WO2022007352A1 - 一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置 - Google Patents

一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置 Download PDF

Info

Publication number
WO2022007352A1
WO2022007352A1 PCT/CN2020/139962 CN2020139962W WO2022007352A1 WO 2022007352 A1 WO2022007352 A1 WO 2022007352A1 CN 2020139962 W CN2020139962 W CN 2020139962W WO 2022007352 A1 WO2022007352 A1 WO 2022007352A1
Authority
WO
WIPO (PCT)
Prior art keywords
choroidal
image
oct
choroid
dimensional
Prior art date
Application number
PCT/CN2020/139962
Other languages
English (en)
French (fr)
Inventor
周翔天
沈梅晓
吴昊
陈思思
金梓
Original Assignee
温州医科大学
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
Application filed by 温州医科大学 filed Critical 温州医科大学
Priority to AU2020458040A priority Critical patent/AU2020458040B2/en
Publication of WO2022007352A1 publication Critical patent/WO2022007352A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • 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]
    • 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
    • A61B3/12—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes
    • A61B3/1225—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes using coherent radiation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/12—Edge-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10101—Optical tomography; Optical coherence tomography [OCT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20048—Transform domain processing
    • G06T2207/20056—Discrete and fast Fourier transform, [DFT, FFT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30041—Eye; Retina; Ophthalmic
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30101—Blood vessel; Artery; Vein; Vascular

Definitions

  • the invention relates to the technical field of OCT, in particular to a choroidal three-dimensional blood vessel imaging and quantitative analysis method and device based on an optical coherence tomography system.
  • the choroid is a highly vascularized and pigment-rich tissue located between the retina and sclera.
  • the main function is to provide oxygen and nutrients to the RPE and the outer layer of the retina.
  • the oxygen transported accounts for about 90% of the retinal oxygen consumption and is necessary to maintain the high metabolic activity of photoreceptor cells in the outer layer of the retina; especially in the fovea of the macula without blood vessels
  • the choroid is the only way of its material exchange. Many diseases are closely related to the abnormal vascular structure of the choroid, such as age-related macular degeneration, high myopia macular degeneration, diabetic retinopathy and so on. Therefore, it is of great significance to realize the quantitative analysis of choroidal vessels.
  • Frequency domain optical coherence tomography technology can realize three-dimensional imaging of fundus. After averaging image enhancement technology or using high-penetration swept frequency light source, it can display the complete choroidal vascular tissue and stromal tissue, which is very beneficial for choroidal vascular analysis. imaging technology.
  • the current system does not solve the problem of poor signal-to-noise ratio of blood vessels in the deep choroid caused by backscatter attenuation, which leads to the difficulty of segmentation of blood vessel boundaries. For this reason, most of the quantitative choroidal blood vessel analysis algorithms in OCT images only use traditional image processing methods. However, because the choroid is located below the RPE, the light passes through the RPE and is attenuated by pigment absorption.
  • Attenuation of the light signal occurred at the depth of the choroid, resulting in blurring of the border between the choroid and the pigment epithelium and the border between the choroid and the sclera.
  • the contrast with the inner and outer boundaries is low, and the boundary continuity is poor, which makes it difficult for the traditional segmentation method to use the shortest path graph theory algorithm to achieve automatic boundary segmentation. Therefore, the accuracy of automatic differentiation is low, and manual correction after a large number of boundary detections is required, which is time-consuming.
  • the previous method to separate the choroidal blood vessels and stroma based on the fixed threshold method has great limitations in theory. Poor adaptability.
  • the present invention provides a choroid three-dimensional vascular imaging and quantitative analysis method and device based on an optical coherence tomography system, which is suitable for all OCT systems that have high penetration to the tissue and can obtain the choroid. And its images are automatic, universal, and can reflect the abnormality of choroidal three-dimensional blood vessels with high precision.
  • the technical solution adopted in the present invention is: a choroidal three-dimensional blood vessel imaging and quantitative analysis method based on an optical coherence tomography system, comprising the following steps:
  • the described step (2) intelligently divides the inner and outer boundaries of the choroid based on deep learning, including the following steps:
  • the upper and lower boundaries of the choroid in the images are accurately depicted; the labeled choroid image set is randomly divided into training according to 8:2 set and test set;
  • the training optimization algorithm is set to stochastic gradient descent (SGD), the algorithm learning ratio is 1.0e-5, the iterative momentum is 0.9, the iterative cost function is the Dice coefficient, the iteration The number of traversals is 150, and the batch size is 8.
  • SGD stochastic gradient descent
  • the algorithm learning ratio is 1.0e-5
  • the iterative momentum is 0.9
  • the iterative cost function is the Dice coefficient
  • the iteration The number of traversals
  • the batch size is 8.
  • necessary enhancement processing is performed on the input image to improve the robustness of the model.
  • the calculation formula of the Dice coefficient is:
  • X represents the choroidal boundary prediction set
  • Y represents the choroidal boundary annotation set
  • represents the intersection or overlap between the two sets
  • represents the total amount of both sets .
  • the step (3) of separating the choroidal vessels and non-vessels by self-adaptive threshold includes the following steps:
  • a square box with a length and width of 2*w+1 pixel blocks is used as a local window.
  • w is a positive integer that does not exceed half of the length and width of the image
  • the coordinates (x, y) are the geometric center of the square frame
  • the brightness information of all pixel blocks in the frame is counted to obtain the mean value m(x, y) and variance s(x,y).
  • the threshold value T(x, y) in the frame can be obtained, and the calculation formula is as follows:
  • all pixel blocks in the frame can be binarized, as shown below:
  • i and j are the relative coordinates representing the relative geometric center (x, y) of the pixel block
  • a x+i, y+j represents the brightness of the pixel block with coordinates (x+i, y+j)
  • a x+i, y +j represents the binarized data of the coordinate (x+i, y+j) pixel block;
  • Moving the local windows in turn can realize the binarization processing of the brightness matrix A of all pixel blocks in the OCT image, and obtain the binarization matrix B.
  • Described step (4) three-dimensional global and each area quantitative index comprises the following steps:
  • Image acquisition take the fovea of the fundus as the center, perform imaging in radial scanning mode, and obtain m pieces of two-dimensional cross-sectional data of the choroid;
  • Image registration By horizontally translating the position of the fovea, the abscissa of the fovea in all OCT images is the same in the spatial coordinates, so as to achieve image registration;
  • iii Three-dimensional spatial reconstruction of the choroid: Based on the upper and lower bounds of the choroid obtained by the deep learning neural network model, the choroid region template M(x, y, z) in the registered image is extracted, and combined with the adaptive threshold method, the blood vessels and the stroma are automatically separated. , get the binarized three-dimensional space choroidal structure matrix V(x,y,z); iiiiii, index establishment and testing: we establish choroidal vascular volume CVV, choroidal non-vascular volume SV, choroidal vascular index CVI and choroidal ischemia index CII and other indicators to quantify. The calculation formula of each indicator is as follows:
  • P x , P y , and P z represent the physical geometric lengths of voxels along the x, y and z axes in three-dimensional space, respectively.
  • n, m and k represent the number of pixels along the x, y and z axes in the three-dimensional space matrix, respectively.
  • Described step (1) obtains the OCT image of the choroid after image signal preprocessing, comprises the following steps:
  • Reverse attenuation compensation choroidal signal By extracting the principle and law of choroidal scattered light attenuation, and constructing a signal compensation and enhancement algorithm, the visualization and contrast of choroidal images can be improved.
  • the photoelectric signal of the interference between the reference arm and the sample arm can be expressed by the following formula:
  • k is the wave number, wave number of signal acquisition is divided into equally spaced m
  • [rho] is the photoelectric conversion efficiency OCT probe
  • S [k m] refers to the wavelength band corresponding radiant energy source
  • R R and R S are the reflectances of the reference arm and the sample arm, respectively;
  • the reflectivity profile function S(z) along the depth direction can be obtained by performing inverse discrete Fourier transform on the above formula (2):
  • Attenuation correction compensation can be performed on the OCT signal of each pixel:
  • N is the number of pixels in the A scan
  • ⁇ can be adjusted according to the tissue.
  • the premise of the above formula (3) is to assume that most of the beam energy is attenuated within the imaging depth range, and the attenuation outside the imaging depth range can be ignored.
  • the signal intensity of each pixel is:
  • S ac (z) is the signal after attenuation correction.
  • a choroidal three-dimensional blood vessel imaging and quantitative analysis device based on an optical coherence tomography system comprising the following modules:
  • Choroidal OCT image acquisition and image signal preprocessing module The deep choroidal blood vessels and stroma acquired by OCT are reversely compensated due to backscattering attenuation signals to enhance the signal-to-noise ratio of choroidal blood vessels; the enhanced OCT choroidal image signals are obtained;
  • Intelligent segmentation module of choroid inner and outer boundary based on deep learning using the segmentation method based on deep learning, intelligently segment the boundary between choroid and retinal epithelial layer and between choroid and sclera;
  • Choroidal vessel and non-vascular adaptive threshold segmentation module using adaptive threshold segmentation method, automatically separates three-dimensional choroidal vessels and non-vascular tissue;
  • Three-dimensional global and regional quantitative index module According to the distribution and proportion of blood vessels in the three-dimensional volume space in the image, the global and regional quantitative indicators that can characterize choroidal ischemia are calculated.
  • Described choroid OCT image acquisition and image signal preprocessing module include following algorithm model:
  • Reverse attenuation compensation for choroidal signals By extracting the principle and law of choroidal scattered light attenuation, an attenuation correction processing algorithm for OCT signals is constructed, which can improve the visualization and contrast of choroidal images.
  • the attenuation correction processing algorithm for OCT signals includes two steps, respectively.
  • the photoelectric signal of the interference between the reference arm and the sample arm can be expressed by the following formula:
  • k is the wave number, wave number of signal acquisition is divided into equally spaced m
  • [rho] is the photoelectric conversion efficiency OCT probe
  • S [k m] refers to the wavelength band corresponding radiant energy source
  • R R and R S are the reflectances of the reference arm and the sample arm, respectively;
  • the reflectivity profile function S(z) along the depth direction can be obtained by performing inverse discrete Fourier transform on the above formula (2):
  • Attenuation correction compensation can be performed on the OCT signal of each pixel:
  • N is the number of pixels in the A scan
  • ⁇ can be adjusted according to the tissue.
  • the premise of the above formula (3) is to assume that most of the beam energy is attenuated within the imaging depth range, and the attenuation outside the imaging depth range can be ignored.
  • the signal intensity of each pixel is:
  • S ac (z) is the signal after attenuation correction.
  • the described choroid inner and outer boundary intelligent segmentation module based on deep learning includes the following algorithm models:
  • the marked choroid image set is randomly divided into training set and test set according to 8:2 part;
  • the training optimization algorithm is set to stochastic gradient descent (SGD), the algorithm learning ratio is 1.0e-5, the iterative momentum is 0.9, the iterative cost function is the Dice coefficient, the iteration The number of traversals is 150, and the batch size is 8.
  • SGD stochastic gradient descent
  • the algorithm learning ratio is 1.0e-5
  • the iterative momentum is 0.9
  • the iterative cost function is the Dice coefficient
  • the iteration The number of traversals
  • the batch size is 8.
  • necessary enhancement processing is performed on the input image to improve the robustness of the model.
  • the calculation formula of the Dice coefficient is:
  • X represents the choroidal boundary prediction set
  • Y represents the choroidal boundary annotation set
  • represents the intersection or overlap between the two sets
  • represents the total amount of both sets .
  • the described choroidal vessel and non-vessel adaptive threshold segmentation module includes the following algorithm models:
  • i and j are the relative coordinates representing the relative geometric center (x, y) of the pixel block
  • a x+i, y+j represents the brightness of the pixel block with coordinates (x+i, y+j)
  • a x+i, y +j represents the binarized data of the coordinate (x+i, y+j) pixel block;
  • the three-dimensional global and each regional quantitative index module includes the following algorithm models:
  • Image acquisition take the fovea of the fundus as the center, perform imaging in radial scanning mode, and obtain m pieces of two-dimensional cross-sectional data of the choroid;
  • Marking of feature points Mark the fovea as the feature point of the OCT image for subsequent image registration
  • Image registration By horizontally translating the position of the fovea, the abscissa of the fovea in all OCT images is the same in the spatial coordinates, so as to achieve image registration;
  • iii Three-dimensional spatial reconstruction of the choroid: Based on the upper and lower bounds of the choroid obtained by the deep learning neural network model, the choroid region template M(x, y, z) in the registered image is extracted, and combined with the adaptive threshold method, the blood vessels and the stroma are automatically separated. , get the binarized three-dimensional space choroidal structure matrix V(x,y,z); iiiiii, index establishment and testing: we establish choroidal vascular volume CVV, choroidal non-vascular volume SV, choroidal vascular index CVI and choroidal ischemia index CII and other indicators are quantified, and the calculation formula of each indicator is as follows:
  • P x , P y , and P z represent the physical geometric lengths of the voxel along the x, y and z axes in the three-dimensional space, respectively, and n, m and k represent the number of pixels along the x, y and z axes in the three-dimensional space matrix, respectively .
  • the present invention provides a method and device for choroidal three-dimensional vascular imaging and quantitative analysis based on an optical coherence tomography system. After acquiring the enhanced OCT choroidal image signal, a segmentation method based on deep learning is further adopted.
  • the threshold segmentation method automatically separates the three-dimensional choroidal blood vessels and non-vascular tissues, and calculates the global and regional quantitative indicators that can characterize choroidal ischemia according to the distribution and proportion of blood vessels in the three-dimensional volume space in the image. This method is applicable to all OCT imaging systems capable of acquiring choroidal vascular and non-vascular signals with high tissue penetration.
  • FIG. 1 is a flow chart of the technical solution of the present invention.
  • Figure 2 shows OCT fundus image acquisition and preprocessing.
  • Figure 3 is a schematic diagram of choroidal signal attenuation correction and image contrast de-enhancement; the left image is the original image, and the right image is the image after attenuation correction to remove artifacts.
  • Figure 4 shows the comparison between the traditional dynamic programming algorithm and the deep learning automatic segmentation of the inner and outer boundaries of the choroid; the left picture is the result of the traditional algorithm, and the right picture is the result of the deep learning algorithm.
  • Figure 5 shows the self-adaptive threshold to separate the choroidal vessels and non-vessels; the upper image is a high myopia, and the lower image is an emmetropic eye.
  • Figure 6 is a diagram of the choroidal blood vessels after three-dimensional reconstruction.
  • the fundus images were obtained using current commercial instruments or self-built OCT, and the images were preprocessed including appropriate cropping, and the OCT choroid intensity map was retained, as shown in Figure 2.
  • the signals received by the OCT detector are backscattered and reflected signals. Due to the influence of RPE and choroid's own pigment on light absorption, the light scattering of a certain wavelength is lost. By extracting the principle and law of choroid scattered light attenuation, and constructing a signal compensation and enhancement algorithm, the visualization and contrast of choroid images can be improved.
  • the attenuation correction processing algorithm of OCT signal includes two steps, which are the attenuation compensation of light and the contrast of image enhancement.
  • the photoelectric signal of the interference between the reference arm and the sample arm can be expressed by the following formula:
  • Equation (1) where k is the wave number, wave number of signal acquisition is divided into equally spaced m, [rho] is the photoelectric conversion efficiency OCT probe, S [k m] refers to the wavelength band corresponding radiant energy source, is [Delta] x between the reference arm and the sample arm.
  • the optical path difference, R R and R S are the reflectances of the reference arm and the sample arm, respectively.
  • the interaction term H[k m ] between the reference arm and the sample arm can be obtained from equation (1):
  • the reflectivity profile function S(z) along the depth direction can be obtained by performing the inverse discrete Fourier transform on equation (2):
  • the attenuated signal DA is obtained after OCT direct data processing, not the original signal S(z). This is the main reason why shadows appear in strongly attenuated tissues such as blood vessels and pigments, that is, the reason for the appearance of artifacts.
  • the attenuation term needs to be removed, which also removes artifacts.
  • N is the number of pixels in the A scan
  • ⁇ can be adjusted according to the tissue.
  • the premise of the above formula is that most of the beam energy is attenuated within the imaging depth range, and the attenuation outside the imaging depth range is negligible.
  • the present invention will perform an exponentiation operation on the original signal after the attenuation correction and compensation to enhance the image contrast, and the signal intensity of each pixel is now:
  • S ac (z) is the signal after attenuation correction, as shown in FIG. 3 .
  • the suprachoroidal border was defined as the dividing line between Bruch's membrane and the retinal pigment epithelium (RPE), and the inferior choroidal border was defined as the dividing line between the choroid and the sclera. Since the RPE layer appears as a high signal band in the OCT image, based on the traditional shortest path graph theory algorithm, the automatic segmentation of the suprachoroidal border can be well achieved; however, the contrast of the dividing line between the choroid and the sclera on the OCT image is poor, It is difficult to realize the automatic segmentation of the lower boundary by the shortest path graph theory algorithm. As the most important breakthrough in the field of artificial intelligence, deep learning has made great breakthroughs in the field of computer vision. The efficiency of deep learning-based algorithms is significantly better than traditional algorithms.
  • the present invention will establish a deep learning segmentation model to realize the automatic segmentation of the upper and lower boundaries of the choroid, and the specific steps are as follows:
  • the upper and lower boundaries of the choroid in the images are accurately depicted; the labeled choroid image set is randomly divided into the training set according to 8:2 , two parts of the test set.
  • the training optimization algorithm is set to stochastic gradient descent (SGD), the algorithm learning ratio is 1.0e-5, the iterative momentum is 0.9, and the iterative cost function is the Dice coefficient (dice coefficient). ), the number of iterations (epoch) is 150, and the batch size (batch size) is 8.
  • necessary enhancement processing such as translation, rotation and flipping, etc. is performed on the input image to improve the robustness of the model.
  • the calculation formula of Dice coefficient is:
  • X represents the choroidal boundary prediction set
  • Y represents the choroidal boundary annotation set
  • represents the intersection or overlap between the two sets
  • represents the total amount of both.
  • the choroidal vessels and stroma show different characteristics in OCT images, among which the vessels are dominated by low-intensity signals, while the stroma is dominated by high-intensity signals. Since the brightness of the choroidal signal in the OCT image is affected by related factors such as the RPE layer, the type of instrument, and the focusing situation during the operation, the separation of choroidal blood vessels and stroma based on the fixed threshold method has great limitations in theory, and the universality is poor. . In contrast, based on the adaptive threshold, the interference effect caused by the overall shift of the choroidal signal brightness can be better avoided, and the automatic separation of choroidal blood vessels and stroma can be better achieved.
  • the process of the method is as follows:
  • a square box with a length and width of 2*w+1 pixel blocks is used as a local window.
  • w is a positive integer that does not exceed half of the length and width of the image
  • the coordinates (x, y) are the geometric center of the square frame
  • the brightness information of all pixel blocks in the frame is counted to obtain the mean value m(x, y) and variance s(x,y).
  • the threshold value T(x, y) in the frame can be obtained, and the calculation formula is as follows:
  • i and j are the relative coordinates representing the relative geometric center (x, y) of the pixel block
  • Ax+i, y+j represents the brightness of the pixel block with coordinates (x+i, y+j)
  • Ax+i, y+j Binarized data representing a pixel block of coordinates (x+i, y+j).
  • the present invention will image the fundus through the radial scanning mode, obtain the choroid three-dimensional space data, register and reconstruct the image, and realize the choroid three-dimensional space reconstruction.
  • the fovea is used as the reference point, and the position of the fovea is horizontally shifted, so that the abscissa of the fovea in all OCT images is the same in space coordinates, and the three-dimensional image reconstruction is realized.
  • the distribution and proportion of choroidal ischemia were calculated, and the global and regional quantitative indicators that could characterize choroidal ischemia were calculated. Specific steps are as follows:
  • 1Image acquisition Take the fovea of the fundus as the center, perform imaging in radial scanning mode, and obtain m pieces of two-dimensional cross-sectional data of the choroid.
  • Feature point marking Since all OCT images contain the characteristic structure of the fovea, the present invention uses the fovea as the feature point of the OCT image for marking for subsequent image registration.
  • 3Image registration By translating the position of the fovea horizontally, the fovea in all OCT images has the same abscissa in the spatial coordinates to achieve image registration.
  • P x , P y , and P z represent the physical geometric lengths of voxels along the x, y and z axes in three-dimensional space, respectively.
  • n, m and k represent the number of pixels along the x, y and z axes in the three-dimensional space matrix, respectively.
  • the same operator repeatedly performs OCT imaging on the same subject twice to test the repeatability of the index, and compare it with the corresponding two-dimensional index.
  • the test results are shown in the following table.
  • the repeatability of the three-dimensional index is significantly better than that of the two-dimensional index.
  • the repeatability of vertical two-dimensional indicators is better than that of horizontal two-dimensional indicators.
  • the invention After preprocessing the choroidal image signal acquired by the system, the invention establishes a reverse signal attenuation compensation model based on the idea of reverse compensation of the OCT backscattered signal, reversely compensates the choroidal signal, and enhances the signal-to-noise ratio between blood vessels and non-vascular tissues. And further use the segmentation method based on deep learning to intelligently segment the inner and outer boundaries of the choroid; on the basis of boundary segmentation, an improved adaptive threshold method is further used to automatically separate the three-dimensional choroidal vessels and non-vascular tissues, and according to the blood vessels in the image. The distribution and proportion in the three-dimensional volume space, the global and regional quantitative indicators that can characterize choroidal ischemia are calculated. The method is suitable for all OCT systems and their images that can acquire the choroid with high penetration to the tissue.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Surgery (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Veterinary Medicine (AREA)
  • Public Health (AREA)
  • Ophthalmology & Optometry (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • Animal Behavior & Ethology (AREA)
  • Radiology & Medical Imaging (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Eye Examination Apparatus (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

一种基于光学相干断层扫描成像技术的脉络膜三维血管成像及定量化分析方法与装置，基于OCT后向散射信号逆向补偿的思想，将OCT获取的深部脉络膜血管及基质因后向散射衰减信号进行逆向补偿，增强脉络膜血管的信噪比；在获取增强的OCT脉络膜图像信号后，进一步采用基于深度学习的分割方法，对脉络膜与视网膜上皮层以及脉络膜与巩膜之间的边界进行智能化分割，克服传统算法在三维脉络膜边界分割的耗时、准确性低的缺点；在边界分割基础上，进一步采用改进的自适应阈值分割方法，自动分离出三维脉络膜血管与非血管组织，并根据图像中血管在三维体空间的分布及占比，计算出能够表征脉络膜缺血的全局和各区域定量化指标。

Description

一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置 技术领域
本发明涉及OCT技术领域，具体涉及一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置。
背景技术
脉络膜是位于视网膜和巩膜之间高度血管化且富含色素的组织。主要功能是向RPE和视网膜外层提供氧气和营养物质，其输送的氧气大约占视网膜氧耗的90％，是维持视网膜外层感光细胞高代谢活动所必须的；尤其在黄斑中心凹无血管区，脉络膜是其物质交换的唯一途径。很多疾病与脉络膜的血管结构异常密切相关，如年龄相关性黄斑病变、高度近视黄斑病变、糖尿病视网膜病变等。因此，实现脉络膜血管的定量分析具有重要的意义。
频域光学相干断层成像技术可实现眼底三维成像，经过平均化的图像增强技术或采用高穿透性的扫频光源，能够显示完整的脉络膜血管组织和基质组织，为脉络膜血管分析提供了很有利的影像技术。但目前的系统并没有解决后向散射衰减而导致的脉络膜深部血管信噪比差的问题，从而导致血管边界分割的困难。基于此原因，目前针对OCT图像中定量化脉络膜血管分析算法大部分也只是采用传统图像处理方法，但由于脉络膜位RPE下方，光经RPE之后因色素吸收衰减，同时脉络膜自身富含色素，光经脉络膜深度上发生了光信号的衰减，导致脉络膜与色素上皮层的边界及脉络膜与巩膜边界较模糊。此外，由于脉络膜内存在血管(低信号)与非血管(高信号)成分，与内外边界的对比度低、边界连续性差，导致传统的分割方法入最短路径图论算法较难实现边界的自动分割。因此，自动分化准确性低，需要大量的边界探测后手动校正，较耗时。由于脉络膜信号在OCT图像中的亮度受RPE层、OCT仪器种类和操作过程中聚焦情况等相关因素的影响，因此既往基于固定阈值法来分离脉络膜血管和基质在理论上存在较大局限性，普适性差。
发明内容
为了现有技术存在的问题，本发明提供了一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置，适用于对组织具有较高穿透的能够获取脉络膜的所有OCT系统及其图像，具有自动、普适性强、且能高精度反映脉络膜三维血管异常。
本发明采用的技术解决方案是：一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法，包括以下步骤：
(1)获得图像信号预处理后的脉络膜的OCT图像；
(2)基于深度学习智能化分割脉络膜内外边界；
(3)自适应阈值分离出脉络膜血管与非血管；
(4)三维全局和各区域定量化指标。
所述的步骤(2)基于深度学习智能化分割脉络膜内外边界，包括以下步骤：
a、基于传统的最短路径图论算法和对数千张OCT影像脉络膜的内外边界进行半自动标注，准确描绘出图像中脉络膜的上下边界；将已标注的脉络膜图像集按8：2随机分为训练集、测试集两个部分；
b、将训练图像输入开源的深度学习神经网络模型中，训练优化算法设为随机梯度下降法(SGD)，算法学习比率为1.0e-5，迭代动量为0.9，迭代代价函数为Dice系数，迭代遍历数为150，批样本量为8，此外，对输入图像进行必要的增强处理，改善模型的鲁棒性，其中，Dice系数计算公式为：
Figure PCTCN2020139962-appb-000001
其中X表示的是脉络膜边界预测集，Y表示的是脉络膜边界标注集，|X∩Y|表示两个集合之间的相交部分或重叠部分，|X|+|Y|表示两者的总量。Dice系数越大，表明两个集合相似度越高，模型越准确；当预测集与标注集完全相同时，Dice系数为1；当预测集与标注集不相关时，Dice系数为0。在模型训练 过程中，设置Dice系数大于0.95作为目标函数；
c、将测试集图像输入深度学习神经网络模型中，计算脉络膜边界的输出结果和标注集之间的Dice系数、边界误差来评估深度学习神经网络模型的分割性能。
所述的步骤(3)自适应阈值分离出脉络膜血管与非血管，包括以下步骤：
A、在OCT图像中，以长度和宽度都为2*w+1个像素块的正方形框作为局部窗口。w为最大不超过图像的长度和宽度的一半的正整数，坐标(x,y)为这个正方形框的几何中心，统计框内所有像素块的亮度信息，求得其均值m(x,y)和方差s(x,y)。根据参数k，可求得框内阈值T(x,y)，计算公式如下：
T(x,y)＝m(x,y)+k*s(x,y)；
B、根据框内阈值T(x,y)，可对框内所有像素块进行二值化处理，如下所示：
Figure PCTCN2020139962-appb-000002
其中i和j是表征像素块相对几何中心(x，y)的相对坐标，A x+i,y+j表示坐标(x+i,y+j)像素块的亮度，A x+i,y+j表示坐标(x+i,y+j)像素块的的二值化数据；
C、依次移动局部窗口，可以实现OCT图像中所有像素块亮度矩阵A的二值化处理，获得二值化矩阵B。
所述的步骤(4)三维全局和各区域定量化指标，包括以下步骤：
i、图像获取：以眼底黄斑中心凹为中心，通过放射状扫描模式进行成像，获取m张脉络膜二维横断面数据；
ii、特征点标记：由于所有OCT图像都包含黄斑中心凹这一特征结构，因此本发明以黄斑中心凹作为OCT图像的特征点进行标记，用于后续图像配准；
iii、图像配准：通过水平平移黄斑中心凹的位置，使所有OCT图像中黄斑中 心凹在空间坐标中横坐标相同，实现图像的配准；
iiii、脉络膜三维空间重建：基于深度学习神经网络模型所获得的脉络膜上下界，提取配准后的图像中脉络膜区域模板M(x,y,z)，并结合自适应阈值方法自动分离血管和基质，得到二值化后的三维空间脉络膜结构矩阵V(x,y,z)；iiiii、指标建立及测试：我们建立脉络膜血管体积CVV、脉络膜非血管体积SV、脉络膜血管指数CVI和脉络膜缺血指数CII等指标进行量化。各指标的计算公式如下所示：
Figure PCTCN2020139962-appb-000003
Figure PCTCN2020139962-appb-000004
Figure PCTCN2020139962-appb-000005
Figure PCTCN2020139962-appb-000006
其中P x,P y,P z分别表征在三维空间中体像素沿x，y和z轴的物理几何长度。n,m和k分别表示三维空间矩阵中沿x，y和z轴的像素数目。
所述的步骤(1)获得图像信号预处理后的脉络膜的OCT图像，包括以下步骤：
(一)数据的获取：利用OCT获取眼底图像，对图片进行预处理包括适当裁剪，保留OCT脉络膜强度图；
(二)逆向衰减补偿脉络膜信号：通过提取脉络膜散射光衰减原理和规律，构建信号补偿和增强算法，可提高脉络膜图像的可视化和对比度，OCT信号的衰减校正处理算法包含两个步骤，分别对光的衰减补偿和图像增强对比，在OCT 系统中，参考臂和样品臂间干涉的光电信号可以用如下公式表示：
Figure PCTCN2020139962-appb-000007
式中k为波数，采集信号被分割成m个等间距的波数，ρ是OCT探测器的光电转化效能，S[k m]指所对应波段光源的辐射能量，Δx是参考臂和样本臂间光程差，R R和R S分别为参考臂和样品臂的反射率；
参考臂和样品臂间的交互项H[k m]可由上式(1)得到:
Figure PCTCN2020139962-appb-000008
通过对上式(2)进行离散傅里叶逆变换可以得到沿深度方向的反射率剖面函数S(z):
Figure PCTCN2020139962-appb-000009
对公式进行离散化处理，可以对每个像素点的OCT信号进行衰减校正补偿：
Figure PCTCN2020139962-appb-000010
其中N是A扫描的像素点数，α可以根据组织进行调整，上述公式(3)成立的前提是假设大部分光束能量是在成像深度范围内衰减的，成像深度范围外的衰减可以忽略不计。
将对衰减校正补偿后的原始信号进行取幂运算增强图像对比度，此时每个像素信号强度为:
Figure PCTCN2020139962-appb-000011
式(5)中S ac(z)是衰减校正后的信号。
一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析装置，包括以下模块：
脉络膜OCT图像获取及图像信号预处理模块：OCT获取的深部脉络膜血 管及基质因后向散射衰减信号进行逆向补偿，增强脉络膜血管的信噪比；在获取增强的OCT脉络膜图像信号；
基于深度学习的脉络膜内外边界智能化分割模块：采用基于深度学习的分割方法，对脉络膜与视网膜上皮层以及脉络膜与巩膜之间的边界进行智能化分割；
脉络膜血管与非血管自适应阈值分割模块：采用自适应阈值分割方法，自动分离出三维脉络膜血管与非血管组织；
三维全局和各区域定量化指标模块：根据图像中血管在三维体空间的分布及占比，计算出能够表征脉络膜缺血的全局和各区域定量化指标。
所述的脉络膜OCT图像获取及图像信号预处理模块包括以下算法模型:
(一)数据的获取：利用OCT获取眼底图像，对图片进行适当裁剪，保留OCT脉络膜强度图；
(二)逆向衰减补偿脉络膜信号：通过提取脉络膜散射光衰减原理和规律，构建OCT信号的衰减校正处理算法，可提高脉络膜图像的可视化和对比度，OCT信号的衰减校正处理算法包含两个步骤，分别对光的衰减补偿和图像增强对比，在OCT系统中，参考臂和样品臂间干涉的光电信号可以用如下公式表示：
Figure PCTCN2020139962-appb-000012
式中k为波数，采集信号被分割成m个等间距的波数，ρ是OCT探测器的光电转化效能，S[k m]指所对应波段光源的辐射能量，Δx是参考臂和样本臂间光程差，R R和R S分别为参考臂和样品臂的反射率；
参考臂和样品臂间的交互项H[k m]可由上式(1)得到:
Figure PCTCN2020139962-appb-000013
通过对上式(2)进行离散傅里叶逆变换可以得到沿深度方向的反射率剖面函数S(z):
Figure PCTCN2020139962-appb-000014
对公式进行离散化处理，可以对每个像素点的OCT信号进行衰减校正补偿：
Figure PCTCN2020139962-appb-000015
其中N是A扫描的像素点数，α可以根据组织进行调整，上述公式(3)成立的前提是假设大部分光束能量是在成像深度范围内衰减的，成像深度范围外的衰减可以忽略不计。
将对衰减校正补偿后的原始信号进行取幂运算增强图像对比度，此时每个像素信号强度为:
Figure PCTCN2020139962-appb-000016
式(5)中S ac(z)是衰减校正后的信号。
所述的基于深度学习的脉络膜内外边界智能化分割模块包括以下算法模型:
a、基于最短路径图论算法和对OCT影像脉络膜的内外边界进行半自动标注，准确描绘出图像中脉络膜的上下边界；将已标注的脉络膜图像集按8：2随机分为训练集、测试集两个部分；
b、将训练图像输入开源的深度学习神经网络模型中，训练优化算法设为随机梯度下降法(SGD)，算法学习比率为1.0e-5，迭代动量为0.9，迭代代价函数为Dice系数，迭代遍历数为150，批样本量为8，此外，对输入图像进行必要的增强处理，改善模型的鲁棒性，其中，Dice系数计算公式为：
Figure PCTCN2020139962-appb-000017
其中X表示的是脉络膜边界预测集，Y表示的是脉络膜边界标注集，|X∩Y|表示两个集合之间的相交部分或重叠部分，|X|+|Y|表示两者的总量。Dice系数越大，表明两个集合相似度越高，模型越准确；当预测集与标注集完全相同 时，Dice系数为1；当预测集与标注集不相关时，Dice系数为0。在模型训练过程中，设置Dice系数大于0.95作为目标函数；
c、将测试集图像输入深度学习神经网络模型中，计算脉络膜边界的输出结果和标注集之间的Dice系数、边界误差来评估深度学习神经网络模型的分割性能。
所述的脉络膜血管与非血管自适应阈值分割模块包括以下算法模型:
A、在OCT图像中，以长度和宽度都为2*w+1个像素块的正方形框作为局部窗口，w为最大不超过图像的长度和宽度的一半的正整数，坐标(x,y)为这个正方形框的几何中心，统计框内所有像素块的亮度信息，求得其均值m(x,y)和方差s(x,y)，根据参数k，求得框内阈值T(x,y)，计算公式如下：
T(x,y)＝m(x,y)+k*s(x,y)；
B、根据框内阈值T(x,y)，对框内所有像素块进行二值化处理，如下所示：
Figure PCTCN2020139962-appb-000018
其中i和j是表征像素块相对几何中心(x，y)的相对坐标，A x+i,y+j表示坐标(x+i,y+j)像素块的亮度，A x+i,y+j表示坐标(x+i,y+j)像素块的的二值化数据；
C、依次移动局部窗口，实现OCT图像中所有像素块亮度矩阵A的二值化处理，获得二值化矩阵B。
所述的三维全局和各区域定量化指标模块包括以下算法模型:
i、图像获取：以眼底黄斑中心凹为中心，通过放射状扫描模式进行成像，获取m张脉络膜二维横断面数据；
ii、特征点标记：以黄斑中心凹作为OCT图像的特征点进行标记，用于后续图像配准；
iii、图像配准：通过水平平移黄斑中心凹的位置，使所有OCT图像中黄斑中 心凹在空间坐标中横坐标相同，实现图像的配准；
iiii、脉络膜三维空间重建：基于深度学习神经网络模型所获得的脉络膜上下界，提取配准后的图像中脉络膜区域模板M(x,y,z)，并结合自适应阈值方法自动分离血管和基质，得到二值化后的三维空间脉络膜结构矩阵V(x,y,z)；iiiii、指标建立及测试：我们建立脉络膜血管体积CVV、脉络膜非血管体积SV、脉络膜血管指数CVI和脉络膜缺血指数CII等指标进行量化，各指标的计算公式如下所示：
Figure PCTCN2020139962-appb-000019
Figure PCTCN2020139962-appb-000020
Figure PCTCN2020139962-appb-000021
Figure PCTCN2020139962-appb-000022
其中P x,P y,P z分别表征在三维空间中体像素沿x，y和z轴的物理几何长度，n,m和k分别表示三维空间矩阵中沿x，y和z轴的像素数目。
本发明的有益效果是：本发明提供了一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置，在获取增强的OCT脉络膜图像信号后，进一步采用基于深度学习的分割方法，对脉络膜与视网膜上皮层以及脉络膜与巩膜之间的边界进行智能化分割，克服传统算法在三维脉络膜边界分割的耗时、准确性低的缺点；在边界分割基础上，进一步采用改进的自适应阈值分割方法，自动分离出三维脉络膜血管与非血管组织，并根据图像中血管在三维体空间的分布及占比，计算出能够表征脉络膜缺血的全局和各区域定量化指 标。该方法适用于对组织具有较高穿透的能够获取脉络膜血管和非血管信号的所有OCT成像系统。
附图说明
图1为本发明技术方案流程图。
图2为OCT眼底图像获取及预处理。
图3为脉络膜信号衰减校正及图像对比去增强示意图；其中左图为原图，右图为衰减校正去除伪影后的图。
图4为传统动态规划算法和深度学习自动分割脉络膜内外边界对比；其中左图为传统算法结果，右图为深度学习算法结果。
图5为自适应阈值分离出脉络膜血管与非血管；其中上图为高度近视，下图为正视眼。
图6为三维重建后的脉络膜血管图。
具体实施方式
下面结合附图以及一些具体实施例，可更好的说明本发明。
(1)数据的获取和预处理：
利用目前商业仪器或自行搭建的OCT获取眼底图像，对图片进行预处理包括适当裁剪，保留OCT脉络膜强度图，如附图2所示。
(2)逆向衰减补偿脉络膜信号：
OCT探测器接收的信号为后向散射和反射的信号。由于RPE和脉络膜自身色素对光吸收的影响导致一定波长的光散射出现损耗，通过提取脉络膜散射光衰减原理和规律，构建信号补偿和增强算法，可提高脉络膜图像的可视化和对比度。
OCT信号的衰减校正处理算法包含两个步骤，分别对光的衰减补偿和图像增强对比。在OCT系统中，参考臂和样品臂间干涉的光电信号可以用如下公式表示：
Figure PCTCN2020139962-appb-000023
式中k为波数，采集信号被分割成m个等间距的波数，ρ是OCT探测器的光电转化效能，S[k m]指所对应波段光源的辐射能量，Δx是参考臂和样本臂间光程差，R R和R S分别为参考臂和样品臂的反射率。参考臂和样品臂间的交互项H[k m]可由式(1)得到:
Figure PCTCN2020139962-appb-000024
通过对式(2)进行离散傅里叶逆变换可以得到沿深度方向的反射率剖面函数S(z):
Figure PCTCN2020139962-appb-000025
而公式(1)-(3)成立的条件是假设光束在传播过程中没有发生能量衰减。但是光束在传播过程中会发生能量衰减，因此我们需要改进上述公式。当光束穿透组织样本时，一小部分转化为热量而其余部分散射，此时光束是以吸收的形式衰减。假设光束的局部衰减仅与光的散射相关，并以固定比例背向散射，因此可以计算在深度Z时光束的局部衰减正比于该位置的反射率R和背向散射常数α。
实际情况下，OCT直接数据处理后获得的是衰减后的信号DA，而不是原始信号S(z)。这就是暗影出现在血管和色素等强衰减组织的主要原因，即伪影出现的原因。为了矫正衰减信号，衰减项需要被消除，同时也去除了伪影。对公式进行离散化处理，可以对每个像素点的OCT信号进行衰减校正补偿:
Figure PCTCN2020139962-appb-000026
其中N是A扫描的像素点数，α可以根据组织进行调整。上述公式成立的前提是假设大部分光束能量是在成像深度范围内衰减的，成像深度范围外的衰减可以忽略不计。
后期，本发明将对衰减校正补偿后的原始信号进行取幂运算增强图像对比度，此时每个像素信号强度为:
Figure PCTCN2020139962-appb-000027
式中S ac(z)是衰减校正后的信号，如附图3所示。
(3)深度学习智能化分割：
脉络膜内外边界准确识别是准确量化脉络膜三维血管指标的重要步骤。
脉络膜上边界定义为Bruch’s膜和视网膜色素上皮层(retinal pigment epithelium，RPE)间的分界线，脉络膜下边界定义为脉络膜和巩膜之间的分界线。由于RPE层在OCT图像中表现为一条高信号带，基于传统的最短路径图论算法，可以很好地实现脉络膜上边界的自动分割；但是OCT图像上脉络膜和巩膜之间的分界线对比度差，通过最短路径图论算法较难实现下边界的自动分割。深度学习作为人工智能领域最重要的突破,在计算机视觉领域中取得了很大突破,基于深度学习的算法效率明显好于传统算法。本发明将建立深度学习分割模型来实现脉络膜上下边界的自动分割，具体步骤如下：
①基于传统的最短路径图论算法和对数千张OCT影像脉络膜的内外边界进行半自动标注，准确描绘出图像中脉络膜的上下边界；将已标注的脉络膜图像集按8：2随机分为训练集、测试集两个部分。
②将训练图像输入开源的深度学习神经网络模型中，训练优化算法设为随机梯度下降法(SGD)，算法学习比率为1.0e-5，迭代动量为0.9，迭代代价函数为Dice系数(dice coefficient)，迭代遍历数(epoch)为150，批样本量(batch size)为8。此外，对输入图像进行必要的增强处理(如平移、旋转和翻转等)，改善模型的鲁棒性。其中，Dice系数计算公式为：
Figure PCTCN2020139962-appb-000028
X表示的是脉络膜边界预测集，Y表示的是脉络膜边界标注集，|X∩Y|表示两个集合之间的相交部分或重叠部分，|X|+|Y|表示两者的总量。Dice系数越大，表明两个集合相似度越高，模型越准确；当预测集与标注集完全相同时， Dice系数为1；当预测集与标注集不相关时，Dice系数为0。在模型训练过程中，设置Dice系数大于0.95作为目标函数。
②将测试集图像输入深度学习神经网络模型中，计算脉络膜边界的输出结果和标注集之间的Dice系数、边界误差来评估深度学习神经网络模型的分割性能。结果如附图4和表1所示。
③表1深度学习分割OCT图像脉络膜边界的效能
Figure PCTCN2020139962-appb-000029
(4)自适应阈值分离出脉络膜血管与非血管：脉络膜血管和基质在OCT图像中表现出不同的特征，其中血管以低亮度信号为主，而基质以高亮度信号为主。由于脉络膜信号在OCT图像中的亮度受RPE层、仪器种类和操作过程中聚焦情况等相关因素的影响，因此基于固定阈值法来分离脉络膜血管和基质在理论上存在较大局限性，普适性差。相对地，基于自适应阈值可以较好地避免脉络膜信号亮度整体的偏移所带来的干扰作用，可以较好地实现脉络膜血管和基质的自动分离，该方法流程如下：
①在OCT图像中，以长度和宽度都为2*w+1个像素块的正方形框作为局部窗口。w为最大不超过图像的长度和宽度的一半的正整数，坐标(x,y)为这个正方形框的几何中心，统计框内所有像素块的亮度信息，求得其均值m(x,y)和方差s(x,y)。根据参数k，可求得框内阈值T(x,y)，计算公式如下：
T(x,y)＝m(x,y)+k*s(x,y)   (7)
②根据框内阈值T(x,y)，可对框内所有像素块进行二值化处理，如下所示：
Figure PCTCN2020139962-appb-000030
其中i和j是表征像素块相对几何中心(x，y)的相对坐标，Ax+i,y+j表示坐标(x+i,y+j)像素块的亮度，Ax+i,y+j表示坐标(x+i,y+j)像素块的的二值化数据。
③依次移动局部窗口，可以实现OCT图像中所有像素块亮度矩阵A的二值化处理，获得二值化矩阵B。结果如图5所示。
(5)三维全局和各区域定量化指标：
基于单张OCT图像仅能获取脉络膜某个横断面上的特征信息，但是由于某些脉络膜病变空间分布不均匀，因此通过单张OCT图像分析模式难以精确评估疾病的病情进展。本发明将通过放射状扫描模式对眼底进行成像，获取脉络膜三维空间数据，并对图像进行配准及重建，实现脉络膜三维空间重建。OCT图像中，以黄斑中心凹作为参考点，通过水平平移黄斑中心凹的位置，使所有OCT图像中黄斑中心凹在空间坐标中横坐标相同，实现图像三维重建，根据图像中血管在三维体空间的分布及占比，计算出能够表征脉络膜缺血的全局和各区域定量化指标。具体步骤如下：
①图像获取：以眼底黄斑中心凹为中心，通过放射状扫描模式进行成像，获取m张脉络膜二维横断面数据。
②特征点标记：由于所有OCT图像都包含黄斑中心凹这一特征结构，因此本发明以黄斑中心凹作为OCT图像的特征点进行标记，用于后续图像配准。
③图像配准：通过水平平移黄斑中心凹的位置，使所有OCT图像中黄斑中心凹在空间坐标中横坐标相同，实现图像的配准。
④脉络膜三维空间重建：基于深度学习神经网络模型所获得的脉络膜上下界，提取配准后的图像中脉络膜区域模板M(x,y,z)，并结合自适应阈值方法自动分离血管和基质，得到二值化后的三维空间脉络膜结构矩阵V(x,y,z)。
⑤指标建立及测试：我们建立脉络膜血管体积CVV、脉络膜非血管体积 SV、脉络膜血管指数CVI和脉络膜缺血指数CII等指标进行量化。各指标的计算公式如下所示：
Figure PCTCN2020139962-appb-000031
Figure PCTCN2020139962-appb-000032
Figure PCTCN2020139962-appb-000033
Figure PCTCN2020139962-appb-000034
其中P x,P y,P z分别表征在三维空间中体像素沿x，y和z轴的物理几何长度。n,m和k分别表示三维空间矩阵中沿x，y和z轴的像素数目。
本发明通过同一操作者重复对同一受试者进行2次OCT成像，来测试指标的重复性，并和二维相应的指标进行比较，测试结果如下表所示。
    0-6mm 左侧3-6mm 左侧1-3mm 0-1mm 右侧1-3mm 右侧3-6mm
CVV 垂直 0.995 0.989 0.987 0.985 0.963 0.964
  水平 0.994 0.984 0.98 0.99 0.96 0.989
  三维 0.998 0.997 0.998 0.997 0.998 0.999
CVI 垂直 0.883 0.736 0.815 0.814 0.714 0.731
  水平 0.865 0.664 0.798 0.864 0.786 0.541
  三维 0.991 0.986 0.965 0.975 0.983 0.982
CII 垂直 0.883 0.736 0.815 0.814 0.714 0.731
  水平 0.865 0.664 0.798 0.864 0.786 0.541
  三维 0.991 0.986 0.965 0.975 0.983 0.982
SV 垂直 0.968 0.86 0.962 0.871 0.927 0.934
  水平 0.935 0.694 0.964 0.939 0.91 0.768
  三维 0.994 0.977 0.989 0.990 0.994 0.992
可见，三维指标的重复性要显著优于二维指标。其中二维指标中，垂直二维指标的重复性优于水平二维指标。
本发明对系统获取的脉络膜图像信号进行预处理后，基于OCT后向散射信号逆向补偿的思想，建立逆向信号衰减补偿模型，逆向补偿脉络膜信号，增强血管与非血管组织之间的信噪比，并进一步采用基于深度学习的分割方法，对脉络膜内外边界进行智能化分割；在边界分割基础上，进一步采用改进的自适应阈值方法，自动分离出三维脉络膜血管与非血管组织，并根据图像中血管在三维体空间的分布及占比，计算出能够表征脉络膜缺血的全局和各区域定量化指标。该方法适用于对组织具有较高穿透的能够获取脉络膜的所有OCT系统及其图像，具有自动、普适性强、且能高精度反映脉络膜三维血管异常。
以上所述仅是本发明的优选实施方式，本发明的保护范围并不仅局限于上述实施例，凡属于本发明思路下的技术方案均属于本发明的保护范围。应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明原理前提下的若干改进和润饰，这些改进和润饰也应视为本发明的保护范围。

Claims (10)

  1. 一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法，其特征在于，包括以下步骤：
    (1)获得图像信号预处理后的脉络膜的OCT图像；
    (2)基于深度学习智能化分割脉络膜内外边界；
    (3)自适应阈值分离出脉络膜血管与非血管；
    (4)三维全局和各区域定量化指标。
  2. 根据权利要求1所述的一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法，其特征在于，所述的步骤(2)基于深度学习智能化分割脉络膜内外边界，包括以下步骤：
    a、基于最短路径图论算法和对OCT影像脉络膜的内外边界进行半自动标注，准确描绘出图像中脉络膜的上下边界；将已标注的脉络膜图像集按8：2随机分为训练集、测试集两个部分；
    b、将训练图像输入开源的深度学习神经网络模型中，训练优化算法设为随机梯度下降法(SGD)，算法学习比率为1.0e-5，迭代动量为0.9，迭代代价函数为Dice系数，迭代遍历数为150，批样本量为8，此外，对输入图像进行必要的增强处理，改善模型的鲁棒性，其中，Dice系数计算公式为：
    Figure PCTCN2020139962-appb-100001
    其中X表示的是脉络膜边界预测集，Y表示的是脉络膜边界标注集，|X∩Y|表示两个集合之间的相交部分或重叠部分，|X|+|Y|表示两者的总量，Dice系数越大，表明两个集合相似度越高，模型越准确；当预测集与标注集完全相同时，Dice系数为1；当预测集与标注集不相关时，Dice系数为0，在模型训练过程中，设置Dice系数大于0.95作为目标函数；
    c、将测试集图像输入深度学习神经网络模型中，计算脉络膜边界的输出结果和标注集之间的Dice系数、边界误差来评估深度学习神经网络模型的分割性能。
  3. 根据权利要求1所述的一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法，其特征在于，所述的步骤(3)自适应阈值分离出脉络膜血管与非血管，包括以下步骤：
    A、在OCT图像中，以长度和宽度都为2*w+1个像素块的正方形框作为局部窗口，w为最大不超过图像的长度和宽度的一半的正整数，坐标(x,y)为这个正方形框的几何中心，统计框内所有像素块的亮度信息，求得其均值m(x,y)和方差s(x,y)，根据参数k，求得框内阈值T(x,y)，计算公式如下：
    T(x,y)＝m(x,y)+k*s(x,y)；
    B、根据框内阈值T(x,y)，对框内所有像素块进行二值化处理，如下所示：
    Figure PCTCN2020139962-appb-100002
    其中i和j是表征像素块相对几何中心(x，y)的相对坐标，A x+i,y+j表示坐标(x+i,y+j)像素块的亮度，A x+i,y+j表示坐标(x+i,y+j)像素块的的二值化数据；
    C、依次移动局部窗口，实现OCT图像中所有像素块亮度矩阵A的二值化处理，获得二值化矩阵B。
  4. 根据权利要求1所述的一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法，其特征在于，所述的步骤(4)三维全局和各区域定量化指标，包括以下步骤：
    i、图像获取：以眼底黄斑中心凹为中心，通过放射状扫描模式进行成像，获取m张脉络膜二维横断面数据；
    ii、特征点标记：以黄斑中心凹作为OCT图像的特征点进行标记，用于后续图像配准；
    iii、图像配准：通过水平平移黄斑中心凹的位置，使所有OCT图像中黄斑中心凹在空间坐标中横坐标相同，实现图像的配准；
    iiii、脉络膜三维空间重建：基于深度学习神经网络模型所获得的脉络膜上下界，提取配准后的图像中脉络膜区域模板M(x,y,z)，并结合自适应阈值方法自动分离血管和基质，得到二值化后的三维空间脉络膜结构矩阵V(x,y,z)；
    iiiii、指标建立及测试：我们建立脉络膜血管体积CVV、脉络膜非血管体积SV、脉络膜血管指数CVI和脉络膜缺血指数CII等指标进行量化，各指标的计算公式如下所示：
    Figure PCTCN2020139962-appb-100003
    Figure PCTCN2020139962-appb-100004
    Figure PCTCN2020139962-appb-100005
    Figure PCTCN2020139962-appb-100006
    其中P x,P y,P z分别表征在三维空间中体像素沿x，y和z轴的物理几何长度，n,m和k分别表示三维空间矩阵中沿x，y和z轴的像素数目。
  5. 根据权利要求1所述的一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法，其特征在于，所述的步骤(1)获得图像信号预处理后的脉络膜的OCT图像，包括以下步骤：
    (一)数据的获取：利用OCT获取眼底图像，对图片进行适当裁剪，保留OCT脉络膜强度图；
    (二)逆向衰减补偿脉络膜信号：通过提取脉络膜散射光衰减原理和规律，构建OCT信号的衰减校正处理算法，提高脉络膜图像的可视化和对比度，OCT信号的衰减校正处理算法包含两个步骤，分别对光的衰减补偿和图像增强 对比，在OCT系统中，参考臂和样品臂间干涉的光电信号用如下公式表示：
    Figure PCTCN2020139962-appb-100007
    式中k为波数，采集信号被分割成m个等间距的波数，ρ是OCT探测器的光电转化效能，S[k m]指所对应波段光源的辐射能量，Δx是参考臂和样本臂间光程差，R R和R S分别为参考臂和样品臂的反射率；
    参考臂和样品臂间的交互项H[k m]由上式(1)得到:
    Figure PCTCN2020139962-appb-100008
    通过对上式(2)进行离散傅里叶逆变换得到沿深度方向的反射率剖面函数S(z):
    Figure PCTCN2020139962-appb-100009
    对公式进行离散化处理，对每个像素点的OCT信号进行衰减校正补偿：
    Figure PCTCN2020139962-appb-100010
    其中N是A扫描的像素点数，α根据组织进行调整，上述公式(3)成立的前提是假设大部分光束能量是在成像深度范围内衰减的，成像深度范围外的衰减忽略不计；
    将对衰减校正补偿后的原始信号进行取幂运算增强图像对比度，此时每个像素信号强度为:
    Figure PCTCN2020139962-appb-100011
    式(5)中S ac(z)是衰减校正后的信号。
  6. 一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析装置，其特征在于，包括以下模块：
    脉络膜OCT图像获取及图像信号预处理模块：通过OCT获取的深部脉络膜血管及基质因后向散射衰减信号进行逆向补偿，增强脉络膜血管的信噪比； 用于获取增强的OCT脉络膜图像信号；
    基于深度学习的脉络膜内外边界智能化分割模块：通过采用基于深度学习的分割方法，用于对脉络膜与视网膜上皮层以及脉络膜与巩膜之间的边界进行智能化分割；
    脉络膜血管与非血管自适应阈值分割模块：通过采用自适应阈值分割方法，用于自动分离出三维脉络膜血管与非血管组织；
    三维全局和各区域定量化指标模块：通过根据图像中血管在三维体空间的分布及占比，用于计算出能够表征脉络膜缺血的全局和各区域定量化指标。
  7. 根据权利要求6所述的一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析装置，其特征在于，所述的脉络膜OCT图像获取及图像信号预处理模块包括以下算法模型:
    (一)数据的获取单元：利用OCT获取眼底图像，对图片进行适当裁剪，保留OCT脉络膜强度图；
    (二)逆向衰减补偿脉络膜信号单元：通过提取脉络膜散射光衰减原理和规律，构建OCT信号的衰减校正处理算法，可提高脉络膜图像的可视化和对比度，OCT信号的衰减校正处理算法包含两个步骤，分别对光的衰减补偿和图像增强对比，在OCT系统中，参考臂和样品臂间干涉的光电信号用如下公式表示：
    Figure PCTCN2020139962-appb-100012
    式中k为波数，采集信号被分割成m个等间距的波数，ρ是OCT探测器的光电转化效能，S[k m]指所对应波段光源的辐射能量，Δx是参考臂和样本臂间光程差，R R和R S分别为参考臂和样品臂的反射率；
    参考臂和样品臂间的交互项H[k m]由上式(1)得到:
    Figure PCTCN2020139962-appb-100013
    通过对上式(2)进行离散傅里叶逆变换得到沿深度方向的反射率剖面函数 S(z):
    Figure PCTCN2020139962-appb-100014
    对公式进行离散化处理，对每个像素点的OCT信号进行衰减校正补偿：
    Figure PCTCN2020139962-appb-100015
    其中N是A扫描的像素点数，α根据组织进行调整，上述公式(3)成立的前提是假设大部分光束能量是在成像深度范围内衰减的，成像深度范围外的衰减可以忽略不计；
    将对衰减校正补偿后的原始信号进行取幂运算增强图像对比度，此时每个像素信号强度为:
    Figure PCTCN2020139962-appb-100016
    式(5)中S ac(z)是衰减校正后的信号。
  8. 根据权利要求6所述的一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析装置，其特征在于，所述的基于深度学习的脉络膜内外边界智能化分割模块包括以下算法模型:
    a、基于最短路径图论算法和对OCT影像脉络膜的内外边界进行半自动标注，准确描绘出图像中脉络膜的上下边界的运算单元；将已标注的脉络膜图像集按8：2随机分为训练集、测试集两个部分；
    b、将训练图像输入开源的深度学习神经网络模型中，训练优化算法设为随机梯度下降法(SGD)，算法学习比率为1.0e-5，迭代动量为0.9，迭代代价函数为Dice系数，迭代遍历数为150，批样本量为8，此外，对输入图像进行必要的增强处理，改善模型的鲁棒性，其中，Dice系数计算公式为：
    Figure PCTCN2020139962-appb-100017
    其中X表示的是脉络膜边界预测集，Y表示的是脉络膜边界标注集，|X∩Y|表 示两个集合之间的相交部分或重叠部分，|X|+|Y|表示两者的总量，Dice系数越大，表明两个集合相似度越高，模型越准确；当预测集与标注集完全相同时，Dice系数为1；当预测集与标注集不相关时，Dice系数为0，在模型训练过程中，设置Dice系数大于0.95作为目标函数；
    c、将测试集图像输入深度学习神经网络模型中，计算脉络膜边界的输出结果和标注集之间的Dice系数、边界误差来评估深度学习神经网络模型的分割性能的运算单元。
  9. 根据权利要求6所述的一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析装置，其特征在于，所述的脉络膜血管与非血管自适应阈值分割模块包括以下算法模型:
    A、在OCT图像中，以长度和宽度都为2*w+1个像素块的正方形框作为局部窗口，w为最大不超过图像的长度和宽度的一半的正整数，坐标(x,y)为这个正方形框的几何中心，统计框内所有像素块的亮度信息，求得其均值m(x,y)和方差s(x,y)，根据参数k，求得框内阈值T(x,y)，计算公式如下：
    T(x,y)＝m(x,y)+k*s(x,y)；
    B、根据框内阈值T(x,y)，对框内所有像素块进行二值化处理单元，如下所示：
    Figure PCTCN2020139962-appb-100018
    其中i和j是表征像素块相对几何中心(x，y)的相对坐标，A x+i,y+j表示坐标(x+i,y+j)像素块的亮度，A x+i,y+j表示坐标(x+i,y+j)像素块的的二值化数据；
    C、依次移动局部窗口，实现OCT图像中所有像素块亮度矩阵A的二值化处理，获得二值化矩阵B的运算单元。
  10. 根据权利要求6所述的一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析装置，其特征在于，所述的三维全局和各区域定量化指 标模块包括以下算法模型:
    i、图像获取单元：以眼底黄斑中心凹为中心，通过放射状扫描模式进行成像，获取m张脉络膜二维横断面数据；
    ii、特征点标记单元：以黄斑中心凹作为OCT图像的特征点进行标记，用于后续图像配准；
    iii、图像配准单元：通过水平平移黄斑中心凹的位置，使所有OCT图像中黄斑中心凹在空间坐标中横坐标相同，实现图像的配准；
    iiii、脉络膜三维空间重建单元：基于深度学习神经网络模型所获得的脉络膜上下界，提取配准后的图像中脉络膜区域模板M(x,y,z)，并结合自适应阈值方法自动分离血管和基质，得到二值化后的三维空间脉络膜结构矩阵V(x,y,z)；iiiii、指标建立及测试：我们建立脉络膜血管体积CVV、脉络膜非血管体积SV、脉络膜血管指数CVI和脉络膜缺血指数CII等指标进行量化，各指标的计算公式如下所示：
    Figure PCTCN2020139962-appb-100019
    Figure PCTCN2020139962-appb-100020
    Figure PCTCN2020139962-appb-100021
    Figure PCTCN2020139962-appb-100022
    其中P x,P y,P z分别表征在三维空间中体像素沿x，y和z轴的物理几何长度，n,m和k分别表示三维空间矩阵中沿x，y和z轴的像素数目。
PCT/CN2020/139962 2020-07-10 2020-12-28 一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置 WO2022007352A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
AU2020458040A AU2020458040B2 (en) 2020-07-10 2020-12-28 Three-dimensional choroidal blood vessel imaging and quantitative analysis method and device based on optical coherence tomography system

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010660715.5A CN111862114A (zh) 2020-07-10 2020-07-10 一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置
CN202010660715.5 2020-07-10

Publications (1)

Publication Number Publication Date
WO2022007352A1 true WO2022007352A1 (zh) 2022-01-13

Family

ID=73152635

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/139962 WO2022007352A1 (zh) 2020-07-10 2020-12-28 一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置

Country Status (3)

Country Link
CN (1) CN111862114A (zh)
AU (1) AU2020458040B2 (zh)
WO (1) WO2022007352A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114879188A (zh) * 2022-04-20 2022-08-09 北京理工大学 模型自适应的深度学习sar三维成像方法
CN116485820A (zh) * 2023-06-21 2023-07-25 杭州堃博生物科技有限公司 动静脉图像提取方法、装置及非易失性存储介质
CN116563414A (zh) * 2023-07-11 2023-08-08 天津博霆光电技术有限公司 基于oct的心血管成像颤影消除方法及设备
CN116703948A (zh) * 2023-08-03 2023-09-05 杭州脉流科技有限公司 基于深度神经网络的颅内血管树分割方法和分割装置
CN117291920A (zh) * 2023-11-27 2023-12-26 天津恒宇医疗科技有限公司 基于oct成像的脂质斑块检测分析方法及系统
CN117593469A (zh) * 2024-01-17 2024-02-23 厦门大学 一种3d内容创建方法

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111861917A (zh) * 2020-07-10 2020-10-30 温州医科大学 一种基于信号逆向补偿的脉络膜oct图像增强方法及装置
CN111862114A (zh) * 2020-07-10 2020-10-30 温州医科大学 一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置
CN113160160B (zh) * 2021-04-12 2022-07-08 哈尔滨医科大学 一种基于oct光衰图像的钙化识别方法
CN113598703B (zh) * 2021-07-06 2024-02-20 温州医科大学附属眼视光医院 一种基于边界模糊程度的脉络膜新生血管活动性量化方法
CN113723505B (zh) * 2021-08-30 2024-04-23 上海大学 一种基于细胞骨架图像定量分析的细胞图像分类方法及其分类器
CN115456950B (zh) * 2022-08-15 2023-08-11 上海交通大学医学院附属第九人民医院 一种穿支血管自动识别和定位方法、设备及存储介质
CN115456962B (zh) * 2022-08-24 2023-09-29 中山大学中山眼科中心 一种基于卷积神经网络的脉络膜血管指数预测方法和装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110222752A1 (en) * 2008-04-08 2011-09-15 Three Palm Software Microcalcification enhancement from digital mammograms
CN108416793A (zh) * 2018-01-16 2018-08-17 武汉诺影云科技有限公司 基于三维相干断层成像图像的脉络膜血管分割方法及系统
CN108836257A (zh) * 2018-06-15 2018-11-20 杭州富扬科技有限公司 一种眼底oct图像中视网膜分层方法
CN109003284A (zh) * 2018-07-11 2018-12-14 武汉科技大学 基于层厚统计信息模型的相干光断层扫描图像的层分割方法
CN111862114A (zh) * 2020-07-10 2020-10-30 温州医科大学 一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104755908B (zh) * 2012-07-27 2017-12-12 统雷有限公司 敏捷成像系统
US10463247B2 (en) * 2015-06-22 2019-11-05 The Regents Of The University Of California Automatic three-dimensional segmentation method for OCT and doppler OCT angiography
CN106651846B (zh) * 2016-12-20 2019-07-12 中南大学湘雅医院 视网膜血管图像的分割方法
CN106934761A (zh) * 2017-02-15 2017-07-07 苏州大学 一种三维非刚性光学相干断层扫描图像的配准方法
CN108257126B (zh) * 2018-01-25 2022-06-24 苏州大学 三维视网膜oct图像的血管检测和配准方法、设备及应用
CN109509178B (zh) * 2018-10-24 2021-09-10 苏州大学 一种基于改进的U-net网络的OCT图像脉络膜分割方法
CN109730633A (zh) * 2018-12-28 2019-05-10 中国科学院宁波工业技术研究院慈溪生物医学工程研究所 基于光学相干断层成像体扫描的脉络膜血管造影方法及设备
CN110517235B (zh) * 2019-08-19 2021-10-19 苏州大学 一种基于GCS-Net进行OCT图像脉络膜自动分割方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110222752A1 (en) * 2008-04-08 2011-09-15 Three Palm Software Microcalcification enhancement from digital mammograms
CN108416793A (zh) * 2018-01-16 2018-08-17 武汉诺影云科技有限公司 基于三维相干断层成像图像的脉络膜血管分割方法及系统
CN108836257A (zh) * 2018-06-15 2018-11-20 杭州富扬科技有限公司 一种眼底oct图像中视网膜分层方法
CN109003284A (zh) * 2018-07-11 2018-12-14 武汉科技大学 基于层厚统计信息模型的相干光断层扫描图像的层分割方法
CN111862114A (zh) * 2020-07-10 2020-10-30 温州医科大学 一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114879188A (zh) * 2022-04-20 2022-08-09 北京理工大学 模型自适应的深度学习sar三维成像方法
CN116485820A (zh) * 2023-06-21 2023-07-25 杭州堃博生物科技有限公司 动静脉图像提取方法、装置及非易失性存储介质
CN116485820B (zh) * 2023-06-21 2023-09-22 杭州堃博生物科技有限公司 动静脉图像提取方法、装置及非易失性存储介质
CN116563414A (zh) * 2023-07-11 2023-08-08 天津博霆光电技术有限公司 基于oct的心血管成像颤影消除方法及设备
CN116563414B (zh) * 2023-07-11 2023-09-12 天津博霆光电技术有限公司 基于oct的心血管成像颤影消除方法及设备
CN116703948A (zh) * 2023-08-03 2023-09-05 杭州脉流科技有限公司 基于深度神经网络的颅内血管树分割方法和分割装置
CN116703948B (zh) * 2023-08-03 2023-11-14 杭州脉流科技有限公司 基于深度神经网络的颅内血管树分割方法和分割装置
CN117291920A (zh) * 2023-11-27 2023-12-26 天津恒宇医疗科技有限公司 基于oct成像的脂质斑块检测分析方法及系统
CN117291920B (zh) * 2023-11-27 2024-03-12 天津恒宇医疗科技有限公司 基于oct成像的脂质斑块检测分析方法及系统
CN117593469A (zh) * 2024-01-17 2024-02-23 厦门大学 一种3d内容创建方法

Also Published As

Publication number Publication date
AU2020458040A1 (en) 2023-01-19
AU2020458040B2 (en) 2023-04-13
CN111862114A (zh) 2020-10-30

Similar Documents

Publication Publication Date Title
WO2022007352A1 (zh) 一种基于光学相干断层扫描系统的脉络膜三维血管成像及定量化分析方法与装置
Devalla et al. A deep learning approach to digitally stain optical coherence tomography images of the optic nerve head
CN109493954B (zh) 一种基于类别判别定位的sd-oct图像视网膜病变检测系统
US10194866B2 (en) Methods and apparatus for reducing artifacts in OCT angiography using machine learning techniques
US9418423B2 (en) Motion correction and normalization of features in optical coherence tomography
Yu et al. Fast localization and segmentation of optic disk in retinal images using directional matched filtering and level sets
Narasimha-Iyer et al. Robust detection and classification of longitudinal changes in color retinal fundus images for monitoring diabetic retinopathy
Muramatsu et al. Automated segmentation of optic disc region on retinal fundus photographs: Comparison of contour modeling and pixel classification methods
Bagci et al. Thickness profiles of retinal layers by optical coherence tomography image segmentation
Baroni et al. Towards quantitative analysis of retinal features in optical coherence tomography
Chen et al. Automated segmentation of fluid regions in optical coherence tomography B-scan images of age-related macular degeneration
CN103348359A (zh) 使用光学相干断层照相的3d视网膜破裂检测
Mittal et al. Computerized retinal image analysis-a survey
WO2022007353A1 (zh) 一种基于信号逆向补偿的脉络膜oct图像增强方法及装置
CN109829942A (zh) 一种眼底图像视网膜血管管径自动量化方法
Fu et al. Optic disc segmentation by U-net and probability bubble in abnormal fundus images
Zhang et al. A novel technique for robust and fast segmentation of corneal layer interfaces based on spectral-domain optical coherence tomography imaging
CN114092405A (zh) 一种针对黄斑水肿oct图像的视网膜层自动分割方法
Eghtedar et al. An update on choroidal layer segmentation methods in optical coherence tomography images: a review
Garcia-Marin et al. Patch-based CNN for corneal segmentation of AS-OCT images: Effect of the number of classes and image quality upon performance
Yadav et al. Optic nerve head three-dimensional shape analysis
Tan et al. Three-dimensional choroidal vessel network quantification using swept source optical coherence tomography
AU2015245908B2 (en) Method for the analysis of image data representing a three-dimensional volume of biological tissue
CN107657605A (zh) 一种基于主动轮廓和能量约束的筛板前表面深度测量方法
Singh et al. Automatic detection of hard exudates shadow region within retinal layers of OCT images

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: 20944801

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020458040

Country of ref document: AU

Date of ref document: 20201228

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20944801

Country of ref document: EP

Kind code of ref document: A1