enhanced differences between the results of FIG. 4A and FIG. 4C.
  • FIGs. 5A-5D show two vertical and horizontal original B-scans acquired by conventional OCT raster-scan geometry.
  • FIGs. 6B-6D show the same slices as FIG.s 5A-5D taken from a volume with 53% missing data according to an embodiment of the invention.
  • FIGs. 7A-7D show the same slices as FIG.s 5A-5D recovered from the compressive measurements by the CS interpolation method according to an embodiment of the invention.
  • FIGs. 8A-8D show the same slices as FIG.s 5A-5D recovered by a conventional bilinear interpolation of the same B-scans.
  • FIGs. 9A-9D show representative typical B-scans recovered according to an embodiment of the invention utilizing compressive sampling in volumes with 23, 35, 61 and 75 percent missing data.
  • FIG. 10A shows an extracted ILM surface from a segmentation of the original 3D image volume acquired using a conventional OCT raster scan geometry.
  • FIG. 10B shows an extracted ILM surface from a segmentation of a recovered 3D image volume with 53% missing data according to an embodiment of the invention.
  • FIG. IOC shows the average ILM where the color-map represents the variance in position in the three manual segmentations of the original image volume.
  • FIGs. 1 lA-1 IE show topographical Change Analysis (TCA) maps showing variations in the position of an ILM surface relative to mean, overlaid on top of summed- voxels-projection images of recovered volumes at 23%, 35%, 53%, 61% and 75% missing data, respectively, according to an embodiment of the invention.
  • FIG. 12 illustrates a graph plotting the reduction in raster scan time as a function of percent of missing data representing the extent of OCT data under-sampling according to an embodiment of the invention.
  • FIG. 13 illustrates a schematic representation of a radial OCT scanning pattern shown in FIG. 13(a); a schematic representation of radial OCT scan data mapped to a Cartesian transform grid shown in FIG. 13(b); a schematic representation of interpolation and reconstruction of radial OCT scan data using compressive sampling shown in FIG. 13(c); and a schematic representation of interpolated radial OCT scan data mapped to a Cartesian physical grid for analysis and display shown in FIG. 13(d), all according to an embodiment of the invention.
  • FIG. 14 illustrates in its top row, a typical en-face slice image extracted from a full OCT data volume of an optic nerve head acquired using a conventional OCT raster scan geometry, showing the en-face slice represented in three frames overlaid with random radial subsampling masks corresponding to 200 radial scans (61.5% missing data compared to conventional raster scan) in the left frame, 100 radial scans (80.9% missing data) in the center frame, and 60 radial scans (88.3% missing data) in the right frame, according to an embodiment of the invention.
  • the yellow lines on the overlaid en-face slice images show locations of six exemplary extracted B-scans further shown in FIG. 14A to FIG. 14F.
  • FIG. 14A illustrates in a series of frames from left to right, the exemplary B-scan labeled "A" extracted from a full conventionally raster-acquired OCT volume shown in the top row of FIG. 14; the same exemplary extracted B-scan corresponding to a random radial subsampling mask which discards 61.5% of the raster data (corresponding to 200 radial scans); a corresponding exemplary reconstructed B-Scan recovered from the remaining subsampled data in the 200 radial scan volume using compressive sampling; the exemplary B-scan corresponding to a random radial subsampling mask which discards 80.9% of the raster data (corresponding to 100 radial scans); a corresponding exemplary reconstructed B- Scan recovered from the remaining subsampled data in the 100 radial scan volume using compressive sampling; the exemplary B-scan corresponding to a random radial
  • subsampling mask which discards 88.3% of the raster data (corresponding to 60 radial scans); and a corresponding exemplary reconstructed B-Scan recovered from the remaining subsampled data in the 60 radial scan volume using compressive sampling, according to an embodiment of the invention.
  • FIG. 14B illustrates in a series of frames from left to right, the exemplary B-scan labeled "B" extracted from a full conventionally raster-acquired OCT volume shown in the top row of FIG. 14; the same exemplary extracted B-scan corresponding to a random radial subsampling mask which discards 61.5% of the raster data (corresponding to 200 radial scans); a corresponding exemplary reconstructed B-Scan recovered from the remaining subsampled data in the 200 radial scan volume using compressive sampling; the exemplary B-scan corresponding to a random radial subsampling mask which discards 80.9% of the raster data (corresponding to 100 radial scans); a corresponding exemplary reconstructed B- Scan recovered from the remaining subsampled data in the 100 radial scan volume using compressive sampling; the exemplary B-scan corresponding to a random radial
  • subsampling mask which discards 88.3% of the raster data (corresponding to 60 radial scans); and a corresponding exemplary reconstructed B-Scan recovered from the remaining subsampled data in the 60 radial scan volume using compressive sampling, according to an embodiment of the invention.
  • FIG. 14C illustrates in a series of frames from left to right, the exemplary B-scan labeled "C" extracted from a full conventionally raster-acquired OCT volume shown in the top row of FIG. 14 and the corresponding subsampled and recovered B-scan images corresponding to 200, 100 and 60 radial scan volumes, similar to as described above in FIG.s 14A and 14B, according to an embodiment of the invention.
  • FIG. 14D illustrates in a series of frames from left to right, the exemplary B-scan labeled "D" extracted from a full conventionally raster-acquired OCT volume shown in the top row of FIG. 14 and the corresponding subsampled and recovered B-scan images corresponding to 200, 100 and 60 radial scan volumes, similar to as described above in FIG.s 14A and 14B, according to an embodiment of the invention.
  • FIG. 14E illustrates in a series of frames from left to right, the exemplary B-scan labeled "E" extracted from a full conventionally raster-acquired OCT volume shown in the top row of FIG. 14 and the corresponding subsampled and recovered B-scan images corresponding to 200, 100 and 60 radial scan volumes, similar to as described above in FlG.s 14A and 14B, according to an embodiment of the invention.
  • FIG. 14F illustrates in a series of frames from left to right, the exemplary B-scan labeled "F" extracted from a full conventionally raster-acquired OCT volume shown in the top row of FIG. 14 and the corresponding subsampled and recovered B-scan images corresponding to 200, 100 and 60 radial scan volumes, similar to as described above in FlG.s 14A and 14B, according to an embodiment of the invention.
  • FIG. 15A illustrates a Topographical Change Analysis (TCA) map showing the position of an 1LM surface relative to the mean overlaid on top of summed-voxels- projection images of a recovered OCT volume at 61.5% missing data corresponding to a 200 radial B-scan volume, according to an embodiment of the invention.
  • TCA Topographical Change Analysis
  • FIG. 15B illustrates a TCA map showing the position of an 1LM surface relative to the mean overlaid on top of summed-voxels-projection images of a recovered OCT volume at 80.9% missing data corresponding to a 100 radial B-scan volume, according to an embodiment of the invention.
  • FIG. 15C illustrates a TCA map showing the position of an 1LM surface relative to the mean overlaid on top of summed-voxels-projection images of a recovered OCT volume at 88.3% missing data corresponding to a 60 radial B-scan volume, according to an embodiment of the invention.
  • the present invention provides a method for rapid OCT image acquisition using compressive sampling, in which OCT acquisition of a reduced random or pseudo-random subset of the raster scans that make up the 3D volumetric field of view of the entire image of a subject of interest is performed during the scanning process.
  • the recovery of missing samples may desirably be achieved with high fidelity, thereby reducing the overall OCT scan time and the associated patient discomfort and motion artifacts that may typically follow slower conventional OCT scanning schemes, while recovering a full 3D image of the scanned volume.
  • a predetermined transform domain e.g. a shift-invariant wavelet transform domain
  • the systems and methods disclosed in the present specification may be applied to both Fourier domain (also known as spectral domain and/or swept source) and time-domain based OCT image acquisition methodologies and applications according to embodiments of the present invention.
  • systems and methods disclosed in the present specification may also be adapted for use with other imaging systems, and in particular alternative embodiments may be adapted for use with alternative emissive imaging means comprising one or more of: ultrasound, magnetic resonance and x-ray imaging, for example.
  • the Nyquist sampling rate Prior to the introduction of compressive sampling, conventional wisdom as exemplified by the Nyquist sampling theorem states that a signal needs to be sampled at a rate greater than the inverse of twice its bandwidth to permit perfect signal reconstruction, commonly referred to as the Nyquist sampling rate. If the signal is a vector of length N, then the acquisition of at least N number of samples is required to reconstruct that signal according to Nyquist sampling methods.
  • the sensing or measurement matrix is therefore a N*N square matrix.
  • Compressive sampling however permits accurate signal reconstruction from highly under-sampled measurements, i.e. data sampled at significantly below the Nyquist rate.
  • compressive sampling may desirably offer successful recovery with relatively few under- sampled measurements K, where K «N. Accordingly, a sample measurement matrix may therefore be reduced to a K*N matrix.
  • the theory of compressive sampling provides that the measurement matrix needs to be incoherent or uncorrelated with the sparsifying matrix used to transform the signal into its sparse representation, or satisfies the so-called Restricted Isometry Property (RIP).
  • RIP Restricted Isometry Property
  • random measurement matrices are often used since they will most likely be uncorrelated with any specific sparsifying matrix. See E.J. Candes, J. Romberg, and T. Tao, Robust uncertainty principles: Exact recovery from highly incomplete Fourier information, IEEE Trans. Inform. Theory, 52 489-509 (2004).
  • an appropriate non-linear method enforcing signal sparsity and consistency of the reconstruction with the acquired samples can be used to reconstruct the signal from randomly under-sampled data.
  • a successful application of compressive sampling can be said to have three requirements: 1) Transform sparsity - the desired image should have a sparse representation in a known transform domain (i.e. it must be compressible by transform coding); 2) Incoherence of under-sampling artifacts - the artifacts in linear reconstruction caused by K-spare under-sampling should be incoherent in the sparsifying domain; and 3) Nonlinear reconstruction - the image should be reconstructed by a non-linear method which simultaneously enforces sparsity of the image representation and consistency of the reconstruction with the acquired samples.
  • a transform that provides sparse representation of 3D OCT image data in a basis different from one in which the image data is acquired is referred to as a "sparsifying transform", a mathematical operator that maps a vector of image data to a sparse vector.
  • a suitable sparsifying transform for the method for rapid OCT image acquisition using compressive sampling to recover data may be chosen depending on several factors: 1. Sparsity.
  • the sparsifying transform may desirably be able to capture most of the energy of the sampled OCT signal with only a few coefficients.
  • the basis functions of the sparsifying transform may desirably be as incoherent as possible with respect to the domain from which compressive signal measurements are acquired.
  • the sparsifying transform may desirably form a tight frame, or obey a generalized Parseval's identity.
  • the sparsifying transform may desirably pass the dot-test.
  • Suitable sparsifying transforms satisfying the above-mentioned four factors according to an embodiment of the invention may include the curvlet transform, surfacelet transform, and wavelet transform, for example.
  • the sparsifying transform may be a shift-invariant wavelet transform.
  • f is a vector of N entries that denotes the 3D image of a subject of interest, such as a biological tissue structure for example, defined on a 3D Cartesian grid, obtainable by OCT using conventional raster scan geometry.
  • the method for rapid OCT image acquisition using compressive sampling proposes to recover the full 3D image f by taking K measurements of the full 3D image f at a sampling rate significantly below the Nyquist rate.
  • the reduced number of measurements from which the full data vector f may be recovered are hereinafter referred to as compressive measurements or samples, represented by vector y in the equation (1.1) with K entries, where K «N, and where each component of y is a linear measurement of the full data vector f.
  • n represents the noise present in the acquired OCT data.
  • the K compressive measurements represented by y may comprise a subset of measurements containing between 25% and 77% of the number of measurements or samples in the full 3D image f, for example.
  • the matrix M in equation (1.1) refers to the domain from which the compressive measurements y are acquired.
  • the OCT is a Fourier domain or time-domain OCT, and OCT images are acquired in a 3D space with Cartesian geometry.
  • the matrix is a restriction operator, i.e., it extracts those rows from the identity matrix M that represent the number of compressive measurements that are actually acquired.
  • the dimensions of the measurement basis and the restriction operator may be determined by the acquisition geometry, or the raster scanning geometry of the OCT image acquisition system or scanning path.
  • the basis functions of the sparsifying transform selected in embodiments of the invention may desirably be as incoherent as possible with respect to the domain from which compressive measurements are acquired, and thereby represent mutual incoherency. It has been shown that sampling an object of interest in a measurement basis at randomly spaced vertical and horizontal lines in a Cartesian geometry skipping a large portion of the image during acquisition satisfies this mutual incoherency criterion, as random measurement matrices are often used since they will most likely be uncorrelated with any specific sparsifying matrix. See E.J. Candes, J. Romberg, and T. Tao, Robust uncertainty principles: Exact recovery from highly incomplete Fourier information, IEEE Trans. Inform.
  • the acquired subset of OCT image data or measurements, y may be acquired corresponding to random vertical and horizontal lines imposed on the physical volume to be imaged.
  • Such random vertical and horizontal scanning or measurement lines may be generated by a binary restriction mask R comprising vertical and horizontal lines with uniform and random placement, which may be applied during data acquisition such that certain random horizontal and vertical raster-scan lines are skipped during the raster-scanning process, thereby generating randomly sub-sampled OCT image volumes y of the 3D image of a subject of interest f.
  • a restriction mask results in the reduction in the number of samples being acquired as compared to a conventional OCT raster scan, as best illustrated by FIGs. 1 and 2.
  • FIG. 1 which illustrates a schematic of a conventional OCT scan pattern
  • the volume of a subject of interest 1 10 which may comprise a biological body or tissue structure such as an human eye
  • OCT optical coherence tomography
  • a typical raster scan geometry whereby every point in the volume is imaged by scanning every horizontal line 130 in the volume with a light beam 120.
  • 100% of the volume of the subject of interest 1 10 is scanned or imaged.
  • FIG. 2 illustrates a schematic of an OCT scan pattern according to an embodiment of the present method for rapid OCT image acquisition using compressive sampling. As shown in FIG.
  • a volume of a subject of interest 210 is imaged by scanning only selected, randomly-spaced fixed horizontal lines 230 and vertical lines 240, whereby only a subset of the raster scans that make up the 3D volumetric field of view of the entire image of a subject of interest is sampled.
  • compressive measurements represent 25% of the total raster scans that make up the 3D volumetric field of view of the entire image with 75% of the data missing.
  • compressive measurements represent subsets of 77%, 65%, 47% and 39% of the total raster scans of the entire image sampled, with 23%, 35%, 53%, and 61% of the data missing, respectively.
  • the compressive measurements (y) represent a range between 25 and 77 percent of the data present in the 3D volumetric OCT image (f). In other embodiments, the compressive measurements (y) represent less than 50 percent of the data present in the 3D volumetric OCT image (f).
  • the theory of compressive sampling states that adding the constraint that the signal f is compressible enables one to solve this underdetermined system of linear equations. That is, if one can find a sparsifying transform domain represented by matrix S that offers good compressibility, i.e.
  • the unknown sparse coefficients x parameterizing f may be selected such that the reconstructed signal S ⁇ x followed by the application of a known restriction mask is close to the observed under-sampled data y, subject to the constraint that the mean-squared error
  • 2 is small.
  • the desired x ⁇ is one with the smallest t ⁇ norm
  • ⁇ representing a measure of sparsity. Therefore, the recovery of sparsely sampled OCT images may be realized by solving the following optimization problem via i ⁇ minimization:
  • the matrix may be determined as a synthesis matrix in the shift-invariant wavelet domain which offers good compressibility of OCT images.
  • Other sparsifying domains that offer good compressibility of OCT images and satisfy the conditions as discussed in the above-section titled
  • Transform Sparsity may also be used in other embodiments of the invention.
  • the optimization problem in equation (1.2) is a convex optimization problem and may be cast into a linear program to solve the problem in equation (1.2), from which suitable conventional linear programs from the field of convex optimization may be employed.
  • the method for rapid OCT image acquisition using compressive sampling may employ the iterative soft thresholding (1ST) solver to solve the i ⁇ minimization problem given in equation (1.2).
  • ⁇ is a parameter that controls the tradeoff between the sparsity of the solution (minimization of i ⁇ norm) and the data misfit.
  • a matrix A is defined to be A: RMS ⁇ .
  • An exemplary 1ST program code according to an embodiment of the invention for solving the i ⁇ minimization problem may be represented as follows:
  • the iterative update x ⁇ T ⁇ (x + A T (y - Ay)) desirably converges to the solution for a particular value of ⁇ as the number of iterations goes to infinity.
  • the full OCT image f is recovered by the following equation:
  • the full 3D OCT image volume may be recovered with the 1ST solver using approximately six (6) inner loops and forty (40) outer loop iterations, for example.
  • the OCT system is a Fourier domain or time-domain OCT system and the measurement basis in which the compressive measurements are acquired may be the Dirac basis, however in other embodiments of the invention, other measurement bases may be utilized by way of modifying the OCT system hardware by which the OCT data is acquired, as may be obvious to a person skilled in the OCT art. In other embodiments, other measurement bases may be used, such as sparcifying transform matrices which form a tight-frame and pass a dot-test, for example.
  • One way of checking whether a specific measurement matrix will allow for sparse solutions to be recovered and thereby be potentially suitable for utilization according to an embodiment of the present invention, is to check if the measurement matrix follows the Restricted Isometry Property (RIP).
  • RIP Restricted Isometry Property
  • the 1ST solver is selected as the it minimization algorithm for its fast convergence and ease of implementation.
  • any other suitable known it minimization algorithms may be applied, such as iteratively reweighted least squares, SPGL1 solver, projected gradient methods, and iterative hard-thresholding, for example
  • FIG. 3 illustrates a flow diagram of a method for rapid OCT image acquisition using compressive sampling according to exemplary embodiment of the invention.
  • the method begins with the modeling of the signal recovery problem as shown in equation (1.1) in the physical domain, or Fourier domain or time-domain OCT.
  • K compressive measurements (y) representing a subset of the raster scans that make up the 3D volumetric field of view of the entire image (f) of a subject of interest are sampled in the Dirac domain by random under-sampling.
  • the method for rapid OCT image acquisition using compressive sampling proceeds to signal reconstruction.
  • a sparsifying domain that offers good compressibility of the full 3D image f is selected.
  • the sparsifying transform used is the shift-invariant wavelet transform.
  • the recovery of sparsely sampled OCT images modeled by the problem in equation (1.1) may be realized by solving the optimization problem via i minimization in equation (1.2) using a suitable ⁇ minimization algorithm, such as using the 1ST solver to find an approximation of the sparsifying coefficients x at step 340.
  • the full 3D image F is reconstructed using equation (1.4) in the physical domain, or Dirac basis at step 350.
  • the method for rapid OCT image acquisition using compressive sampling of the invention by acquiring a small random or pseudo-random subset of the raster scans that make up the 3D volumetric field of view of the entire OCT image of a subject of interest, followed by exploiting the sparsity of image coefficients in a predetermined transform domain (e.g. shift-invariant wavelet transform domain), enables the recovery of missing samples with high fidelity, thereby advantageously reducing the overall OCT scan time and the associated patient discomfort and motion artifacts that typically follow conventional OCT scanning schemes.
  • a predetermined transform domain e.g. shift-invariant wavelet transform domain
  • a full 3D OCT original image volume (f) was acquired by raster- scanning the optic nerve head of a healthy male subject using OCT.
  • An example of a transversal slice, or C- scan, extracted from the OCT volume is shown in FIG. 4A.
  • a binary restriction mask R is applied to this full OCT image f. This mask R consists of vertical and horizontal lines with random (but fixed) placing.
  • the full 3D OCT image volume for each level of subsampling was recovered according to an embodiment of the present method for rapid OCT image acquisition using compressive sampling employing an exemplary 1ST solver (detail: using 6 inner and 40 outer loop iterations).
  • SNR based on recovered images corresponding to 23, 35, 53, 61, and 75 percent missing data was found to be 14.9, 13.0, 10.6, 9.8, and 7.5 db respectively showing decreasing SNR with increasing degradation due to missing data.
  • FIGs. 5A-5D show two vertical (FIGs. 5 A and 5B) and horizontal (FIGs. 5C and 5D) original B-scans acquired from conventional OCT using raster-scan geometry.
  • FIGs. 6A-6D show the same slices as FIG.s 5A-5D, respectively, taken from an image volume with 53% missing data, corresponding to an imaging sampling rate of 47% of the total volume desired to be imaged.
  • FIGs. 7A-7D show the same slices as FIG.s 5A-5D, respectively, after recovery from the compressed or sparsely sampled scan slices shown in FIGs. 6A-6D using the CS interpolation method as detailed above according to an embodiment of the invention.
  • FIGs. 8A-8D show the same slices as FIG.s 5A-5D, respectively, recovered by bilinear interpolation of the same B-scans of FIGs. 5A-5D.
  • FIGs. 9A-9D show representative typical B-scans recovered using the CS interpolation in imaged volumes with 23, 35, 61 and 75 percent missing data, respectively (corresponding to compressive sampling rates of 77, 65, 29 and 25 percent, respectively), from a region where data was not sampled. It can be seen that the overall image quality begins to deteriorate progressively from FIGs. 9A to 9D as more data is discarded
  • the extracted ILM surface from one of the segmentations of the original 3D image is shown in FIG. 10A, where the color map represents the vertical height (to better present the 3D nature of the cup).
  • the recovered image may be deemed to have passed a test of required fidelity for potential morphometric use if the location of the ILM surface segmented from the recovered image falls within the variability of the surface segmentations performed multiple times on the original image volume by the same rater.
  • the ILM in the original 3D image was segmented thrice, each time by the same rater but without consulting the previous segmentations.
  • an average ILM surface was created representing a baseline or ground truth of the ILM surface.
  • the standard deviation of each point's position as observed in the three manual segmentations of the ILM surface was calculated, and the height or color of the colormap shown in FIG. IOC represents the variance in position in the three manual segmentations of the original image volume.
  • the surface shown in FIG. IOC is therefore the average ILM surface where the colormap represents the variance in position in the three manual segmentations of the original image volume.
  • TCA Topographical Change Analysis
  • Topographical Change Analysis maps showing the position of the ILM surface relative to the mean, overlaid on top of summed- voxels-projection images of recovered volumes at 23%, 35%, 53%, 61% and 75% missing data (corresponding to sampling rates of 77%, 65%, 47%, 39%, and 25% respectively in embodiments of the present invention) are shown in FlGs. 1 lA-1 IE, respectively.
  • the points on the recovered ILM surface lying more than one standard deviation from the mean ILM surface are shown in FlGs. 1 lA-1 IE with red representing posterior surface location variance, and green representing anterior surface location variance in the recovered image.
  • embodiments of the present invention begin to be observable when one discards approximately 25% of the data, or more (corresponding to a sampling rate of 75% or lower). Using a compressive subsampling pattern with sampling rates of 47% or 39% (corresponding to 53% or 61% missing data) roughly halved the acquisition time.
  • a method for rapid OCT image acquisition using radial image collection or scanning geometry and compressive sampling in which OCT acquisition of a reduced random or pseudo-random subset of radially oriented scans that make up the 3D volumetric field of view of the entire image of a subject of interest is performed during the radial scanning process.
  • the data comprised in the reduced random or pseudo-random subset of radial scans are further transformed and mapped to a Cartesian coordinate system with coordinates of (r, ⁇ ) on a rectangular Cartesian grid prior to compressive sampling (CS) interpolation and reconstruction, such as by using the exemplary CS interpolation and reconstruction method described above in reference to raster scanning embodiments, for example.
  • CS compressive sampling
  • the reconstructed radial scan data is thereafter transformed from radial coordinates of (r, ⁇ ) to an (x, y) Cartesian coordinate grid to transform and restore the physically corresponding positions of the reconstructed image to a 3D Cartesian grid to facilitate storage and display of the OCT volume in a conventional 3D Cartesian grid-based system.
  • the procedure comprises an OCT imaging process using sparse radial samples of the subject volume to be imaged, which in one embodiment may comprise at least a portion of the retina and/or optic nerve head of a subject, or other ophthalmological or biological subject of interest for example.
  • the OCT data is initially acquired by randomly-spaced radial B-scans of the subject.
  • random- radial B-scans refer to OCT scans where the angle of a subsequent B-scan, relative to the previous B-scan, is drawn from a random or pseudo-random distribution.
  • Suitable such radial B-scan "spokes" consisting of a fixed number of A-scans may be acquired at a random angular spacing within a range defining a maximum and minimum allowable step size, for example, to provide a randomly subsampled set of radial scans of the subject volume.
  • the sparseness of the random radial B-scan subsampling acquisition may accordingly be determined by selecting the maximum allowable angular step size between radial B-scans.
  • FIG. 13(a) illustrates a schematic representation of an exemplary random radial OCT scanning pattern according to a radial embodiment of the present invention.
  • a suitable random radial B-scan pattern may be constructed by selecting a first radial scan to be oriented in an arbitrary reference direction, such as horizontal or vertical, for example, and thereafter incrementing the radial angle of subsequent radial B-scans in radial angle increments that are drawn from a random or pseudo-random distribution while imposing a suitable upper limit on maximum allowable radial angle increments between any two adjacent radial B-scans, in order to define a random subsampled set of radial B-scans of the subject volume.
  • a radial angle increment between subsequent radial scans may be randomly drawn from a uniform distribution over [0, ⁇ / ⁇ ] where n may be chosen depending on the size of the smallest feature that may be desired to be resolved in the scanned image.
  • selection of a radial scan geometry for OCT image acquisition may desirably allow for concentration of sampling of features in the center of the radial scanning field, while allowing for reduction of sampling concentration in areas radially distant from the center of the radial scanning field.
  • the radial scanning field may be desirably centered on the optic nerve head or optic cup, such that each random radial B-scan inherently collects a denser distribution of samples close to the center of the optic nerve head, and that each radial B- scan contains a cross-section of the optic cup, regardless of the angular direction of the radial B-scan.
  • random radial OCT scanning patterns may desirably provide for improved sampling density and thereby also improved resolution of desired features to be imaged near a central region of interest in the scanned subject volume, and the location of the center of the radial scan pattern may be selected to focus the increased sampling density on a particular part of a subject, such as the optic nerve head, fovea or macula in exemplary ophthalmological applications, or of other features of interest in other applications of such random radial scanning according to further embodiments of the invention.
  • the mapping of the (r, ⁇ ) coordinates of the radial B-scans to a Cartesian grid is substantially equivalent to "stacking" the radial B-scans vertically along the x-axis, where the placement of each radial scan "spoke” in the horizontal direction is based on the angle ⁇ of the radial scan.
  • "missing" radial scans which are not acquired due to the random subsampling of the radial B-scan pattern to be used for CS become equivalent to missing vertical lines or vertical raster scans in the Cartesian grid, as may be seen by the gaps between vertically aligned radial B-scans in the schematic view of FIG. 13(b), according to an embodiment of the invention.
  • the method comprises the step of interpolating the missing equivalent vertical raster scans in the Cartesian grid according to a suitable CS method to reconstruct a series of equivalent vertical raster scans and provide a suitably accurate interpolated set of vertically oriented B-scans in the Cartesian grid, as illustrated in schematic view in FIG. 13(c), for example.
  • the CS interpolation and reconstruction method described above for a raster scanned embodiment may be applied to interpolate and reconstruct the missing vertically oriented radial B-scans, for example.
  • other potentially suitable interpolation methods may be used to interpolate and reconstruct the "missing" vertically oriented scans, such as
  • incorporating CS interpolation of the "missing" vertically oriented scans may be used to denote the full 3D tissue image, R to be a 3D restriction mask that restricts the image samples actually acquired, M to be a measurement basis and y to be the image samples actually acquired.
  • the above constrained optimization problem may be solved using a suitable exemplary Iterative Soft-Thresholding solver, for example.
  • the method comprises the step of mapping the interpolated B-scan and acquired B-scan data back onto a representative physical 3D Cartesian volume by applying the inverse of the earlier mapping, as illustrated in schematic view in FIG. 13(d).
  • a standard 3D Cartesian grid for defining the reconstructed OCT image desirably facilitates the storage, display, manipulation and analysis of the image using standard or conventional imaging formats and systems, for example.
  • a full 3D OCT original image volume (f) was acquired by raster-scanning the optic nerve head of a healthy male subject using Fourier- domain OCT.
  • the exemplary OCT system used for the scan had a line rate of 20kHz, with axial and lateral resolutions of 4 ⁇ and 25 ⁇ , respectively.
  • the cross sectional B-scans consisted of 512 A-scans and the full volume consisted of 400 B-scans, covering an area of approximately 3.7mm x 1.8mm.
  • a first radial B-scan is taken oriented along the horizontal axis, and the radial direction of each subsequent radial B-scan is incremented by a random angle, randomly drawn from a normal distribution over [0, ⁇ / ⁇ ], so that the maximal angular increment between adjacent radial B-scans is 9 max ⁇ 2 ⁇ / ⁇ .
  • FIG. 14 illustrates in its top row, a typical en-face slice image extracted from the full OCT data volume (f) showing the en-face slice represented in three frames overlaid with random radial subsampling masks ( ) corresponding to 200 radial scans (61.5% missing data compared to conventional raster scan) in the left frame, 100 radial scans (80.9% missing data) in the center frame, and 60 radial scans (88.3% missing data) in the right frame, according to an embodiment of the invention.
  • the yellow lines on the overlaid en-face slice images show locations of six exemplary extracted B-scans further shown in FIG. 14A to FIG. 14F.
  • the missing radial B-scan data was interpolated and reconstructed using the CS procedure described above in the previous section, including the mapping and reverse mapping steps described above in connection with FIG. 13.
  • an 1ST solver with 6 inner and 40 outer loop iterations was used to interpolate the sparse radial B-scan samples in the (r, ⁇ ) space.
  • FIG. 14A illustrates in a series of frames from left to right, the exemplary B-scan labeled "A" extracted from the full conventionally raster-acquired OCT volume (f) shown in the top row of FIG. 14; the same exemplary extracted B-scan corresponding to a random radial subsampling mask which discards 61.5% of the raster data (corresponding to 200 radial scans); a corresponding exemplary reconstructed B-Scan recovered from the remaining subsampled data in the 200 radial scan volume using compressive sampling; the exemplary B-scan corresponding to a random radial subsampling mask which discards 80.9% of the raster data (corresponding to 100 radial scans); a corresponding exemplary reconstructed B-Scan recovered from the remaining subsampled data in the 100 radial scan volume using compressive sampling; the exemplary B-scan corresponding to a random radial subsampling mask which discards 88.3% of the raster data (
  • FIG. 14B illustrates in a series of frames from left to right, the exemplary B-scan labeled "B" extracted from the full conventionally raster-acquired OCT volume (f); the same exemplary extracted B-scan corresponding to the random radial subsampling mask comprising 200 radial scans; a corresponding exemplary reconstructed B-Scan recovered from the remaining subsampled data in the 200 radial scan volume using compressive sampling; the exemplary B-scan corresponding to a random radial subsampling mask comprising 100 radial scans; a corresponding exemplary reconstructed B-Scan recovered from the remaining subsampled data in the 100 radial scan volume using compressive sampling; the exemplary B-scan corresponding to a random radial subsampling mask comprising 60 radial scans; and a corresponding exemplary reconstructed B-Scan recovered from the remaining subsampled data in the 60 radial scan volume using compressive sampling, according to an exemplary embodiment of
  • FIG. 14C illustrates in a series of frames from left to right, the exemplary B-scan labeled "C" extracted from the full conventionally raster-acquired OCT volume (f) shown in the top row of FIG. 14 and the corresponding subsampled and recovered B-scan images corresponding to 200, 100 and 60 radial scan volumes, similar to as described above in FIG.s 14A and 14B, according to an exemplary embodiment of the invention.
  • FIG. 14D illustrates in a series of frames from left to right, the exemplary B-scan labeled "D" extracted from the full conventionally raster-acquired OCT volume (f) shown in the top row of FIG. 14 and the corresponding subsampled and recovered B-scan images corresponding to 200, 100 and 60 radial scan volumes, similar to as described above in FIG.s 14A and 14B, according to an exemplary embodiment of the invention.
  • FIG. 14E illustrates in a series of frames from left to right, the exemplary B-scan labeled "E" extracted from the full conventionally raster-acquired OCT volume (f) shown in the top row of FIG. 14 and the corresponding subsampled and recovered B-scan images corresponding to 200, 100 and 60 radial scan volumes, similar to as described above in FlG.s 14A and 14B, according to an exemplary embodiment of the invention.
  • FIG. 14F illustrates in a series of frames from left to right, the exemplary B-scan labeled "F" extracted from the full conventionally raster-acquired OCT volume shown in the top row of FIG. 14 and the corresponding subsampled and recovered B-scan images corresponding to 200, 100 and 60 radial scan volumes, similar to as described above in FlG.s 14A and 14B, according to an exemplary embodiment of the invention.
  • the Inner Limiting Membrane (ILM) of the original raster-scanned volume was independently manually segmented three times and the standard deviation of the segmentation at each point on the resulting ILM surface was computed. Subsequently, the ILM on each of the recovered surfaces was segmented by the same internal segmentation protocol and the points on each ILM surface lying more than one standard deviation away from the mean ILM surface were located. Points on each of the three degraded and recovered ILM surfaces lying more than one standard deviation from the mean ILM surface were shown in a Topographical Change Analysis (TCA) plot overlaid on top of the summed- voxels-projection images of the recovered volume.
  • TCA Topographical Change Analysis
  • FIG. 15A illustrates a topographical Change Analysis (TCA) map showing the position of an ILM surface relative to the mean overlaid on top of summed-voxels- projection images of a recovered OCT volume at 61.5% missing data corresponding to a 200 radial B-scan volume, according to an embodiment of the invention.
  • TCA topographical Change Analysis
  • FIG. 15B illustrates a topographical Change Analysis (TCA) map showing the position of an ILM surface relative to the mean overlaid on top of summed-voxels- projection images of a recovered OCT volume at 80.9% missing data corresponding to a 100 radial B-scan volume, according to an embodiment of the invention.
  • TCA topographical Change Analysis
  • FIG. 15C illustrates a topographical Change Analysis (TCA) map showing the position of an ILM surface relative to the mean overlaid on top of summed-voxels- projection images of a recovered OCT volume at 88.3% missing data corresponding to a 60 radial B-scan volume, according to an embodiment of the invention.
  • TCA topographical Change Analysis
  • the most predominant topographical changes occur in the regions of the recovered image where the ILM exhibits rapid vertical changes in elevation.
  • the recovered volume with 60 radial B-scans less than 9% of the points in the segmented ILM surface were farther than one standard deviation from the mean ILM surface of the original full sampled volume.
  • the advantageous reduction in scan time required to acquire a subsampled random radial OCT B-scan volume according to the method described above according to a radial scanning embodiment of the invention, relative to the scan time required to acquire a full raster scanned volume of the same subject, is directly proportional to the number of radial B-scans acquired according to the exemplary embodiment of the invention compared to the number of original raster scans required in the raster scanned volume.
  • the percentage reduction in scan time can be calculated by the formula (1 — n/N) where n is the number of acquired radial B-scans and N is the number of original raster scans.
  • the embodiment consisted of 382 raster scans, so the subsampled radial scan volume consisting of 60 radial scans corresponds to an 82.3% reduction in scan time.
  • the method described above using a random radial subsampling pattern for acquiring a reduced set of radial B-scans and use of radial to Cartesian coordinate mapping and reverse transformation in connection with CS interpolation and reconstruction of missing data may also be applied to radial imaging of other non-OCT image acquisition systems, such as ultrasound, magnetic resonance (M 1) and x-ray (including Computed Tomography) imaging systems, for example.
  • non-OCT image acquisition systems such as ultrasound, magnetic resonance (M 1) and x-ray (including Computed Tomography) imaging systems, for example.
  • the method of radial subsampling according to an embodiment of the invention may also be advantageously be applied to imaging of particular subject volumes where the critical or desired features to be imaged may be concentrated in one area which may be used to center a random radial scanning pattern according to the invention, so as to advantageously concentrate sampling data acquired by random radial scans of the central area and associate features, while reducing sampling of features distant from the center of the radial scan pattern, for example.
  • random radial scan and CS image acquisition methods may also advantageously be used in applications where it is desired to reduce the exposure to imaging radiation, such as in x-ray based imaging systems, for example, where the reduced number of scans and scan time offered by the present inventive methods may significantly reduce the total exposure of a subject to imaging radiation, for example.
  • the above-described random radial scanning methods using CS may also be applied to scanning subject volumes where radial symmetry or access to the subject are limited to or optimized for radial scanning over raster imaging patterns, such as in trans-rectal imaging of the prostate using ultrasound, for example, where a point source imager may be rotated through a radial scan pattern advantageously over a raster scan pattern, for example.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Ophthalmology & Optometry (AREA)
  • Mathematical Physics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Algebra (AREA)
  • Radiology & Medical Imaging (AREA)
  • Discrete Mathematics (AREA)
  • Biophysics (AREA)
  • Mathematical Optimization (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

L'invention porte sur un procédé d'acquisition rapide d'image OCT consistant à acquérir par OCT une pluralité de mesures compressives (y) représentant un ensemble de données OCT sous-échantillonnées dans un domaine de Dirac au-dessous d'une fréquence de Nyquist par échantillonnage d'un objet d'intérêt au niveau de lignes verticales et horizontales aléatoirement espacées dans une géométrie cartésienne à l'aide d'un balayage récurrent, et reconstituer une image OCT volumétrique 3D (f) à partir des mesures compressives (y) à l'aide d'un échantillonnage compressif. Le procédé peut également consister à reconstituer l'image OCT volumétrique 3D (f) à partir des mesures compressives (y) sur la base au moins en partie d'une matrice de réduction de densité (S) apte à transformer l'image OCT volumétrique 3D (f) en une représentation creuse, telle qu'une représentation matricielle de l'image OCT volumétrique 3D (f) dans un domaine des transformées en ondelettes invariantes par décalage. Le procédé peut également être appliqué à des motifs de balayage OCT radial.
PCT/CA2011/050532 2010-08-31 2011-08-31 Système et procédé d'acquisition rapide d'image oct à l'aide d'un échantillonnage compressif WO2012027849A1 (fr)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US13/819,713 US9014451B2 (en) 2010-08-31 2011-08-31 System and method for rapid OCT image acquisition using compressive sampling
CA2844087A CA2844087C (fr) 2010-08-31 2011-08-31 Systeme et procede d'acquisition rapide d'image oct a l'aide d'un echantillonnage compressif

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
US37893110P 2010-08-31 2010-08-31
US61/378,931 2010-08-31
CA2,737,822 2011-04-21
CA2737822A CA2737822C (fr) 2010-08-31 2011-04-21 Systeme et procede d'acquisition rapide d'images de tomographie par coherence optique (oct) au moyen d'un echantillonnage compressif

Publications (1)

Publication Number Publication Date
WO2012027849A1 true WO2012027849A1 (fr) 2012-03-08

Family

ID=45772054

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CA2011/050532 WO2012027849A1 (fr) 2010-08-31 2011-08-31 Système et procédé d'acquisition rapide d'image oct à l'aide d'un échantillonnage compressif

Country Status (1)

Country Link
WO (1) WO2012027849A1 (fr)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103914823A (zh) * 2012-12-31 2014-07-09 复旦大学 基于稀疏表示的快速精确非线性配准立体医学影像的方法
WO2014071221A3 (fr) * 2012-11-02 2014-07-10 Optimedica Corporation Identification de surface optique pour chirurgie au laser
CN104220007A (zh) * 2012-04-10 2014-12-17 株式会社国际电气通信基础技术研究所 物体观测装置、物体观测方法以及记录介质
US9014451B2 (en) 2010-08-31 2015-04-21 Simon Fraser University System and method for rapid OCT image acquisition using compressive sampling
US9299139B2 (en) 2012-03-26 2016-03-29 The Cleveland Clinic Foundation Volumetric analysis of pathologies
JP2019107569A (ja) * 2014-07-30 2019-07-04 興和株式会社 眼底撮影装置

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
E. CANDES ET AL.: "Robust Uncertainty Principles: Exact Signal Reconstruction from Highly Incomplete Frequency Information", IEEE TRANS. INFORM. THEORY, vol. 52, 2006, pages 489 - 509 *
E. CANDES: "Compressive Sampling", PROCEEDINGS OF THE INTERNATIONAL CONGRESS OF MATHEMATICIANS, 2006, MADRID, SPAIN *

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9014451B2 (en) 2010-08-31 2015-04-21 Simon Fraser University System and method for rapid OCT image acquisition using compressive sampling
US9299139B2 (en) 2012-03-26 2016-03-29 The Cleveland Clinic Foundation Volumetric analysis of pathologies
CN104220007A (zh) * 2012-04-10 2014-12-17 株式会社国际电气通信基础技术研究所 物体观测装置、物体观测方法以及记录介质
US9297752B2 (en) 2012-04-10 2016-03-29 Advanced Telecommunications Research Institute International Object observing apparatus, object observing method, and storage medium
WO2014071221A3 (fr) * 2012-11-02 2014-07-10 Optimedica Corporation Identification de surface optique pour chirurgie au laser
US9549670B2 (en) 2012-11-02 2017-01-24 Optimedica Corporation Optical surface identification for laser surgery
US10441464B2 (en) 2012-11-02 2019-10-15 Optimedica Corporation Optical surface identification for laser eye surgery
US11540945B2 (en) 2012-11-02 2023-01-03 Amo Development, Llc Optical surface identification for laser eye surgery
US11752037B2 (en) 2012-11-02 2023-09-12 Amo Development, Llc Optical surface identification for laser eye surgery
CN103914823A (zh) * 2012-12-31 2014-07-09 复旦大学 基于稀疏表示的快速精确非线性配准立体医学影像的方法
JP2019107569A (ja) * 2014-07-30 2019-07-04 興和株式会社 眼底撮影装置

Similar Documents

Publication Publication Date Title
US9014451B2 (en) System and method for rapid OCT image acquisition using compressive sampling
US11170258B2 (en) Reducing noise in an image
US11120582B2 (en) Unified dual-domain network for medical image formation, recovery, and analysis
Arridge et al. Accelerated high-resolution photoacoustic tomography via compressed sensing
US8611626B2 (en) System and methods for fast implementation of equally-sloped tomography
US8699773B2 (en) Method for image reconstruction using low-dimensional-structure self-learning and thresholding
Fang et al. Fast acquisition and reconstruction of optical coherence tomography images via sparse representation
JP6731919B2 (ja) 光コヒーレンス断層撮影像におけるスペックル低減
WO2012027849A1 (fr) Système et procédé d'acquisition rapide d'image oct à l'aide d'un échantillonnage compressif
Shi et al. DeSpecNet: a CNN-based method for speckle reduction in retinal optical coherence tomography images
US8565502B2 (en) Method and system for reconstruction of tomographic images
US20120308100A1 (en) Method and system for reconstruction of tomographic images
CN111260748B (zh) 基于神经网络的数字合成x射线层析成像方法
JP2004181243A (ja) 横方向のフィルタリング処理を用いたトモシンセシス画像を強調する方法及びシステム
Bai et al. Super-resolution reconstruction of MR brain images
Bousse et al. Systematic review on learning-based spectral CT
Gunduzalp et al. 3d u-netr: Low dose computed tomography reconstruction via deep learning and 3 dimensional convolutions
Lebed et al. Rapid radial optical coherence tomography image acquisition
Khodajou-Chokami et al. PARS-NET: a novel deep learning framework using parallel residual conventional neural networks for sparse-view CT reconstruction
Zarei et al. Harmonizing CT images via physics-based deep neural networks
CN111369638A (zh) 激光反射层析成像欠采样的重建方法、存储介质和系统
Zhu et al. CT image reconstruction from partial angular measurements via compressed sensing
Wong et al. Phase-adaptive superresolution of mammographic images using complex wavelets
Dehner et al. DeepMB: Deep neural network for real-time optoacoustic image reconstruction with adjustable speed of sound
EP4207076A1 (fr) Traitement d'images par apprentissage machine indépendant du filtre de reconstruction

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 11820972

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 13819713

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 11820972

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2844087

Country of ref document: CA