[go: up one dir, main page]

WO2022033580A1 - 视网膜血管动静脉区分方法和装置、设备 - Google Patents

视网膜血管动静脉区分方法和装置、设备 Download PDF

Info

Publication number
WO2022033580A1
WO2022033580A1 PCT/CN2021/112487 CN2021112487W WO2022033580A1 WO 2022033580 A1 WO2022033580 A1 WO 2022033580A1 CN 2021112487 W CN2021112487 W CN 2021112487W WO 2022033580 A1 WO2022033580 A1 WO 2022033580A1
Authority
WO
WIPO (PCT)
Prior art keywords
blood vessel
image
arteriovenous
segments
main
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/CN2021/112487
Other languages
English (en)
French (fr)
Inventor
代黎明
姜泓羊
张冬冬
于雷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Zhenhealth Technology Co Ltd
Original Assignee
Beijing Zhenhealth Technology Co Ltd
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 Beijing Zhenhealth Technology Co Ltd filed Critical Beijing Zhenhealth Technology Co Ltd
Priority to JP2022512438A priority Critical patent/JP7333465B2/ja
Priority to US17/755,560 priority patent/US12307663B2/en
Priority to KR1020227015077A priority patent/KR102826940B1/ko
Publication of WO2022033580A1 publication Critical patent/WO2022033580A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/77—Retouching; Inpainting; Scratch removal
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/14—Vascular patterns
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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 present disclosure relates to the technical field of computer-aided diagnosis, and in particular, to a method, device, and device for distinguishing retinal blood vessels from arteries and veins.
  • Fundus retinal arteriovenous diameter ratio index is an important reference index for doctors when diagnosing arteriosclerotic diseases.
  • doctors actually read the film they still rely on the visual method to give a rough ratio of the arteriovenous diameter ratio: greater than 1/2, greater than 1/3 and less than 1/2, less than 1/3 and so on.
  • the visual inspection method especially for new ophthalmologists or general ophthalmologists in the process of reading images. Therefore, the accurate measurement of the ratio of the arteriovenous diameter ratio of the main retinal vessels by computer is of great help to doctors in clinical diagnosis of arteriosclerosis, and can effectively improve the efficiency of doctors to read pictures.
  • the method based on machine learning method to distinguish retinal vascular arteries and veins has the problem of poor universality and robustness of the algorithm.
  • the present disclosure proposes a method for distinguishing retinal blood vessels from arteries and veins, including:
  • a plurality of described single blood vessel segments are carried out caliber measurement to obtain the caliber width of each described single vascular segment, and a plurality of arteriovenous blood vessel pairs are selected and obtained according to each described caliber width;
  • the unqualified arterial and venous blood vessels are excluded, including:
  • the average image brightness of a single blood vessel segment in each of the arteriovenous blood vessel pairs is acquired, and the unqualified arteriovenous blood vessel pairs are eliminated from the plurality of arteriovenous blood vessel pairs according to the average image brightness.
  • the method further includes the step of removing unqualified arteriovenous blood vessel pairs from a plurality of the arteriovenous blood vessel pairs and retaining qualified arteriovenous blood vessel pairs;
  • the unqualified arterial and venous blood vessels are excluded, including:
  • the average image brightness of a single blood vessel segment in each of the arteriovenous blood vessel pairs is acquired, and the unqualified arteriovenous blood vessel pairs are eliminated from the plurality of arteriovenous blood vessel pairs according to the average image brightness.
  • the extraction of the main blood vessel is performed according to the blood vessel extraction image, the fundus image and the coordinates of the optic disc center, including:
  • the skeleton extraction function in Opencv is used to extract the blood vessel skeleton from the connected blood vessels.
  • rejecting unqualified arterial and venous blood vessel pairs also includes:
  • the distance between a single blood vessel in each of the arteriovenous blood vessel pairs between an arterial blood vessel and a venous blood vessel is obtained, and the unqualified arteriovenous blood vessel pair is eliminated from the plurality of arteriovenous blood vessel pairs according to the distance.
  • extracting the intersection of blood vessels according to the blood vessel skeleton and removing the intersection includes:
  • the pixel value within the radius of four pixels is set to 0.
  • the main blood vessels obtained by removing the small blood vessel segments and non-main blood vessel branches of the blood vessel skeleton according to the optic disc center coordinates and the fundus image include:
  • 3 ⁇ 3 ELLIPSE was used to refine the grayscale image to obtain the main blood vessels.
  • a plurality of single blood vessel segments are obtained by intercepting the main blood vessel based on the main blood vessel image, including:
  • a plurality of the single blood vessel segments are obtained according to a preset angle interval.
  • measuring the diameter of a plurality of the single blood vessel segments to obtain the diameter width of each of the single blood vessel segments, and selecting and obtaining a plurality of arteriovenous blood vessel pairs according to the diameter widths of the individual vessels including: :
  • a first straight line is obtained by performing straight line fitting on a contour line of the single blood vessel segment between the vertical line equations;
  • a second straight line is obtained by performing straight line fitting on another of the contour lines using the least squares method
  • a retinal blood vessel arteriovenous distinguishing device including a data acquisition module, a blood vessel screening module, and an arteriovenous blood vessel pair selection module;
  • the data acquisition module is configured to acquire blood vessel extraction images, fundus images and optic disc center coordinates
  • the blood vessel screening module is configured to extract the main blood vessel according to the blood vessel extraction image, the fundus image and the coordinates of the optic disc center to obtain the main blood vessel image, and perform the main blood vessel image based on the main blood vessel image. Obtain multiple single vessel segments;
  • the arteriovenous blood vessel pair selection module is configured to measure the diameter of a plurality of the single blood vessel segments to obtain the diameter width of each of the single blood vessel segments, and select a plurality of arteriovenous vessels according to the diameter widths of each of the tubes. blood vessel pair;
  • the unqualified arterial and venous blood vessels are excluded, including:
  • the average image brightness of a single blood vessel segment in each of the arteriovenous blood vessel pairs is acquired, and the unqualified arteriovenous blood vessel pairs are eliminated from the plurality of arteriovenous blood vessel pairs according to the average image brightness.
  • a retinal blood vessel arteriovenous differentiation device comprising:
  • memory for storing processor-executable instructions
  • the processor is configured to implement any of the foregoing methods when executing the executable instructions.
  • the main blood vessels are extracted according to the blood vessel extraction image, the fundus image and the optic disc center coordinates to obtain the main blood vessel image, and the main blood vessel is intercepted based on the main blood vessel image to obtain multiple single blood vessels.
  • the diameter of each single blood vessel segment is measured to obtain the diameter width of each single blood vessel segment, and a plurality of arteriovenous blood vessel pairs are selected according to the width of each tube diameter.
  • FIG. 1 shows a flowchart of a retinal blood vessel arteriovenous differentiation method according to an embodiment of the present disclosure
  • FIG. 2 shows a schematic diagram of a fundus image with a size of 512 ⁇ 512 in a method for distinguishing retinal blood vessels and arteries and veins according to an embodiment of the present disclosure
  • FIG. 3 shows a schematic diagram of a 512 ⁇ 512 size blood vessel extraction image of the retinal blood vessel arteriovenous distinction method according to an embodiment of the present disclosure
  • FIG. 4 shows a schematic diagram of a blood vessel extraction test image of the retinal blood vessel arteriovenous differentiation method according to an embodiment of the present disclosure
  • FIG. 5 shows a schematic diagram of a blood vessel extraction test result image of the retinal blood vessel arteriovenous distinction method according to an embodiment of the present disclosure
  • FIG. 6 shows another flowchart of the retinal blood vessel arteriovenous differentiation method according to an embodiment of the present disclosure
  • FIG. 7 shows a schematic diagram of a fundus image of a retinal blood vessel arteriovenous differentiation method according to an embodiment of the present disclosure
  • FIG. 8 shows a schematic diagram of an optic disc detection test image of the retinal blood vessel arteriovenous differentiation method according to an embodiment of the present disclosure
  • FIG. 9 shows a schematic diagram of an optic disc detection test result image of the retinal blood vessel arteriovenous distinction method according to an embodiment of the present disclosure
  • FIG. 10 shows a schematic diagram of the direction of blood vessels of the method for distinguishing retinal blood vessel arteries and veins according to an embodiment of the present disclosure
  • FIG. 11 shows a schematic diagram of the macular area of the retinal blood vessel arteriovenous differentiation method according to an embodiment of the present disclosure
  • FIG. 12 shows a schematic diagram of a ring template of the retinal blood vessel arteriovenous differentiation method according to an embodiment of the present disclosure
  • FIG. 13 shows a schematic diagram of a small blood vessel segment of a method for distinguishing retinal blood vessels from arteries and veins according to an embodiment of the present disclosure
  • FIG. 14 shows a schematic diagram of contour line fitting of the retinal blood vessel arteriovenous distinction method according to an embodiment of the present disclosure
  • FIG. 15 shows a schematic diagram of the tube diameter of the retinal blood vessel arteriovenous differentiation method according to an embodiment of the present disclosure
  • FIG. 16 shows a block diagram of a retinal blood vessel arteriovenous discriminating device according to an embodiment of the present disclosure
  • FIG. 17 shows a block diagram of a retinal blood vessel arteriovenous discriminating apparatus according to an embodiment of the present disclosure.
  • FIG. 1 shows a flowchart of a method for distinguishing retinal blood vessels from arteries and veins according to an embodiment of the present disclosure.
  • the retinal blood vessel arteriovenous distinction method includes:
  • Step S100 obtaining the blood vessel extraction image, fundus image and optic disc center coordinates
  • step S200 extracting the main blood vessel according to the blood vessel extraction image, the fundus image and the optic disc center coordinates, obtaining the main blood vessel image, and intercepting the main blood vessel based on the main blood vessel image
  • a plurality of single blood vessel segments are obtained.
  • step S300 the diameter of the plurality of single blood vessel segments is measured to obtain the diameter width of each single blood vessel segment, and a plurality of arteriovenous blood vessel pairs are selected according to the diameter widths of the individual blood vessels.
  • the main blood vessels are extracted according to the blood vessel extraction image, the fundus image and the optic disc center coordinates to obtain the main blood vessel image, and the main blood vessel is intercepted based on the main blood vessel image to obtain multiple single blood vessels.
  • the diameter of each single blood vessel segment is measured to obtain the diameter width of each single blood vessel segment, and a plurality of arteriovenous blood vessel pairs are selected according to the width of each tube diameter.
  • the method for distinguishing retinal blood vessel arteries and veins in the embodiments of the present application is implemented based on machine learning.
  • the extraction of blood vessels and the acquisition of the coordinates of the center of the optic disc may be realized by using a corresponding network model.
  • the blood vessel extraction image can be realized by using the U-net network structure, and the center coordinates of the optic disc can also be realized by using the U-net network structure.
  • the training data set for blood vessel extraction and optic disc detection is first constructed.
  • 20 training sets and 18 test sets are used in the DRIVE public data set.
  • the training set is the refuge2018 training set and the validation set 800 Zhang, 50 drishti-gsl training data sets, 159 rim-one-r3 training data sets, a total of 4036 after data augmentation, the test set includes 400 refuge2018 test sets, 51 drishti-gsl test sets, a total of 451 pieces , Further, in the blood vessel extraction layer, the training data set is first cropped with black borders, then the G channel is extracted, and then CLAHE (Limited Contrast Adaptive Histogram Equalization) image enhancement is used, followed by normalization, and the image is horizontally inverted. Transfer, contrast adjustment and random cropping to expand the training data set to 3004.
  • CLAHE Lited Contrast Adaptive Histogram Equalization
  • the network structure of the model training of the blood vessel extraction layer is U-Net
  • the input size is 512 ⁇ 512
  • the optimization method is Adam
  • the learning rate is 0.0001
  • Batch_size is 2
  • loss function is dice_lose
  • training times is 150epoch
  • early_stopping mechanism is added, see Figure 6 for model test images, and Figure 7 for model result images.
  • the training set data image is first cropped with black borders, then median filtered, the image is scaled to 512 ⁇ 512 size, then the G channel is extracted, and then CLAHE (limited contrast adaptive histogram is used) Balanced) image enhancement, followed by normalization, horizontal inversion and contrast adjustment of the image, and the training data set is expanded to 4036.
  • CLAHE limited contrast adaptive histogram is used
  • the network structure of the model training of the optic disc detection layer is U-Net, and the input size is 512 ⁇ 512 , the optimization method is Adam, the learning rate is 0.0001, the Batch_size is 4, the loss function is dice_lose, the number of training times is 150epoch, and the early_stopping mechanism is added.
  • the model test image is shown in Figure 8, and the test result image is shown in Figure 9.
  • step S100 is executed to acquire the blood vessel extraction image, the fundus image and the coordinates of the center of the optic disc.
  • first input a color image of the fundus retina see Figure 3 and perform preprocessing operations on it, and then perform blood vessel extraction and optic disc detection on it to obtain the blood vessel extraction results and optic disc detection results.
  • the above operations After all is completed, the preprocessed color fundus image, retinal blood vessel extraction image, and optic disc coordinates are sent as input to the arterial and venous distinguishing module to distinguish the veins.
  • step S100 is performed, referring to FIG. Figure 5.
  • the 512 ⁇ 512 color retinal fundus image, the 512 ⁇ 512 size blood vessel extraction image and the optic disc center coordinates based on the 512 ⁇ 512 size image can be obtained directly from the blood vessel extraction results and the optic disc detection results. .
  • step S200 is performed to extract the main blood vessel according to the blood vessel extraction image, the fundus image and the optic disc center coordinates to obtain the main blood vessel image, and based on the main blood vessel image, the main blood vessel is intercepted to obtain a plurality of single blood vessel segments.
  • the extraction of main blood vessels is performed according to the blood vessel extraction image, the fundus image and the optic disc center coordinates, including: referring to FIG. 2, step S200a is performed to remove the disconnected blood vessels in the blood vessel extraction image to obtain the largest connected blood vessel , step S200b, extracting the blood vessel skeleton according to the connected blood vessels, step S200c, step S200d and step S200e, extracting the intersection of the blood vessels according to the blood vessel skeleton and removing the intersection, according to the optic disc center coordinate and the fundus image
  • the small blood vessel segments of the blood vessel skeleton are The main blood vessels are obtained by removing the branches of non-main blood vessels and non-main blood vessels.
  • the skeleton extraction function in Opencv is used to extract the blood vessel skeleton from the connected blood vessels.
  • extracting the intersection of blood vessels according to the blood vessel skeleton and removing the intersection includes: binarizing the blood vessel skeleton to obtain a grayscale image, traversing the grayscale image with a 3 ⁇ 3 template, and extracting the grayscale image by an enumeration method.
  • the intersection of take the intersection as the center, and set the pixel value within the radius of four pixels to 0.
  • removing the small blood vessel segments and non-main blood vessel branches of the vascular skeleton according to the optic disc center coordinates and the fundus image to obtain the main blood vessels includes: connecting the optic disc center coordinates and the macular area of the fundus image as the positive semi-axis, and retaining the positive semi-axis. 110° clockwise and 110° anti-clockwise of the positive semi-axis, use 3 ⁇ 3 ELLIPSE to refine the grayscale image to obtain the main vessels.
  • intercepting the main blood vessel based on the image of the main blood vessel to obtain a plurality of single blood vessel segments including: referring to FIG. 2, step S200f, taking the center coordinate of the optic disc as the center of the circle, traversing the main vessel outward with a predetermined step size within a preset radius to intercept the main vessel.
  • a plurality of small blood vessel segments are obtained from the blood vessel.
  • step S200g is performed to divide the small blood vessel segments into upper arches and lower arches
  • step S200h is further performed to perform convex hull detection on the plurality of small blood vessel segments to obtain a plurality of smooth blood vessels segment, remove the smooth blood vessel segments greater than the preset curvature threshold among the multiple smooth blood vessel segments to obtain straight blood vessel segments, use the rectangle detection API in Opencv to detect the straight blood vessel segments to obtain the blood vessel angle, and obtain multiple single blood vessels according to the preset angle interval part.
  • the largest connected blood vessel in the blood vessel extraction image is first retained, and then the branches of the small blood vessels are gradually eliminated.
  • a 3 ⁇ 3 template is used to traverse the grayscale image of the complete blood vessel skeleton, see Figure 10. Since there are three types of intersections and four directions in the 3 ⁇ 3 area, there are twelve types in total.
  • the extraction of all the intersection points in the vascular skeleton can be realized by using the method, and the gray image (the binary image of the blood vessel) is refined by using the 3 ⁇ 3 ELLIPSE type morphological operation Kernel.
  • the state of the refined small branch blood vessels is: completely eroded and there are discontinuities.
  • the preliminary removal of small blood vessel branches can be realized. Further, since the above operations cannot completely remove all small blood vessel branches, here we loop through the intersections of the main blood vessels, and take the intersection as the center and a circular template with a radius of 4 pixels to convert the binary image of the main blood vessels. The pixels in the circular area at the mid-intersection are set to 0, and the grayscale image (the blood vessel binary image) in the circular area is retained. The small area is removed to realize the secondary elimination of the branches of the remaining small blood vessel segments. The image restoration of the intersection position of the blood vessel image is realized by filling the image in the circular area of the position of the blood vessel intersection to the original position. Next, referring to FIG.
  • the irregular blood vessel segment includes: the position of the bifurcation point, the position of the burr, the extremely irregular shape of the blood vessel, and the like.
  • the small blood vessel segment that is qualified for convex hull detection that is, the smooth blood vessel segment, extract its center line, and perform a straight line fitting operation. Set the threshold, and eliminate the small blood vessel segments that are too curved to obtain straight blood vessel segments. Then, through the conditional restriction of the direction of the blood vessel segment, the straight blood vessel segment on the main blood vessel that is contrary to the overall direction of the blood vessel of the main blood vessel is eliminated.
  • the angle measurement of the blood vessel segment is realized through the rectangle detection API in Opencv, and the small blood vessel segment is limited to 5- If it is retained within the range of 85 degrees, a single vessel segment is obtained.
  • step S300 is performed to measure the diameter of a plurality of single blood vessel segments to obtain the diameter width of each single blood vessel segment, and select a plurality of arteriovenous blood vessel pairs according to the width of each tube diameter.
  • step S300a is performed, the diameter of a plurality of single blood vessel segments is measured to obtain the diameter width of each single blood vessel segment, and step S300b is further performed, and selected according to the width of each tube diameter to obtain Multiple arteriovenous vessel pairs include:
  • the first straight line is obtained by straight-line fitting on one contour line of The diameter of the vessel is obtained from the distance of , and the one with the largest diameter among multiple single vessel segments is selected as the vein, the distance between the vein and other single vessel segments is calculated, and the single vessel segment closest to the vein is selected as the artery.
  • the midline of the single vessel segment is cut from the main vessel, and a straight line fitting operation is performed on it, and then the position coordinates of the midline of the vessel segment are located and extended to both sides, see Figure 14 , retain the midline segment of five pixels, calculate the vertical line equation of the blood vessel segment at the two endpoints of the blood vessel midline segment, and intercept the vertical line segment of a fixed length from the vertical line equation of the blood vessel segment to draw the actual effect display.
  • the outline pixel points of a single blood vessel segment are intercepted by two vertical line segments, and then one of the two intercepted blood vessel contours is selected for straight line fitting, and the slope after the straight line fitting of the above single contour is fixed, using The least squares method realizes the straight line fitting of the other contour between the two vertical line segments of the single blood vessel segment, and the diameter measurement can be realized by calculating the distance between the two straight lines, and then traverses the single blood vessel segment in a loop to calculate its diameter respectively.
  • the above-mentioned reason for selecting the midline segment of the five pixel points is based on the setting of the cut length of the actual blood vessel segment, which can be modified according to the actual length of the blood vessel segment.
  • the step S300c is executed to return the building mechanism, and further includes the step of removing unqualified arteriovenous blood vessel pairs from a plurality of arteriovenous blood vessel pairs and retaining qualified arterial and venous blood vessel pairs, wherein the unqualified arteriovenous blood vessel pairs are eliminated. , including: obtaining the average image brightness of a single vessel segment in each arteriovenous vessel pair, and eliminating unqualified arteriovenous vessel pairs from multiple arteriovenous vessel pairs according to the average image brightness. For example, the average image brightness of a single blood vessel segment in an arteriovenous blood vessel pair identified by a calculation algorithm, if the brightness of the arterial blood vessel is higher than that of the venous blood vessel, the arteriovenous blood vessel pair is retained.
  • removing the unqualified arteriovenous blood vessel pair further includes: acquiring the angle between the arterial blood vessel and the venous blood vessel of a single blood vessel in each arteriovenous blood vessel pair, and according to the included angle, a plurality of arteriovenous blood vessel Eliminate unqualified arteriovenous blood vessel pairs in centering, obtain the distance between arterial blood vessels and venous blood vessels of a single blood vessel pair in each arteriovenous blood vessel pair, and eliminate unqualified arteriovenous blood vessel pairs from multiple arteriovenous blood vessel pairs according to the distance.
  • the angle threshold is set to 30 degrees to eliminate the algorithm misjudging a main vessel and its branch vessels as a pair of arteriovenous vessels, that is, if the arterial vessel and the venous vessel in the arteriovenous vessel pair are sandwiched between the arterial and venous vessels. If the angle is greater than 30 degrees, the pair of arterial and venous vessels is removed. Further, if the distance between the arterial vessel segment and the venous vessel segment of a pair of arterial and venous vessel pairs identified by the algorithm is large, the pair of arterial and venous vessel segments is extremely polar. If the probability is not a pair of arteriovenous blood vessels, the arteriovenous blood vessel pair is eliminated, and step S300d is performed to retain the remaining arteriovenous blood vessel pairs.
  • the main blood vessels are extracted according to the blood vessel extraction image, the fundus image and the optic disc center coordinates, so as to obtain the main blood vessel image, and based on the main blood vessel image, the main blood vessel is intercepted to obtain multiple
  • the diameter of a plurality of single blood vessel segments is measured to obtain the diameter width of each single blood vessel segment, and a plurality of arteriovenous blood vessel pairs are selected according to the width of each tube diameter.
  • an apparatus 100 for distinguishing retinal blood vessels from arteries and veins is also provided. Since the working principle of the apparatus 100 for distinguishing retinal blood vessel arteries and veins in the embodiment of the present disclosure is the same as or similar to the principle of the method for distinguishing retinal blood vessel arteries and veins in the embodiment of the present disclosure, the repetition will not be repeated.
  • the apparatus 100 for distinguishing retinal blood vessels from arteries and veins according to an embodiment of the present disclosure includes: a data acquisition module 110, a blood vessel screening module 120, and an arteriovenous blood vessel pair selection module 130,
  • the data acquisition module 110 is configured to acquire the blood vessel extraction image, the fundus image and the coordinates of the center of the optic disc,
  • the blood vessel screening module 120 is configured to extract the main blood vessel according to the blood vessel extraction image, the fundus image and the optic disc center coordinates to obtain the main blood vessel image, and to intercept the main blood vessel based on the main blood vessel image to obtain a plurality of single blood vessel segments,
  • the arteriovenous blood vessel pair selection module 130 is configured to measure the diameter of a plurality of single blood vessel segments to obtain the diameter width of each single blood vessel segment, and select a plurality of arteriovenous blood vessel pairs according to the diameter width of each tube.
  • a retinal blood vessel arteriovenous discriminating device 200 is also provided.
  • an apparatus 200 for distinguishing retinal blood vessels from arteries and veins includes a processor 210 and a memory 220 for storing instructions executable by the processor 210 .
  • the processor 210 is configured to implement any of the foregoing retinal blood vessel arteriovenous differentiation methods when executing the executable instructions.
  • the number of processors 210 may be one or more.
  • the apparatus 200 for distinguishing retinal blood vessels and arteries and veins in the embodiment of the present disclosure may further include an input device 230 and an output device 240 .
  • the processor 210, the memory 220, the input device 230, and the output device 240 may be connected through a bus, or may be connected in other ways, which are not specifically limited here.
  • the memory 220 can be used to store software programs, computer-executable programs, and various modules, such as programs or modules corresponding to the retinal blood vessel arteriovenous differentiation method according to the embodiment of the present disclosure.
  • the processor 210 executes various functional applications and data processing of the retinal blood vessel arteriovenous discriminating apparatus 200 by running the software programs or modules stored in the memory 220 .
  • the input device 230 may be used to receive input numbers or signals. Wherein, the signal may be the generation of a key signal related to user setting and function control of the device/terminal/server.
  • the output device 240 may include a display device such as a display screen.
  • a non-volatile computer-readable storage medium on which computer program instructions are stored, and when the computer program instructions are executed by the processor 210, any of the foregoing retinal blood vessel arteriovenous distinctions are implemented method.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Geometry (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Medical Informatics (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Vascular Medicine (AREA)
  • Eye Examination Apparatus (AREA)
  • Image Analysis (AREA)

Abstract

一种视网膜血管动静脉区分方法和装置、设备。方法包括，获取血管提取图像、眼底图像和视盘中心坐标（S100）；依据血管提取图像、眼底图像和视盘中心坐标进行主干血管的提取，得到主干血管图像，并基于主干血管图像对主干血管进行截取得到多个单一血管段（S200）；将多个单一血管段进行管径测量得到各单一血管段的管径宽度，并根据各管径宽度选取得到多个动静脉血管对（S300）。以使在实际场景眼底视网膜彩色图像上测试效果表现优异，对于不同类型、不同品牌、不同等级图像质量的眼底图像皆有较好效果，算法鲁棒性和普适性较强，为视网膜血管动静脉区分及管径测量算法实际场景落地提供可能。

Description

视网膜血管动静脉区分方法和装置、设备
本申请要求于2020年08月14日提交中国专利局、申请号为202010815698.8、发明名称为“视网膜血管动静脉区分方法和装置、设备”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本公开涉及计算机辅助诊断技术领域，尤其涉及一种视网膜血管动静脉区分方法和装置、设备。
背景技术
眼底视网膜血管动静脉管径比指标是医生在诊断动脉硬化疾病时的重要参考指标。然而医生实际阅片时，依旧凭借目测的方法，给出动静脉管径比的一个大致比值：大于1/2、大于1/3且小于1/2、小于1/3等。然而目测的方法毕竟存在较大误差，尤其对于新晋眼科医生或普通眼科医生在阅片过程中难免存在误判等情况。因此，通过计算机准确测量出视网膜主干血管动静脉管径比的比值对于医生临床诊断动脉硬化具有极大帮助，可有效提高医生阅片效率。另外，对于进一步实现计算机辅助诊断动脉硬化等存在血管形态异常表现的疾病同样具有重大意义。
基于机器学习方法区分视网膜血管动静脉的方法存在算法普适性和鲁棒性差问题。
发明内容
有鉴于此，本公开提出了一种视网膜血管动静脉区分方法，包括：
获取血管提取图像、眼底图像和视盘中心坐标；
依据所述血管提取图像、所述眼底图像和所述视盘中心坐标进行主干血管的提取，得到主干血管图像，并基于所述主干血管图像对所述主干血管进行截取得到多个单一血管段；
将多个所述单一血管段进行管径测量得到各所述单一血管段的管径 宽度，并根据各所述管径宽度选取得到多个动静脉血管对；
由多个所述动静脉血管对中剔除不合格动静脉血管对，保留合格动静脉血管对的步骤；
其中，剔除不合格动静脉血管对，包括：
获取各所述动静脉血管对中单一血管段的平均图像亮度，并根据所述平均图像亮度由多个所述动静脉血管对中剔除所述不合格动静脉血管对。
在一种可能的实现方式中，还包括由多个所述动静脉血管对中剔除不合格动静脉血管对，保留合格动静脉血管对的步骤；
其中，剔除不合格动静脉血管对，包括：
获取各所述动静脉血管对中单一血管段的平均图像亮度，并根据所述平均图像亮度由多个所述动静脉血管对中剔除所述不合格动静脉血管对。
在一种可能的实现方式中，依据所述血管提取图像、所述眼底图像和所述视盘中心坐标进行主干血管的提取，包括：
去除所述血管提取图像中的断开血管得到最大的连通血管；
依据所述连通血管提取出血管骨架；
依据所述血管骨架提取出血管的交叉点并将所述交叉点去除；
依据所述视盘中心坐标和所述眼底图像将所述血管骨架的细小血管段和非主干血管分支去除得到主干血管；
其中，依据所述连通血管提取出血管骨架时，使用Opencv中的骨架提取函数将所述连通血管提取出所述血管骨架。
在一种可能的实现方式中，剔除不合格动静脉血管对，还包括：
获取各所述动静脉血管对中单一血管对中动脉血管和静脉血管之间的夹角，根据所述夹角由多个所述动静脉血管对中剔除所述不合格动静脉血管对；
获取各所述动静脉血管对中单一血管对中动脉血管和静脉血管之间的距离，根据所述距离由多个所述动静脉血管对中剔除所述不合格动静脉血管对。
在一种可能的实现方式中，依据所述血管骨架提取出血管的交叉点并 将所述交叉点去除包括：
将所述血管骨架进行二值化得到灰度图像；
以3×3模板遍历所述灰度图像，通过枚举法提取所述灰度图像中的所述交叉点；
以所述交叉点为圆心，将半径为四个像素点内的像素值设置为0。
在一种可能的实现方式中，依据所述视盘中心坐标和所述眼底图像将所述血管骨架的细小血管段和非主干血管分支去除得到主干血管包括：
将所述视盘中心坐标和所述眼底图像的黄斑区连线为正半轴；
保留以所述正半轴的顺时针110°和所述正半轴的逆时针110°的区域；
采用3×3ELLIPSE对所述灰度图像进行细化得到主干血管。
在一种可能的实现方式中，基于所述主干血管图像对所述主干血管进行截取得到多个单一血管段，包括：
以所述视盘中心坐标为圆心，在预设半径内以预设步长向外遍历截取所述主干血管得到多个小血管段；
对多个所述小血管段进行凸包检测得到多个光滑血管段；
去除多个所述光滑血管段中大于预设弯曲度阈值的所述光滑血管段得到直线血管段；
将所述直线血管段使用Opencv中矩形检测API进行检测得到血管角度；
依据预设角度区间得到多个所述单一血管段。
在一种可能的实现方式中，将多个所述单一血管段进行管径测量得到各所述单一血管段的管径宽度，并根据各所述管径宽度选取得到多个动静脉血管对包括：
取各所述单一血管段的中线，将所述中线的中点向所述单一血管段的两端延伸预设像素得到中线线段；
计算所述中线线段的两端端点的垂线方程；
将所述垂线方程之间的所述单一血管段的一条轮廓线进行直线拟合 得到第一直线；
基于所述第一直线的斜率采用最小二乘法对另一条所述轮廓线进行直线拟合得到第二直线；
计算所述第一直线与第二直线之间的距离得到血管管径；
选择多个所述单一血管段中管径最大的作为静脉；
计算所述静脉到其他单一血管段之间的距离，选取距离所述静脉最近的所述单一血管段作为动脉。
根据本公开的另一方面，提供了一种视网膜血管动静脉区分装置，包括数据获取模块、血管筛选模块和动静脉血管对选取模块；
所述数据获取模块，被配置为获取血管提取图像、眼底图像和视盘中心坐标；
所述血管筛选模块，被配置为依据所述血管提取图像、所述眼底图像和所述视盘中心坐标进行主干血管的提取，得到主干血管图像，并基于所述主干血管图像对所述主干血管进行截取得到多个单一血管段；
所述动静脉血管对选取模块，被配置为将多个所述单一血管段进行管径测量得到各所述单一血管段的管径宽度，并根据各所述管径宽度选取得到多个动静脉血管对；
由多个所述动静脉血管对中剔除不合格动静脉血管对，保留合格动静脉血管对的步骤；
其中，剔除不合格动静脉血管对，包括：
获取各所述动静脉血管对中单一血管段的平均图像亮度，并根据所述平均图像亮度由多个所述动静脉血管对中剔除所述不合格动静脉血管对。
根据本公开的另一方面，提供了一种视网膜血管动静脉区分设备，包括：
处理器；
用于存储处理器可执行指令的存储器；
其中，所述处理器被配置为执行所述可执行指令时实现前面任一所述的方法。
通过获取血管提取图像、眼底图像和视盘中心坐标，依据血管提取图像、眼底图像和视盘中心坐标进行主干血管的提取，得到主干血管图像，并基于主干血管图像对主干血管进行截取得到多个单一血管段，将多个单一血管段进行管径测量得到各单一血管段的管径宽度，并根据各管径宽度选取得到多个动静脉血管对。以使在实际场景眼底视网膜彩色图像上测试效果表现优异，对于不同类型、不同品牌、不同等级图像质量的眼底图像皆有较好效果，算法鲁棒性和普适性较强，为视网膜血管动静脉区分及管径测量算法实际场景落地提供可能。
根据下面参考附图对示例性实施例的详细说明，本公开的其它特征及方面将变得清楚。
说明书附图
包含在说明书中并且构成说明书的一部分的附图与说明书一起示出了本公开的示例性实施例、特征和方面，并且用于解释本公开的原理。
图1示出本公开实施例的视网膜血管动静脉区分方法的流程图；
图2示出本公开实施例的视网膜血管动静脉区分方法的512×512尺寸眼底图像示意图；
图3示出本公开实施例的视网膜血管动静脉区分方法的512×512尺寸血管提取图像示意图；
图4示出本公开实施例的视网膜血管动静脉区分方法的血管提取测试图像示意图；
图5示出本公开实施例的视网膜血管动静脉区分方法的血管提取测试结果图像示意图；
图6示出本公开实施例的视网膜血管动静脉区分方法的另一流程图；
图7示出本公开实施例的视网膜血管动静脉区分方法的眼底图像示意图；
图8示出本公开实施例的视网膜血管动静脉区分方法的视盘检测测试图像示意图；
图9示出本公开实施例的视网膜血管动静脉区分方法的视盘检测测试结果图像示意图；
图10示出本公开实施例的视网膜血管动静脉区分方法的血管走向示意图；
图11示出本公开实施例的视网膜血管动静脉区分方法的黄斑区示意图；
图12示出本公开实施例的视网膜血管动静脉区分方法的圆环模板示意图；
图13示出本公开实施例的视网膜血管动静脉区分方法的小血管段示意图；
图14示出本公开实施例的视网膜血管动静脉区分方法的轮廓直线拟合示意图；
图15示出本公开实施例的视网膜血管动静脉区分方法的管径示意图；
图16示出本公开实施例的视网膜血管动静脉区分装置的框图；
图17示出本公开实施例的视网膜血管动静脉区分设备的框图。
具体实施方式
以下将参考附图详细说明本公开的各种示例性实施例、特征和方面。附图中相同的附图标记表示功能相同或相似的元件。尽管在附图中示出了实施例的各种方面，但是除非特别指出，不必按比例绘制附图。
在这里专用的词“示例性”意为“用作例子、实施例或说明性”。这里作为“示例性”所说明的任何实施例不必解释为优于或好于其它实施例。
另外，为了更好的说明本公开，在下文的具体实施方式中给出了众多的具体细节。本领域技术人员应当理解，没有某些具体细节，本公开同样可以实施。在一些实例中，对于本领域技术人员熟知的方法、手段、元件和电路未作详细描述，以便于凸显本公开的主旨。
图1示出根据本公开一实施例的视网膜血管动静脉区分方法的流程图。如图1所示，该视网膜血管动静脉区分方法包括：
步骤S100，获取血管提取图像、眼底图像和视盘中心坐标，步骤S200，依据血管提取图像、眼底图像和视盘中心坐标进行主干血管的提取，得到主干血管图像，并基于主干血管图像对主干血管进行截取得到多个单一血管段，步骤S300，将多个单一血管段进行管径测量得到各单一血管段的管径宽度，并根据各管径宽度选取得到多个动静脉血管对。
通过获取血管提取图像、眼底图像和视盘中心坐标，依据血管提取图像、眼底图像和视盘中心坐标进行主干血管的提取，得到主干血管图像，并基于主干血管图像对主干血管进行截取得到多个单一血管段，将多个单一血管段进行管径测量得到各单一血管段的管径宽度，并根据各管径宽度选取得到多个动静脉血管对。以使在实际场景眼底视网膜彩色图像上测试效果表现优异，对于不同类型、不同品牌、不同等级图像质量的眼底图像皆有较好效果，算法鲁棒性和普适性较强，为视网膜血管动静脉区分及管径测量算法实际场景落地提供可能。
首先，需要说明的是，本申请实施例的视网膜血管动静脉区分方法是基于机器学习来实现的。其中，对应本申请实施例的视网膜血管动静脉区分方法中，血管提取图像和视盘中心坐标的获取可以采用相应的网络模型来实现。
具体的，血管提取图像可以采用U-net网络结构来实现，视盘中心坐标同样也可以采用U-net网络结构来实现。
在采用U-net网络结构进行血管提取图像和视盘中心坐标的获取之前，需要先对所采用的网络模型进行训练，得到最终收敛后的网络结构。
其中，在对所采用的网络模型进行训练时，首先构建血管提取和视盘检测的训练数据集，在进行血管提取的训练数据集的构建中，使用DRIVE公开数据集中训练集20张及测试集18张，数据扩增后共3004张，测试集2张，共2张，其中图片均为565×584像素，在进行视盘检测的训练数据集的构建中，训练集为refuge2018训练集及验证集800张，drishti-gsl训练数据集50张、rim-one-r3训练数据集159张，数据扩增后共4036张，测试集包括refuge2018测试集400张，drishti-gsl测试集51张，共451张，进一步的，在血管提取层中，首先对训练数据集进行黑边裁剪，然后提取 G通道，然后使用CLAHE(限制对比度自适应直方图均衡)图像增强，接着进行归一化，将图像水平反转、对比度调整和随机裁剪，将训练数据集扩增至3004张，其中，血管提取层的模型训练的网络结构为U-Net，输入尺寸为512×512，优化方法为Adam，学习率为0.0001，Batch_size为2，损失函数为dice_lose，训练次数为150epoch，并且添加early_stopping机制，模型测试图像参见图6，模型结果图像参见图7。同样的，在视盘检测层中，首先对训练集数据图像进行黑边裁剪，然后进行中值滤波，将图像缩放至512×512尺寸，接着提取G通道，然后使用CLAHE(限制对比度自适应直方图均衡)图像增强，接着进行归一化，将图像水平反转和对比度调整，将训练数据集扩增至4036张，视盘检测层的模型训练的网络结构为U-Net，输入尺寸为512×512，优化方法为Adam，学习率为0.0001，Batch_size为4，损失函数为dice_lose，训练次数为150epoch，并且添加early_stopping机制，模型测试图像参见图8，测试结果图像参见图9。
在对网络模型训练完成之后，即可采用训练好的网络模型进行血管提取图像和视盘中心坐标的获取。即，参见图1，执行步骤S100，获取血管提取图像、眼底图像和视盘中心坐标。
在一种可能的实现方式中，首先输入一张眼底视网膜彩色图像，参见图3并对其进行预处理操作，之后将其分别进行血管提取和视盘检测获得血管提取结果和视盘检测结果，上述操作全部完成后，将预处理后的彩色眼底图像、视网膜血管提取图像、视盘坐标作为输入送至动静脉区分模块进行静动脉的区分，举例来说，参见图2，执行步骤S100，参见图4和图5，可以直接从血管提取结果和视盘检测结果得到512×512尺寸且经过黑边裁剪后的彩色的视网膜眼底图像、512×512尺寸的血管提取图像以及基于512×512尺寸图像上视盘中心坐标。
进一步的，参见图1，执行步骤S200，依据血管提取图像、眼底图像和视盘中心坐标进行主干血管的提取，得到主干血管图像，并基于主干血管图像对主干血管进行截取得到多个单一血管段。
在一种可能的实现方式中，依据血管提取图像、眼底图像和视盘中心坐标进行主干血管的提取，包括：参见图2，执行步骤S200a，去除血管 提取图像中的断开血管得到最大的连通血管，步骤S200b，依据连通血管提取出血管骨架，步骤S200c、步骤S200d以及步骤S200e，依据血管骨架提取出血管的交叉点并将交叉点去除，依据视盘中心坐标和眼底图像将血管骨架的细小血管段和非主干血管分支去除得到主干血管，其中，依据连通血管提取出血管骨架时，使用Opencv中的骨架提取函数将连通血管提取出血管骨架。
具体的，依据血管骨架提取出血管的交叉点并将交叉点去除包括：将血管骨架进行二值化得到灰度图像，以3×3模板遍历灰度图像，通过枚举法提取灰度图像中的交叉点，以交叉点为圆心，将半径为四个像素点内的像素值设置为0。
进一步的，依据视盘中心坐标和眼底图像将血管骨架的细小血管段和非主干血管分支去除得到主干血管包括：将视盘中心坐标和眼底图像的黄斑区连线为正半轴，保留以正半轴的顺时针110°和正半轴的逆时针110°的区域，采用3×3ELLIPSE对灰度图像进行细化得到主干血管。
进一步的，基于主干血管图像对主干血管进行截取得到多个单一血管段，包括：参见图2，步骤S200f，以视盘中心坐标为圆心，在预设半径内以预设步长向外遍历截取主干血管得到多个小血管段，为保证精细化剔除，执行步骤S200g，将小血管段分为上弓和下弓，进一步执行步骤S200h，对多个小血管段进行凸包检测得到多个光滑血管段，去除多个光滑血管段中大于预设弯曲度阈值的光滑血管段得到直线血管段，将直线血管段使用Opencv中矩形检测API进行检测得到血管角度，依据预设角度区间得到多个单一血管段。
示例性的，在进行主干血管的提取时，首先保留血管提取图像中最大的连通血管，再逐步剔除细小血管的分支，其中，通过Opencv形态学操作morphology库中骨架提取函数skeletonize()实现对血管骨架的提取，采用3×3模板遍历完整血管骨架的灰度图像，参见图10，由于交叉点在3×3区域内呈现状态有三种类型及四个方向，共十二种类型，故通过枚举方法即可实现血管骨架中全部交叉点的提取，结合采用3×3ELLIPSE类型形态学操作Kernel对灰度图像(血管二值图像)细化，细化后的细小分支血 管状态为：完全腐蚀掉和存在间断。此时结合移除小区域操作即可实现细小血管分支的初步剔除。进一步的，由于上述操作无法彻底移除全部细小血管分支，故此处我们循环遍历主干血管的交叉点，并以交叉点为圆心，以4个像素点为半径的圆形模板将主干血管二值图像中交叉点位置圆形区域内像素点置0，并保留此圆环区域内的灰度图像(血管二值图像)，通过移除小区域操作实现遗留细小血管段分支的二次剔除，之后再通过保留的血管交叉点位置圆形区域内图像填充至原始位置，实现血管图像交叉点位置图像还原。接着，参见图11，选取视盘中心坐标，以视盘中心坐标和黄斑区连线为横轴正半轴，截取上下0-110度区域血管，进一步剔除残余的视盘侧非主干血管分支即得到主干血管，接着以视盘中心坐标为圆心，固定步长圆环区域向外遍历，在剔除交叉点后的主干血管上截取小血管段。圆环截取完整范围参见图12中的圆环围成区域，其中大圆环表示圆环截取范围的下边界，小圆环表示圆环截取范围的上边界，圆环遍历过程中截取的小血管段参见图13所示。进一步的，还需要对小血管段进行凸包检测，即对提取出来的小血管段计算其凸包，并分别统计凸包轮廓像素点及该血管段像素点个数，通过计算两者之间的差值，并设定阈值实现非规则血管段的剔除得到光滑血管段。其中，非规则血管段包括：分叉点位置、存在毛刺位置、血管形状极其不规则等情况。接着，对凸包检测合格的小血管段即光滑血管段，提取其中心线，并进行直线拟合操作，通过计算中心线上像素点到拟合直线之间距离的累加和及偏差，并设定阈值，剔除过于弯曲的小血管段得到直线血管段。再通过血管段走向的条件限制，剔除主干血管上的与主干血管的血管整体走向相悖的直线血管段，具体的，通过Opencv中矩形检测API实现血管段角度测量，并限制小血管段在5-85度区间范围内时保留，即得到了单一血管段。另外，还可以进行血管段长度限制剔除圆环模板截取出的由于剔除交叉点操作导致不完整的小血管段。
进一步的，执行步骤S300，将多个单一血管段进行管径测量得到各单一血管段的管径宽度，并根据各管径宽度选取得到多个动静脉血管对。
在一种可能的实现方式中，参见图2，执行步骤S300a，将多个单一血管段进行管径测量得到各单一血管段的管径宽度，进一步执行步骤S300b，并根据各管径宽度选取得到多个动静脉血管对包括：
取各单一血管段的中线，将中线的中点向单一血管段的两端延伸预设像素得到中线线段，计算中线线段的两端端点的垂线方程，将垂线方程之间的单一血管段的一条轮廓线进行直线拟合得到第一直线，基于第一直线的斜率采用最小二乘法对另一条轮廓线进行直线拟合得到第二直线，计算第一直线与第二直线之间的距离得到血管管径，选择多个单一血管段中管径最大的作为静脉，计算静脉到其他单一血管段之间的距离，选取距离静脉最近的单一血管段作为动脉。举例来说，通过灰度图像作为模板，在主干血管上截取该单一血管段中线，并对其进行直线拟合操作，之后定位该血管段中线中点位置坐标，向两侧延伸，参见图14，保留五个像素点的中线线段，分别在血管中线线段的两个端点处计算血管段的垂线方程，并分别在血管段垂线方程上截取固定长度的垂线段绘制图中进行实际效果展示，进一步的，参见图15，通过两条垂线段截取单一血管段轮廓像素点，再对截取出来的两条血管轮廓选择其中一条进行直线拟合，固定上述单一轮廓直线拟合后的斜率，采用最小二乘法实现该单一血管段两条垂线段之间另一条轮廓的直线拟合，通过计算两条直线之间的距离即可实现管径测量，接着循环遍历单一血管段，分别计算其管径值，选取管径值最大的假设其为静脉血管，计算该静脉血管段中心点到其他血管段中心点的距离，选取距离静脉血管段最近的单一血管段假设其为动脉，两个血管段构成一对动静脉血管对，保存相关数据。
需要说明的是，上述选择五个像素点的中线线段的原因是基于实际血管段截取长度设定，可根据实际血管段长度对其进行修改。
另外的，参见图2，执行步骤S300c回楼机制，还包括由多个动静脉血管对中剔除不合格动静脉血管对，保留合格动静脉血管对的步骤，其中，剔除不合格动静脉血管对，包括：获取各动静脉血管对中单一血管段的平均图像亮度，并根据平均图像亮度由多个动静脉血管对中剔除不合格动静脉血管对。举例来说，通过计算算法识别出的动静脉血管对中单一血管段 的平均图像亮度，若其中动脉血管的亮度高于静脉血管，即保留该动静脉血管对。
需要说明的是，亮度的计算算法可以使用现有技术进行计算，此处不再进行赘述。
在一种可能的实现方式中，剔除不合格动静脉血管对还包括：获取各动静脉血管对中单一血管对中动脉血管和静脉血管之间的夹角，根据夹角由多个动静脉血管对中剔除不合格动静脉血管对，获取各动静脉血管对中单一血管对中动脉血管和静脉血管之间的距离，根据距离由多个动静脉血管对中剔除不合格动静脉血管对。举例来说，设定角度阈值为30度来剔除算法将一个主干血管和其分支血管误判为一对动静脉血管对情况，即若动静脉血管对中的动脉血管和静脉血管之间的夹角大于30度，去除这一对动静脉血管对，进一步的，若算法识别出的一对动静脉血管对中动脉血管段和静脉血管段之间距离较大，则该对动静脉血管段极大概率为非一对动静脉血管，则将该动静脉血管对剔除，并执行步骤S300d，保留其余的动静脉血管对。
需要说明的是，尽管以上述各个步骤作为示例介绍了如上所述的视网膜血管动静脉区分方法，但本领域技术人员能够理解，本公开应不限于此。事实上，用户完全可根据个人喜好和/或实际应用场景灵活设定视网膜血管动静脉区分方法，只要达到所需功能即可。
这样，通过获取血管提取图像、眼底图像和视盘中心坐标，依据血管提取图像、眼底图像和视盘中心坐标进行主干血管的提取，得到主干血管图像，并基于主干血管图像对主干血管进行截取得到多个单一血管段，将多个单一血管段进行管径测量得到各单一血管段的管径宽度，并根据各管径宽度选取得到多个动静脉血管对。以使在实际场景眼底视网膜彩色图像上测试效果表现优异，对于不同类型、不同品牌、不同等级图像质量的眼底图像皆有较好效果，算法鲁棒性和普适性较强，为视网膜血管动静脉区分及管径测量算法实际场景落地提供可能。
进一步的，根据本公开的另一方面，还提供了一种视网膜血管动静脉区分装置100。由于本公开实施例的视网膜血管动静脉区分装置100的工 作原理与本公开实施例的视网膜血管动静脉区分方法的原理相同或相似，因此重复之处不再赘述。参见图16，本公开实施例的视网膜血管动静脉区分装置100包括：数据获取模块110、血管筛选模块120和动静脉血管对选取模块130，
数据获取模块110，被配置为获取血管提取图像、眼底图像和视盘中心坐标，
血管筛选模块120，被配置为依据血管提取图像、眼底图像和视盘中心坐标进行主干血管的提取，得到主干血管图像，并基于主干血管图像对主干血管进行截取得到多个单一血管段，
动静脉血管对选取模块130，被配置为将多个单一血管段进行管径测量得到各单一血管段的管径宽度，并根据各管径宽度选取得到多个动静脉血管对。
更进一步地，根据本公开的另一方面，还提供了一种视网膜血管动静脉区分设备200。参阅图17，本公开实施例视网膜血管动静脉区分设备200包括处理器210以及用于存储处理器210可执行指令的存储器220。其中，处理器210被配置为执行可执行指令时实现前面任一的视网膜血管动静脉区分方法。
此处，应当指出的是，处理器210的个数可以为一个或多个。同时，在本公开实施例的视网膜血管动静脉区分设备200中，还可以包括输入装置230和输出装置240。其中，处理器210、存储器220、输入装置230和输出装置240之间可以通过总线连接，也可以通过其他方式连接，此处不进行具体限定。
存储器220作为一种计算机可读存储介质，可用于存储软件程序、计算机可执行程序和各种模块，如：本公开实施例的视网膜血管动静脉区分方法所对应的程序或模块。处理器210通过运行存储在存储器220中的软件程序或模块，从而执行视网膜血管动静脉区分设备200的各种功能应用及数据处理。
输入装置230可用于接收输入的数字或信号。其中，信号可以为产生与设备/终端/服务器的用户设置以及功能控制有关的键信号。输出装置240可以包括显示屏等显示设备。
根据本公开的另一方面，还提供了一种非易失性计算机可读存储介质，其上存储有计算机程序指令，计算机程序指令被处理器210执行时实现前面任一的视网膜血管动静脉区分方法。
以上已经描述了本公开的各实施例，上述说明是示例性的，并非穷尽性的，并且也不限于所披露的各实施例。在不偏离所说明的各实施例的范围和精神的情况下，对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。本文中所用术语的选择，旨在最好地解释各实施例的原理、实际应用或对市场中的技术改进，或者使本技术领域的其它普通技术人员能理解本文披露的各实施例。

Claims (9)

  1. 一种视网膜血管动静脉区分方法，其特征在于，包括：
    获取血管提取图像、眼底图像和视盘中心坐标；
    依据所述血管提取图像、所述眼底图像和所述视盘中心坐标进行主干血管的提取，得到主干血管图像，并基于所述主干血管图像对所述主干血管进行截取得到多个单一血管段；
    将多个所述单一血管段进行管径测量得到各所述单一血管段的管径宽度，并根据各所述管径宽度选取得到多个动静脉血管对；
    由多个所述动静脉血管对中剔除不合格动静脉血管对，保留合格动静脉血管对的步骤；
    其中，剔除不合格动静脉血管对，包括：
    获取各所述动静脉血管对中单一血管段的平均图像亮度，并根据所述平均图像亮度由多个所述动静脉血管对中剔除所述不合格动静脉血管对。
  2. 根据权利要求1所述的方法，其特征在于，依据所述血管提取图像、所述眼底图像和所述视盘中心坐标进行主干血管的提取，包括：
    去除所述血管提取图像中的断开血管得到最大的连通血管；
    依据所述连通血管提取出血管骨架；
    依据所述血管骨架提取出血管的交叉点并将所述交叉点去除；
    依据所述视盘中心坐标和所述眼底图像将所述血管骨架的细小血管段和非主干血管分支去除得到主干血管；
    其中，依据所述连通血管提取出血管骨架时，使用Opencv中的骨架提取函数将所述连通血管提取出所述血管骨架。
  3. 根据权利要求1所述的方法，其特征在于，剔除不合格动静脉血管对，还包括：
    获取各所述动静脉血管对中单一血管对中动脉血管和静脉血管之间的夹角，根据所述夹角由多个所述动静脉血管对中剔除所述不合格动静脉血管对；
    获取各所述动静脉血管对中单一血管对中动脉血管和静脉血管之间的距离，根据所述距离由多个所述动静脉血管对中剔除所述不合格动静脉血管对。
  4. 根据权利要求2所述的方法，其特征在于，依据所述血管骨架提取出血管的交叉点并将所述交叉点去除包括：
    将所述血管骨架进行二值化得到灰度图像；
    以3×3模板遍历所述灰度图像，通过枚举法提取所述灰度图像中的所述交叉点；
    以所述交叉点为圆心，将半径为四个像素点内的像素值设置为0。
  5. 根据权利要求4所述的方法，其特征在于，依据所述视盘中心坐标和所述眼底图像将所述血管骨架的细小血管段和非主干血管分支去除得到主干血管包括：
    将所述视盘中心坐标和所述眼底图像的黄斑区连线为正半轴；
    保留以所述正半轴的顺时针110°和所述正半轴的逆时针110°的区域；
    采用3×3ELLIPSE对所述灰度图像进行细化得到主干血管。
  6. 根据权利要求1所述的方法，其特征在于，基于所述主干血管图像对所述主干血管进行截取得到多个单一血管段，包括：
    以所述视盘中心坐标为圆心，在预设半径内以预设步长向外遍历截取所述主干血管得到多个小血管段；
    对多个所述小血管段进行凸包检测得到多个光滑血管段；
    去除多个所述光滑血管段中大于预设弯曲度阈值的所述光滑血管段得到直线血管段；
    将所述直线血管段使用Opencv中矩形检测API进行检测得到血管角度；
    依据预设角度区间得到多个所述单一血管段。
  7. 根据权利要求1所述的方法，其特征在于，将多个所述单一血管段进行管径测量得到各所述单一血管段的管径宽度，并根据各所述管径宽度选取得到多个动静脉血管对包括：
    取各所述单一血管段的中线，将所述中线的中点向所述单一血管段的两端延伸预设像素得到中线线段；
    计算所述中线线段的两端端点的垂线方程；
    将所述垂线方程之间的所述单一血管段的一条轮廓线进行直线拟合得到第一直线；
    基于所述第一直线的斜率采用最小二乘法对另一条所述轮廓线进行直线拟合得到第二直线；
    计算所述第一直线与第二直线之间的距离得到血管管径；
    选择多个所述单一血管段中管径最大的作为静脉；
    计算所述静脉到其他单一血管段之间的距离，选取距离所述静脉最近的所述单一血管段作为动脉。
  8. 一种视网膜血管动静脉区分装置，其特征在于，包括数据获取模块、血管筛选模块和动静脉血管对选取模块；
    所述数据获取模块，被配置为获取血管提取图像、眼底图像和视盘中心坐标；
    所述血管筛选模块，被配置为依据所述血管提取图像、所述眼底图像和所述视盘中心坐标进行主干血管的提取，得到主干血管图像，并基于所述主干血管图像对所述主干血管进行截取得到多个单一血管段；
    所述动静脉血管对选取模块，被配置为将多个所述单一血管段进行管径测量得到各所述单一血管段的管径宽度，并根据各所述管径宽度选取得到多个动静脉血管对；
  9. 一种视网膜血管动静脉区分设备，其特征在于，包括：
    处理器；
    用于存储处理器可执行指令的存储器；
    其中，所述处理器被配置为执行所述可执行指令时实现权利要求1至7中任意一项所述的方法。
PCT/CN2021/112487 2020-08-14 2021-08-13 视网膜血管动静脉区分方法和装置、设备 Ceased WO2022033580A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
JP2022512438A JP7333465B2 (ja) 2020-08-14 2021-08-13 網膜血管の動静脈を判別する方法及び装置
US17/755,560 US12307663B2 (en) 2020-08-14 2021-08-13 Method, device and apparatus for differentiating arteries and veins of retinal vessels
KR1020227015077A KR102826940B1 (ko) 2020-08-14 2021-08-13 망막 혈관 동정맥 구분 방법 및 장치, 기기

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010815698.8A CN111681242B (zh) 2020-08-14 2020-08-14 视网膜血管动静脉区分方法和装置、设备
CN202010815698.8 2020-08-14

Publications (1)

Publication Number Publication Date
WO2022033580A1 true WO2022033580A1 (zh) 2022-02-17

Family

ID=72438772

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/112487 Ceased WO2022033580A1 (zh) 2020-08-14 2021-08-13 视网膜血管动静脉区分方法和装置、设备

Country Status (5)

Country Link
US (1) US12307663B2 (zh)
JP (1) JP7333465B2 (zh)
KR (1) KR102826940B1 (zh)
CN (1) CN111681242B (zh)
WO (1) WO2022033580A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114973207A (zh) * 2022-08-01 2022-08-30 成都航空职业技术学院 一种基于目标检测的路标识别方法

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111681242B (zh) * 2020-08-14 2020-11-17 北京至真互联网技术有限公司 视网膜血管动静脉区分方法和装置、设备
CN111932535B (zh) * 2020-09-24 2024-12-17 北京百度网讯科技有限公司 用于处理图像的方法、装置、设备以及存储介质
CN112652011B (zh) * 2020-12-25 2023-04-14 北京阅影科技有限公司 目标血管的提取方法、提取装置与计算机可读存储介质
CN116236150A (zh) * 2020-12-28 2023-06-09 深圳硅基智能科技有限公司 基于眼底图像的动静脉血管图像分割方法
CN113269737B (zh) * 2021-05-17 2024-03-19 北京鹰瞳科技发展股份有限公司 一种眼底视网膜动静脉血管直径计算方法及系统
CN113344897B (zh) * 2021-06-24 2022-01-11 推想医疗科技股份有限公司 肺部图像的管径测量方法及装置、图像处理方法及装置
CN114549395B (zh) * 2021-12-14 2025-03-21 依未科技(北京)有限公司 基于视计算的图像特征确定的方法、装置、介质和设备
CN115760873B (zh) * 2022-11-08 2025-10-28 温州谱希医学检验实验室有限公司 基于区域分割的眼底照视网膜血管管径的计算方法
CN118822983B (zh) * 2024-07-03 2025-04-04 山东大学齐鲁医院 用于医学图像中目标网络的主干识别方法及相关系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040230124A1 (en) * 2003-05-12 2004-11-18 Querfurth Henry W. Methods of and systems and devices for assessing intracranial pressure non-invasively
US20080140192A1 (en) * 2006-11-29 2008-06-12 University Of Southern California Reversible thermoresponsive adhesives for implants
CN101999885A (zh) * 2010-12-21 2011-04-06 中国人民解放军国防科学技术大学 一种自动分离动静脉血管的内源光学成像方法
CN108073918A (zh) * 2018-01-26 2018-05-25 浙江大学 眼底视网膜的血管动静脉交叉压迫特征提取方法
CN108765418A (zh) * 2018-06-14 2018-11-06 四川和生视界医药技术开发有限公司 视网膜动静脉等效管径比检测方法及检测装置
CN110874597A (zh) * 2018-08-31 2020-03-10 福州依影健康科技有限公司 一种眼底图像血管特征提取方法、设备、系统和存储介质
CN111681242A (zh) * 2020-08-14 2020-09-18 北京至真互联网技术有限公司 视网膜血管动静脉区分方法和装置、设备

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH10243924A (ja) * 1997-03-05 1998-09-14 Nippon Telegr & Teleph Corp <Ntt> 眼底の動静脈径比の計測方法
JP2007097740A (ja) * 2005-10-03 2007-04-19 Hitachi Omron Terminal Solutions Corp 眼底画像診断支援装置
JP2008022928A (ja) * 2006-07-19 2008-02-07 Gifu Univ 画像解析装置及び画像解析プログラム
US8687862B2 (en) * 2008-04-08 2014-04-01 National University Of Singapore Retinal image analysis systems and methods
CA2825169A1 (en) * 2011-01-20 2012-07-26 University Of Iowa Research Foundation Automated determination of arteriovenous ratio in images of blood vessels
US8355544B2 (en) * 2011-02-01 2013-01-15 Universidade Da Coruna-Otri Method, apparatus, and system for automatic retinal image analysis
FR2970858B1 (fr) * 2011-02-01 2013-02-08 Imagine Eyes Methode et dispositif d'imagerie retinienne a haute resolution
US9089288B2 (en) * 2011-03-31 2015-07-28 The Hong Kong Polytechnic University Apparatus and method for non-invasive diabetic retinopathy detection and monitoring
TWI578977B (zh) * 2011-04-07 2017-04-21 香港中文大學 視網膜圖像分析裝置
JP6215555B2 (ja) * 2013-03-29 2017-10-18 株式会社ニデック 眼底画像処理装置および眼底画像処理プログラム
US9905008B2 (en) * 2013-10-10 2018-02-27 University Of Rochester Automated fundus image field detection and quality assessment
JP6469387B2 (ja) * 2014-08-26 2019-02-13 株式会社トプコン 眼底解析装置
US11461599B2 (en) * 2018-05-07 2022-10-04 Kennesaw State University Research And Service Foundation, Inc. Classification of images based on convolution neural networks
CN109166124B (zh) * 2018-11-20 2021-12-14 中南大学 一种基于连通区域的视网膜血管形态量化方法
CN110956107B (zh) * 2019-11-21 2023-05-30 佛山科学技术学院 一种基于oct扫描系统的三维血管类型区分方法
CN111000563B (zh) * 2019-11-22 2020-12-11 北京理工大学 一种视网膜动脉静脉直径比自动测量方法及装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040230124A1 (en) * 2003-05-12 2004-11-18 Querfurth Henry W. Methods of and systems and devices for assessing intracranial pressure non-invasively
US20080140192A1 (en) * 2006-11-29 2008-06-12 University Of Southern California Reversible thermoresponsive adhesives for implants
CN101999885A (zh) * 2010-12-21 2011-04-06 中国人民解放军国防科学技术大学 一种自动分离动静脉血管的内源光学成像方法
CN108073918A (zh) * 2018-01-26 2018-05-25 浙江大学 眼底视网膜的血管动静脉交叉压迫特征提取方法
CN108765418A (zh) * 2018-06-14 2018-11-06 四川和生视界医药技术开发有限公司 视网膜动静脉等效管径比检测方法及检测装置
CN110874597A (zh) * 2018-08-31 2020-03-10 福州依影健康科技有限公司 一种眼底图像血管特征提取方法、设备、系统和存储介质
CN111681242A (zh) * 2020-08-14 2020-09-18 北京至真互联网技术有限公司 视网膜血管动静脉区分方法和装置、设备

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114973207A (zh) * 2022-08-01 2022-08-30 成都航空职业技术学院 一种基于目标检测的路标识别方法
CN114973207B (zh) * 2022-08-01 2022-10-21 成都航空职业技术学院 一种基于目标检测的路标识别方法

Also Published As

Publication number Publication date
US20240177299A1 (en) 2024-05-30
US12307663B2 (en) 2025-05-20
CN111681242A (zh) 2020-09-18
KR20220076507A (ko) 2022-06-08
JP2022550669A (ja) 2022-12-05
CN111681242B (zh) 2020-11-17
JP7333465B2 (ja) 2023-08-24
KR102826940B1 (ko) 2025-06-27

Similar Documents

Publication Publication Date Title
WO2022033580A1 (zh) 视网膜血管动静脉区分方法和装置、设备
CN112465772B (zh) 眼底彩照图像血管评估方法、装置、计算机设备和介质
CN111932535B (zh) 用于处理图像的方法、装置、设备以及存储介质
SujithKumar et al. Automatic detection of diabetic retinopathy in non-dilated RGB retinal fundus images
CN110276356A (zh) 基于r-cnn的眼底图像微动脉瘤识别方法
WO2022063198A1 (zh) 一种肺部图像处理方法、装置及设备
Bian et al. Optic disc and optic cup segmentation based on anatomy guided cascade network
Zaki et al. Diabetic retinopathy assessment: Towards an automated system
US20240074658A1 (en) Method and system for measuring lesion features of hypertensive retinopathy
CN104537669B (zh) 眼底图像的动静脉视网膜血管分割方法
CN104545792B (zh) 眼底图像的动静脉视网膜血管视盘定位方法
CN108764342B (zh) 一种对于眼底图中视盘和视杯的语义分割方法
Qin et al. Optic disc and cup segmentation based on deep learning
CN111797900B (zh) 一种oct-a图像的动静脉分类方法及装置
Heslinga et al. Quantifying graft detachment after Descemet's membrane endothelial keratoplasty with deep convolutional neural networks
CN105718919B (zh) 一种眼底图像视盘定位方法及系统
CN117876801B (zh) 基于眼底血管特征和人工智能预测糖尿病肾病的方法
CN116309235A (zh) 一种针对糖尿病预测的眼底图像的处理方法和系统
CN114549395B (zh) 基于视计算的图像特征确定的方法、装置、介质和设备
Jaafar et al. Automated detection of red lesions from digital colour fundus photographs
CN107203758A (zh) 糖尿病人视网膜血管图像分割方法
CN113344893B (zh) 一种高精度眼底动静脉识别的方法、装置、介质和设备
CN114387209A (zh) 眼底结构特征确定的方法、装置、介质和设备
Liu et al. Retinal vessel segmentation using densely connected convolution neural network with colorful fundus images
CN117523633A (zh) 一种基于图像深度学习的眼部上睑下垂自动识别方法

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2022512438

Country of ref document: JP

Kind code of ref document: A

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

Ref document number: 21855630

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 17755560

Country of ref document: US

ENP Entry into the national phase

Ref document number: 20227015077

Country of ref document: KR

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: 21855630

Country of ref document: EP

Kind code of ref document: A1

WWG Wipo information: grant in national office

Ref document number: 17755560

Country of ref document: US

WWG Wipo information: grant in national office

Ref document number: 1020227015077

Country of ref document: KR