OCTA-based image processing method

Abstract

The application discloses an OCTA-based image processing method, comprising the following steps: acquiring an OCTA angiogram; pre-processing the OCTA angiogram; performing image segmentation on the pre-processed OCTA angiogram; extracting a blood vessel skeleton from the image segmented OCTA angiogram; and performing parameter measurement based on the extraction result. The OCTA-based image processing method can accurately obtain the skeleton and boundary of the blood vessel by performing a thinning algorithm and branch point detection on the binary image, and can split the blood vessel tree into single-section branch blood vessels, so that the measurement process of each branch blood vessel is simplified and unified, and the efficiency and accuracy of the measurement are ensured.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image processing, and particularly relates to an image processing method based on OCTA. BACKGROUND

[0002] Morphological parameters of blood vessels, such as blood vessel diameter, blood vessel area density and the like, have important reference values in clinical diagnosis of many diseases. Many diseases are proved to be closely related to morphological changes of blood vessels. By measuring the morphological parameters of blood vessels, the condition of blood vessels of a patient can be understood in time, and thus a judgment basis for analyzing and diagnosing related diseases is provided. Therefore, research on a blood vessel morphological parameter measurement method has important significance in clinical and scientific research fields.

[0003] Optical coherence tomography angiography (OCTA) is a functional imaging technology derived from OCT technology. Based on the non-invasive and high-resolution characteristics of the OCT technology, the OCTA technology provides three-dimensional blood vessel information inside a tissue by using red blood cells flowing in the blood vessels as a natural contrast agent, so as to realize visualization of three-dimensional blood vessel networks in a biological tissue without invasion.

[0004] The basic principle of OCTA is to use moving particles in a biological tissue, such as red blood cells, as an internal contrast agent for angiography. This is because the light beam signals backscattered from the biological tissue around the blood vessels are relatively stable, while the light reflectivity at the same position is always changing due to the constant rolling and movement of red blood cells, and thus the light beam signal intensity backscattered from the blood flow inside the blood vessels also constantly changes. OCTA distinguishes static tissues and moving particles by this principle, and thus can generate corresponding blood flow signals, and can realize imaging of blood vessels without an external contrast agent.

[0005] It is often difficult to distinguish subtle changes in microvessels by visually inspecting the OCTA angiography image. At this time, it is necessary to process the OCTA angiography image and quantitatively detect the blood vessel parameters. SUMMARY

[0006] The application provides an image processing method based on OCTA to solve the above technical problems, and specifically adopts the following technical scheme:

[0007] An image processing method based on OCTA, comprising the following steps:

[0008] Obtaining an OCTA angiography image;

[0009] Pretreating the OCTA angiography image;

[0010] performing image segmentation on the preprocessed OCTA angiogram;

[0011] performing vessel skeleton extraction on the OCTA angiogram after image segmentation;

[0012] performing parameter measurement based on the extraction result.

[0013] Further, the specific method for obtaining the OCTA angiogram is as follows:

[0014] adding the square amplitudes of the data of the two detection channels of the OCT system to obtain a linear intensity total amount;

[0015] by selecting a suitable pixel window, calculating each pair of adjacent B scans in the three-dimensional OCT scan to generate a three-dimensional correlation graph;

[0016] on the basis of the three-dimensional correlation graph, calculating a two-dimensional maximum intensity projection to obtain a two-dimensional OCTA angiogram.

[0017] Further, the specific method for preprocessing the OCTA angiogram is as follows:

[0018] performing gray scale conversion on the OCTA angiogram;

[0019] performing image enhancement on the converted OCTA angiogram.

[0020] Further, the specific method for performing image enhancement on the converted OCTA angiogram is as follows:

[0021] performing filtering on the converted OCTA angiogram by using median filtering;

[0022] performing contrast enhancement on the filtered OCTA angiogram by using a contrast-limited adaptive histogram equalization method;

[0023] performing linear gray scale conversion on the contrast-enhanced OCTA angiogram.

[0024] Further, the specific method for performing image segmentation on the preprocessed OCTA angiogram is as follows:

[0025] calculating the segmentation threshold of the preprocessed OCTA angiogram by using the Otsu algorithm;

[0026] performing binaryzation on the OCTA angiogram according to the segmentation threshold;

[0027] performing morphological processing on the binaryzed OCTA angiogram;

[0028] Reconstruct a binary image for measurement by judging the size of connected domain of the OCTA angiogram after morphological processing.

[0029] Further, the specific method for constructing a binary image for measurement by judging the size of connected domain of the OCTA angiogram after morphological processing is:

[0030] Count the number of connected domains contained in the statistical image, the total number of pixels contained in each connected domain, and the linear index position of the pixels;

[0031] Set a judgment threshold, and judge each connected domain. If the total number of pixels in the connected domain is greater than the judgment threshold, record the linear index of the pixels in the connected domain. If the total number of pixels in the connected domain is less than the judgment threshold, the connected domain is classified into the background region and is not processed.

[0032] In a blank image with the same size as the original image, set the pixel value corresponding to the recorded pixel index position to 1, thereby obtaining a binary image for measurement.

[0033] Further, the specific method for extracting the blood vessel skeleton from the OCTA angiogram after image segmentation is:

[0034] Extract the blood vessel skeleton from the constructed binary image for measurement by thinning algorithm;

[0035] Perform single-pixel processing on the extracted blood vessel skeleton;

[0036] Delete stray pixels based on template matching to prune false branches;

[0037] Perform eight-neighborhood filtering to detect skeleton branch points;

[0038] Delete the detected branch point pixels, thereby splitting the originally connected blood vessel skeleton tree into single-segment branches;

[0039] Extract the blood vessel boundary.

[0040] Further, the thinning algorithm is Zhang-Suen algorithm.

[0041] Further, the specific method for parameter measurement based on the extraction result is:

[0042] After skeletonization, select a pixel point on the skeleton in order, and determine a local region as a fitting neighborhood centered on the pixel point;

[0043] Fit a local curve according to the coordinates of the skeleton pixels in the fitting neighborhood;

[0044] The first derivative of the curve at the pixel point is calculated to obtain the tangent direction of the curve at the pixel point, so as to determine the blood vessel orientation;

[0045] A straight line perpendicular to the blood vessel orientation intersects the blood vessel wall at two points, and the distance between the two points is calculated to obtain the blood vessel diameter;

[0046] The density of the blood vessel region is counted.

[0047] Further, the specific method for fitting a local curve according to the coordinates of the skeleton pixels in the fitting neighborhood is as follows:

[0048] The data is fitted into a quadratic polynomial by using the least square method.

[0049] The image processing method based on OCTA provided by the present application can accurately obtain the skeleton and boundary of the blood vessel by performing thinning algorithm and branch point detection on the binary image, and can split the blood vessel tree into single-section branch blood vessels, so that the measurement process of each branch blood vessel is simplified and unified, thereby ensuring the efficiency and accuracy of the measurement. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 is a schematic diagram of an image processing method based on OCTA of the present application;

[0051] Figure 2 is a schematic diagram of the original OCTA angiogram after gray scale conversion;

[0052] Figure 3 is a schematic diagram after median filtering;

[0053] Figure 4 is a schematic diagram after adaptive histogram equalization processing with limited contrast;

[0054] Figure 5 is a schematic diagram after linear gray scale transformation processing;

[0055] Figure 6 is a schematic diagram after threshold segmentation by Otsu algorithm;

[0056] Figure 7 is a schematic diagram after closing operation processing;

[0057] Figure 8 is a schematic diagram after screening;

[0058] Figure 9 is a schematic diagram of skeleton extraction effect;

[0059] Figure 10 is a schematic diagram of blood vessel skeleton diameter measurement. DETAILED DESCRIPTION

[0060] The application will be specifically described below in combination with the drawings and specific embodiments.

[0061] As Figure 1 shown is an OCTA-based image processing method of the application, comprising the following steps: S1: obtaining an OCTA angiogram. S2: preprocessing the OCTA angiogram. S3: performing image segmentation on the preprocessed OCTA angiogram. S4: performing vessel skeleton extraction on the image segmented OCTA angiogram. S5: performing parameter measurement based on the extraction result. The above steps will be specifically described below.

[0062] For step S1: obtaining an OCTA angiogram.

[0063] In the application, a speckle decorrelation algorithm is used to process to obtain the OCTA angiogram.

[0064] The specific method for obtaining the OCTA angiogram is as follows:

[0065] The square amplitudes of the data of the two detection channels of the OCT system are added to obtain the linear intensity total amount.

[0066] By selecting a suitable pixel window, the three-dimensional correlation map is generated by calculating each pair of adjacent B scans in the three-dimensional OCT scan. That is, the correlation map between each pair of adjacent B scans is calculated, wherein the correlation coefficient is calculated by the following formula,

[0067]

[0068] wherein the window size is defined by an MxN grid, I A and I B are adjacent sub-image pairs of adjacent B scans, which are determined by the window in an OCT linear intensity total amount. and are the average values of sub-images I A and I B respectively. The range of the correlation value is [-1, 1]. By selecting a suitable pixel window, the three-dimensional correlation map is generated by calculating each pair of adjacent B scans in the three-dimensional OCT scan.

[0069] On the basis of the three-dimensional correlation map, the two-dimensional OCTA angiogram is obtained by calculating the two-dimensional maximum intensity projection (MIP).

[0070] In the application, the part of the OCTA angiogram obtained is the angiogram of the burned skin part.

[0071] For step S2: pre-processing the OCTA angiogram.

[0072] Specifically, the specific method for pre-processing the OCTA angiogram is:

[0073] The OCTA angiogram is subjected to grayscale conversion.

[0074] Specifically, a color image is composed of a red, green, and blue (R, G, B) three-value vector for each pixel in the computer. In order to facilitate image processing, it is necessary to convert the color image composed of three vectors into a grayscale image of a single vector. In the present application, the following formula is used,

[0075] f(i,j) = 0.299R(i,j) + 0.587G(i,j) + 0.114B(i,j)

[0076] The OCTA angiogram is processed from three-dimensional mapping to one-dimensional, so that the image is converted into a grayscale image, as shown in Figure 2 .

[0077] Subsequently, the converted OCTA angiogram is subjected to image enhancement.

[0078] In the present application, the specific method for enhancing the converted OCTA angiogram is:

[0079] The converted OCTA angiogram is first subjected to filtering using median filtering. The result is shown in Figure 3 . The OCTA angiogram mainly contains salt and pepper noise and noise B scans generated by motion artifacts. The former will affect the subsequent binary threshold setting, and the latter appears as a high-intensity straight line noise on the image. For the former, image filtering is used to suppress noise. Median filtering is a nonlinear signal processing method that can effectively filter out salt and pepper noise and preserve the edge features of the image.

[0080] Then, the filtered OCTA angiogram is subjected to contrast enhancement by a contrast-limited adaptive histogram equalization method. The result is shown in Figure 4 . It can be understood that the grayscale image obtained by median filtering still contains a part of noise. In order to obtain an accurate blood vessel contour, the image needs to be subjected to grayscale equalization processing to improve the dynamic range of the grayscale, enhance the edge features of the image and further reduce the background noise, thereby improving the contrast between the blood vessel structure and the background. The contrast-limited adaptive histogram equalization enhances the blood vessel structure and avoids identifying low-brightness blood vessels as background noise.

[0081] Finally, the contrast-enhanced OCTA angiography image undergoes linear grayscale transformation. In this application, pixels in the [0.01, 0.99] interval are linearly mapped to [0, 1], increasing the image's grayscale dynamic range and further distinguishing blood vessels from the background, facilitating subsequent image binarization. Figure 5 As shown.

[0082] For step S3: perform image segmentation on the preprocessed OCTA angiography image.

[0083] Specifically, the method for image segmentation of the preprocessed OCTA angiography images is as follows:

[0084] The segmentation threshold of the preprocessed OCTA angiography image was calculated using the Otsu algorithm.

[0085] The OCTA angiography images were binarized based on a segmentation threshold. The results are as follows: Figure 6 As shown.

[0086] Morphological processing was performed on the binarized OCTA angiography images. In this application, morphological closing operations were used to process the images. The morphological closing operation first involves dilation, then erosion. The selection of template parameters was crucial to fill small holes in the image while minimizing the impact on the contours of structural holes. In this application, 20 pixels was chosen as the template radius for the closing operation. The results are as follows... Figure 7 As shown.

[0087] A binary image for measurement is reconstructed by determining the size of connected components in the morphologically processed OCTA angiography image. For example... Figure 7 As shown, due to the non-uniformity of vascular signals, many isolated vascular segments exist in the image after threshold segmentation. Some of these vessels are too small, and these small vessels will significantly deviate from the actual diameter in subsequent diameter measurements, affecting the accuracy of the measurement. However, these vessels are objectively existing entities and are meaningful in the calculation of vascular region density, thus needing to be included in the calculation. Therefore, the segmentation result cannot be directly modified; a separate binary image needs to be constructed for diameter measurement. The segmented image will be directly used for calculating the vascular region density.

[0088] In this application, a binary image for diameter measurement is constructed by determining the size of connected components in the image. Specifically, the method for constructing the binary image for measurement by determining the size of connected components in the morphologically processed OCTA angiography image is as follows:

[0089] The number of connected components in the image, the total number of pixels in each connected component, and the linear index position of the pixels are counted.

[0090] A determination threshold is set, and each connected domain is determined. If the total number of pixels in the connected domain is greater than the determination threshold, the linear index of the pixels in the connected domain is recorded. If the total number of pixels in the connected domain is less than the determination threshold, the connected domain is classified into a background region and is not processed.

[0091] In a blank image with the same size as the original image, the pixel value corresponding to the recorded pixel index position is set to 1, thereby obtaining a binary image for measurement.

[0092] Preferably, the determination threshold is 10,000. The result after processing is shown in FIG. 4. Figure 8

[0093] For step S4: Vessel skeleton extraction is performed on the OCTA angiogram after image segmentation.

[0094] The morphological skeleton is a topological description of the geometric shape features of an image. The vessel skeleton can reflect the connection state, structural information and direction information of the blood vessels. The vessel skeleton is often located in the center of the blood vessels and is also called the vessel centerline.

[0095] In the present application, the specific method for performing vessel skeleton extraction on the OCTA angiogram after image segmentation is as follows:

[0096] The vessel skeleton is extracted from the constructed binary image for measurement through a thinning algorithm. Specifically, in the present application, a Zhang-Suen fast parallel algorithm is used to perform thinning processing on the image. The Zhang-Suen algorithm needs to traverse all pixels that are not 0 in the image each time it is run. Whether to delete the target pixel is determined by judging whether the eight-neighbor domain of the target pixel meets certain rules. As shown in FIG. 5, the left image is the original image, and the right image is the result of the thinning processing. Figure 9

[0097] The extracted vessel skeleton is subjected to single-pixel processing. The Zhang-Suen algorithm cannot guarantee that the extracted skeleton is single-pixel. Non-single-pixel skeletons are not conducive to the simplification of diameter measurement, and therefore further single-pixel processing needs to be performed on the skeleton extracted by the Zhang-Suen algorithm.

[0098] Based on template matching, stray pixels are deleted to trim the pseudo-branches. During the thinning process, many pseudo-branches, also known as burrs, are generated. These pseudo-branches do not have actual corresponding physical structures in the blood vessels, and therefore need to be removed to avoid misleading the diameter measurement. The principle of removing the burrs is also based on template matching to delete stray pixels. After comparing the effects of different trimming times, the skeleton of the image is trimmed 200 times in the present application.

[0099] ​​Skeleton branch points are detected by octagonal filtering. A pixel at a branch point has a common feature that there are three adjacent points in the octagonal neighborhood. According to this feature, the octagonal filtering can be used to detect the skeleton branch points. The pixels with a value of 3 in the octagonal filtered image are the suspected points, which include the background points with a value of 0 before filtering and the skeleton with a value of 1. The coordinates of the suspected points and the coordinates of the skeleton are compared to obtain the coordinates of the branch points.

[0100] The detected branch point pixels are deleted to split the originally connected vessel skeleton tree into single segment branches. This lays a foundation for subsequent diameter detection.

[0101] Finally, the blood vessel boundary is extracted. In order to more simply determine the intersection of the detection line and the blood vessel wall when measuring the diameter, the boundary position of the blood vessel needs to be further extracted. The edge extraction examines the gray difference of the image. For a binary image, only the relationship between the center point and its neighborhood pixels needs to be examined, that is, the boundary extraction.

[0102] Specifically, the pixels located at the boundary have two conditions: (1) the pixel value is 1, that is, located on the blood vessel target; and (2) the octagonal neighborhood of the pixel must contain a pixel with a value of 0, that is, the background. According to this feature, the calculation steps for extracting the boundary are as follows:

[0103] 1. In the binary image, the pixel points with a value of 1 are detected in order. If the point has not been recorded, it is taken as a center point, and step 2 is entered;

[0104] 2. The upper left corner of the center point is taken as a starting point, and the pixel values in the octagonal neighborhood are detected in a clockwise direction;

[0105] 3. If a neighborhood point with a value of 0 is detected in step 2, the coordinates of the center point in this detection are recorded;

[0106] 4. Steps 1 to 3 are repeated until all pixels have been detected.

[0107] For step S5: parameter measurement based on the extraction result.

[0108] The measurement of the blood vessel diameter needs to be along the direction perpendicular to the blood vessel direction to best reflect the true situation of the blood vessel. Specifically, the specific method for parameter measurement based on the extraction result is as follows:

[0109] After skeletonization, a pixel point on the skeleton is selected in order, and a local region is determined as a fitting neighborhood with the pixel point as the center. As shown in Figure 10 .

[0110] A local curve is fitted according to the coordinates of the skeleton pixels in the fitting neighborhood. In the present application, the least square method is used to fit the data into a quadratic polynomial.

[0111] The selection of the size of the fitting neighborhood region will affect the measurement result. If the region is too small, the coordinate data used for fitting is less, and the fitted curve will sometimes deviate from the actual curve, and cannot truly reflect the direction of the blood vessel, resulting in that the direction of the detected diameter is not perpendicular to the direction of the blood vessel, and the measured diameter data is larger. If the region is too large, the shape of the skeleton cannot be well fitted as a quadratic polynomial, and the calculation amount will also increase. The fitting neighborhood size selected by the present application is 19x19, the maximum preset diameter is set to 321 pixel lengths, and the measurement interval along the blood vessel skeleton is more than 5 pixels each time.

[0112] The first derivative of the curve at the pixel point is calculated to obtain the tangent direction of the curve at the pixel point, so as to determine the direction of the blood vessel.

[0113] The line perpendicular to the direction of the blood vessel intersects the blood vessel wall at two points, and the distance between the two points is calculated to obtain the diameter of the blood vessel. In actual application, it is necessary to convert the pixel distance into physical length according to the proportion.

[0114] Finally, the density of the blood vessel region is calculated. Specifically, the area occupied by the blood vessel can be represented by the number of pixels with a value of 1 in the image, and the density of the blood vessel region can be obtained by dividing the area of the blood vessel by the size of the image.

[0115] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the above examples do not limit the present application in any form, and any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present application.

Claims

1. An image processing method based on OCTA, characterized in that, Includes the following steps: Obtain OCTA angiography images; The OCTA angiography images are preprocessed, including grayscale conversion, median filtering, contrast-limited adaptive histogram equalization, and linear grayscale transformation. Image segmentation is performed on the preprocessed OCTA angiography images; The vascular skeleton is extracted from the OCTA angiography image after image segmentation; Parameters are measured based on the extraction results; The specific method for extracting the vascular skeleton from the OCTA angiography image after image segmentation is as follows: The vascular skeleton was extracted from the constructed binary image used for measurement using the Zhang-Suen thinning algorithm; The extracted vascular skeleton is processed into single pixels; Pruning of pseudo-branches is achieved by removing stray pixels based on template matching; Eight-neighbor filtering is used to detect skeleton branch points; The detected branch point pixels are deleted, thereby splitting the originally connected vascular skeleton tree into single branches; Extracting blood vessel boundaries; The specific method for image segmentation of the preprocessed OCTA angiography image is as follows: The segmentation threshold of the preprocessed OCTA angiography image was calculated using the Otsu algorithm; The OCTA angiography image is binarized according to the segmentation threshold. Morphological processing was performed on the binarized OCTA angiography images; A binary image for measurement is reconstructed by determining the size of the connected components in the morphologically processed OCTA angiography image. The specific method for parameter measurement based on the extraction results is as follows: After skeletonization, select the pixel on the skeleton in sequence, and determine a local region centered on that pixel as the fitting neighborhood. A local curve is fitted based on the coordinates of the skeleton pixels in the fitted neighborhood, including: fitting the data into a quadratic polynomial using the least squares method. Calculate the first derivative of the curve at that pixel to obtain the tangent direction of the curve at that pixel, thereby determining the orientation of the blood vessel. A straight line perpendicular to the direction of the blood vessel intersects the blood vessel wall at two points. The distance between the two points is used to calculate the diameter of the blood vessel. Statistical analysis of vascular density.

2. The image processing method based on OCTA according to claim 1, characterized in that, The specific method for obtaining OCTA angiography images is as follows: The total linear intensity is obtained by summing the squared amplitude values ​​of the data from the two detection channels of the OCT system. By selecting an appropriate pixel window, calculations are performed on each pair of adjacent B-scans in the 3D OCT scan to generate a 3D correlation map; Based on the three-dimensional correlation map, the two-dimensional OCTA angiography map is obtained by calculating the two-dimensional maximum intensity projection.

3. The image processing method based on OCTA according to claim 1, characterized in that, The specific method for preprocessing the OCTA angiography images is as follows: The OCTA angiography images were converted to grayscale. Image enhancement was performed on the converted OCTA angiography images.

4. The image processing method based on OCTA according to claim 3, characterized in that, The specific method for image enhancement of the converted OCTA angiography image is as follows: Median filtering was applied to the converted OCTA angiography images. The contrast of the filtered OCTA angiography image is enhanced by a contrast-limited adaptive histogram equalization method. The contrast-enhanced OCTA angiography image was then subjected to linear grayscale transformation.

5. The image processing method based on OCTA according to claim 4, characterized in that, The specific method for constructing a binary image for measurement by determining the size of the connected components in the morphologically processed OCTA angiography image is as follows: The number of connected components in the image, the total number of pixels in each connected component, and the linear index position of the pixels are counted. Set a judgment threshold and judge each connected component. If the total number of pixels in the connected component is greater than the judgment threshold, record the linear index of the pixels in the connected component. If the total number of pixels in the connected component is less than the judgment threshold, classify the connected component into the background region and do not perform any processing. In a blank image of the same size as the original image, the pixel value corresponding to the recorded pixel index position is set to 1, thus obtaining a binary image for measurement.

Images