An OCTA image blood vessel segmentation method, quantitative analysis method and program product

Abstract

The present application is an image segmentation method. In view of the technical problem that image data collected by different devices are difficult to segment and analyze on a unified model, an OCTA image vascular segmentation method, a quantitative analysis method and a program product are provided. When training the current segmentation model used for OCTA image vascular segmentation, the training sample data is style enhanced and structure enhanced to make the sample data richer. After the segmentation model is trained, the segmentation model can recognize and segment OCTA images from multiple centers and multiple devices. The enhancement method in the present application not only expands the training data range of the segmentation model, but also improves the robustness of the segmentation model to changes in image details, scale and morphology in complex scenes, so that the segmentation model can maintain stable segmentation performance in various application scenarios.

Description

Technical Field

[0001] The present application relates to an image segmentation method, and to an OCTA image blood vessel segmentation method, a quantitative analysis method and a program product. Background Art

[0002] OCTA (Optical Coherence Tomography Angiography) has been widely used in the screening and monitoring of various fundus diseases in ophthalmology clinics because of its advantages of being non-invasive, real-time, high-resolution, and able to image retinal microvascular structures. A large number of patients around the world have undergone retinal microvascular detection and analysis through OCTA. Unlike traditional OCT (Optical Coherence Tomography), OCTA can not only provide the anatomical structure of the retina, but also display the flow information of microvessels, and has extremely high clinical application value. The accurate segmentation and quantitative analysis of vascular structures in OCTA images are of great significance for evaluating early lesions of the disease, predicting progression, and evaluating the efficacy of intervention.

[0003] As OCTA technology is increasingly used in ophthalmology, the demand for automated and intelligent OCTA vascular analysis and diagnosis is increasing. However, due to differences in hardware and parameter settings among different OCTA devices, OCTA images acquired on different devices for the same patient often have significant differences in vascular structure and image quality, which directly affects the stability and generalization ability of automated vascular segmentation. Therefore, it is difficult to segment and analyze image data acquired by different devices on a unified model, which has become one of the main bottlenecks for the further promotion and application of OCTA vascular analysis technology. Summary of the invention

[0004] The present application aims to solve the technical problem that image data collected by different devices are difficult to segment and analyze on a unified model, and provides an OCTA image blood vessel segmentation method, a quantitative analysis method and a program product.

[0005] In order to achieve the above objectives, this application adopts the following technical solutions:

[0006] In a first aspect, the present application proposes a method for OCTA image blood vessel segmentation, comprising:

[0007] Acquire the acquired OCTA images;

[0008] The acquired OCTA image is input into the trained segmentation model to obtain the OCTA image segmentation result; the method for obtaining the trained segmentation model includes:

[0009] Style enhancement is performed on multiple original OCTA images through NURBS curves to obtain a style-enhanced image group; the control point generation of the NURBS curve is constrained by the pixel value of the OCTA image, the node vector of the NURBS curve includes four groups of nodes, the first group of nodes in the four groups of nodes is equal to the actual minimum normalized result value of the pixels of the OCTA image, the fourth group of nodes is equal to the actual maximum normalized result value of the pixels of the OCTA image, the second group of nodes and the third group of nodes are randomly generated between the first group of nodes and the fourth group of nodes, and the second group of nodes is greater than or equal to the third group of nodes;

[0010] The multiple original OCTA images are respectively divided into different retinal vascular layers, and each retinal vascular layer is then divided into different vascular branch regions; each vascular branch region is subjected to local elastic deformation to obtain a local elastic deformation field, and then the local elastic deformation fields of different vascular branch regions in the same retinal vascular layer are superimposed to obtain the final result of the local elastic deformation, and a plurality of OCTA images after local elastic deformation are correspondingly obtained as a structure-enhanced image group;

[0011] The style enhanced image group, the structure enhanced image group and a plurality of original OCTA images are input into a segmentation model, and the segmentation model is trained to obtain a trained segmentation model.

[0012] Furthermore, the node vectors of the NURBS curve include:

[0013]

[0014] in, represents the node vector, represents a node in the second set of nodes, Represents a node in the third set of nodes.

[0015] Furthermore, the style enhancement of the plurality of original OCTA images by using NURBS curves includes:

[0016] Multiply the control points in the NURBS curve by the pixels of the original OCTA image.

[0017] Furthermore, the locally elastically deforming each blood vessel branch region includes:

[0018] Construct vascular functions of different vascular branching regions through segmentation labels;

[0019] The vascular functions of different vascular branching regions, the sine functions of different vascular branching regions, and the random variables are summed to obtain fitting functions of different vascular branching regions;

[0020] The fitting function is applied to the corresponding blood vessel branch regions respectively to obtain the local elastic deformation field of the corresponding blood vessel branch regions.

[0021] In a second aspect, the present application proposes a method for quantitative analysis of OCTA images, comprising:

[0022] Using the above OCTA image blood vessel segmentation method, an OCTA image segmentation result is obtained;

[0023] Based on the OCTA image segmentation result, at least one of a blood vessel curvature parameter, a blood vessel diameter parameter and a blood vessel density parameter is calculated.

[0024] Further, the method for calculating the blood vessel curvature parameter includes:

[0025] Morphological operations are used to extract the vascular skeleton from the OCTA image segmentation results, and the central axis of each segment of the vascular skeleton corresponding to the blood vessel is obtained;

[0026] For each segment of the vascular skeleton, a high-order curve fitting technique is used to fit the central axis of the blood vessel corresponding to each segment of the vascular skeleton to obtain a fitting curve;

[0027] Calculating the geometric derivative of the fitting curve;

[0028] According to the relationship between the second-order derivative in the geometric derivative and the tangential change rate of the fitting curve, the local curvature of each point on the fitting curve is calculated to obtain the curvature value of each point on the fitting curve;

[0029] By statistically analyzing the curvature values, the vascular curvature parameters corresponding to each segment of the vascular skeleton are extracted.

[0030] After obtaining the curvature value of each point on the fitting curve, the method further includes:

[0031] Set the upper and lower limits of the curvature value. If the curvature value exceeds the upper and lower limits, it will be deleted.

[0032] Furthermore, the method for calculating the blood vessel diameter parameter includes:

[0033] Perform edge detection on the OCTA image segmentation results to extract the outer edge of the blood vessels and obtain the blood vessel skeleton;

[0034] By sampling the points on the vascular skeleton in the vertical direction and measuring the distance between the corresponding points on the outer edge of the blood vessel, the local blood vessel diameter is obtained;

[0035] Statistical analysis is performed on the local blood vessel diameter to obtain the blood vessel diameter parameters.

[0036] Furthermore, the method for calculating the blood vessel density parameter includes:

[0037] The OCTA image segmentation result is divided into a plurality of regions of equal size to obtain a plurality of divided regions;

[0038] For each divided area, the ratio of blood vessel pixels in the divided area to the total pixels in the divided area is calculated as the blood vessel density of the divided area;

[0039] The vascular density of each divided area is mapped to the OCTA image segmentation result in a color-coded manner to obtain a heat map of the density distribution as the calculation result of the vascular density parameter.

[0040] In a third aspect, the present application proposes a computer program product, which includes instructions, and when the instructions are executed by a processor, they implement the above-mentioned OCTA image blood vessel segmentation method, or when the instructions are executed by a processor, they implement the above-mentioned OCTA image quantitative analysis method.

[0041] Compared with the prior art, this application has the following beneficial effects:

[0042] The present application proposes a method for OCTA image vascular segmentation. When training the current segmentation model used for OCTA image vascular segmentation, the sample data used for training is style enhanced and structure enhanced to make the sample data richer. After the segmentation model is trained, the segmentation model can recognize and segment OCTA images from multiple centers and multiple devices. The enhancement method in the present application not only expands the training data range of the segmentation model, but also improves the robustness of the segmentation model to changes in image details, scale, and morphology in complex scenes, so that the segmentation model can maintain stable segmentation performance in various application scenarios. In particular, the segmentation method of the present application also shows strong adaptability to the refined segmentation of microvessels and lesion areas.

[0043] This application also proposes a quantitative analysis method for OCTA images, which has all the advantages of the above-mentioned OCTA image vascular segmentation method. In addition, this application can quickly and accurately perform quantitative analysis of vascular curvature parameters, vascular diameter parameters and vascular density parameters on the OCTA image segmentation results through an improved image segmentation method, which greatly improves the actual clinical application value of the OCTA image segmentation results. The quantitative analysis results can also provide doctors with accurate pathological information and assist in the diagnosis and treatment evaluation of the disease. Furthermore, this application uses an optimized quantitative analysis method to ensure high-precision quantitative results while being able to respond quickly to demand, providing more efficient technical support for real-time analysis and clinical applications, shortening the diagnosis time, and helping to make timely treatment decisions.

[0044] The present application also proposes a computer program product, which has all the advantages of the above-mentioned OCTA image blood vessel segmentation method and OCTA image quantitative analysis method. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0046] Figure 1 A schematic diagram of a flow chart of an OCTA image blood vessel segmentation method of the present application;

[0047] Figure 2 A schematic diagram of a flow chart of an OCTA image quantitative analysis method of the present application. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0049] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0050] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0051] In the description of the embodiments of the present application, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. appear, the orientation or position relationship indicated is based on the orientation or position relationship shown in the drawings, or the orientation or position relationship in which the invented product is usually placed when used. It is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0052] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0053] In the description of the embodiments of the present application, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0054] OCTA is an advanced medical imaging technology that can be used to observe microvascular structures in living tissues non-invasively and with high resolution. OCTA combines the principles of OCT and angiography to display the blood flow of tissues such as the retina and choroid in real time and in three dimensions.

[0055] At present, the vascular segmentation and analysis of OCTA images mainly rely on manual annotation by experts, which is not only time-consuming and laborious, but also has differences in segmentation consistency among experts, which affects the accuracy and reliability of vascular quantitative analysis based on OCTA images. With the increasing application of OCTA technology in ophthalmology, the demand for automated and intelligent OCTA vascular analysis and diagnosis is increasing. However, due to the differences in hardware and parameter settings of different OCTA devices, OCTA images collected on different devices for the same patient often have significant differences in vascular structure and image quality, which will directly affect the stability and generalization ability of automated vascular segmentation. Image data collected by different devices are difficult to segment and analyze on a unified model, which has become one of the main bottlenecks for the further promotion and application of OCTA vascular analysis technology. Therefore, studying the generalized segmentation technology in the OCTA image domain based on deep learning to solve the problem of image consistency under multi-device conditions has become a core scientific issue and technical direction in this field.

[0056] In recent years, domestic and foreign research teams have conducted extensive exploration in the research and development of OCTA vascular segmentation algorithms and have made initial progress. However, most existing algorithms are mainly trained based on a single device or a single data set, facing obvious overfitting problems, and are difficult to promote and apply in complex clinical scenarios with multiple centers and multiple devices. For example, the team of the Ningbo Institute of Materials, Chinese Academy of Sciences, Cixi Institute of Medical Engineering, proposed a coarse-to-fine two-stage vascular segmentation network OCTA-Net, which is used for vascular segmentation and extraction in OCTA, aiming to detect blood vessels of different sizes separately. Another example is the efficient vascular network proposed by the Institute of Automation, Chinese Academy of Sciences. Theodoros Pissas et al. used convolutional neural networks to solve the task of vascular segmentation of OCTA superficial vascular layer projection images.

[0057] Solving the device dependence of OCTA image segmentation and developing a vascular segmentation algorithm with good domain generalization capabilities have become the main technical challenges in this field. At the same time, accurate quantitative analysis of the vascular structure after OCTA segmentation to achieve auxiliary diagnosis of related fundus diseases is also of great clinical significance. Therefore, designing an OCTA vascular segmentation model based on domain generalization, combined with accurate vascular quantification analysis methods, to achieve fast and accurate segmentation and quantitative analysis of multi-center and multi-device OCTA images is a key direction of current clinical needs and technological development. This application not only promotes the application of OCTA technology in complex clinical environments, but also provides technical support for the automated diagnosis of large-scale OCTA data. It is also in line with the current development trend of artificial intelligence technology in the field of medical image analysis.

[0058] The present application is further described in detail below with reference to the embodiments and drawings:

[0059] like Figure 1 As shown, it is a flowchart of a method for segmenting blood vessels in an OCTA image according to the present application, which may include:

[0060] S101, acquiring the collected OCTA image.

[0061] The collected OCTA images are obtained from the OCTA device and provide raw data for subsequent vascular segmentation.

[0062] S102, inputting the acquired OCTA image into the trained segmentation model to obtain the OCTA image segmentation result.

[0063] It should be noted that the segmentation model is a machine learning or deep learning model specifically designed for image segmentation tasks. In OCTA image vascular segmentation, the goal of the segmentation model is to accurately identify and segment the vascular structures in the image to determine whether they belong to blood vessels. In practical applications, models such as convolutional neural networks, U-Net, SegNet, etc. can be used to extract image features through multi-layer convolution and pooling operations, and restore image resolution through upsampling or deconvolution operations, and finally output vascular segmentation results.

[0064] The method for obtaining the trained segmentation model comprises:

[0065] (1) Style enhancement is performed on multiple original OCTA images through NURBS curves to obtain a style enhanced image group; the control point generation of the NURBS curve is constrained by the pixel value of the OCTA image, and the node vector of the NURBS curve includes four groups of nodes, the first group of nodes of the four groups of nodes is equal to the actual minimum normalized result value of the pixels of the OCTA image, the fourth group of nodes is equal to the actual maximum normalized result value of the pixels of the OCTA image, the second group of nodes and the third group of nodes are randomly generated between the first group of nodes and the fourth group of nodes, and the second group of nodes is greater than or equal to the third group of nodes.

[0066] NURBS (Non-Uniform Rational B-spline) curve is a mathematical model widely used in computer graphics, computer-aided design and other engineering fields. It combines the advantages of B-spline curve and rational Bézier curve, and is defined by control points, weight factors and node vectors. It can accurately represent various curves from simple straight lines to complex free-form shapes. Based on the characteristics of NURBS curve, different curves can be efficiently fitted.

[0067] As an embodiment of the present application, a fourth-order NURBS curve can be used as a style mapping function. The style mapping function is constrained by the minimum and maximum values ​​of the non-empty regions of the image. Specifically, a normalized minimum value of an image region can be given. and maximum value , the constrained style mapping function expression is:

[0068]

[0069] in, is the fractional value along the NURBS curve, is a constrained style mapping function, , , is the starting point for defining the range of values. is the end point that defines the value range. and From the interval Two randomly generated control points in .

[0070] However, directly applying the constrained style mapping function to the OCTA image may lose detail information due to the image enhancement process. If the style mapping function is not designed properly, artifacts or distortion may be introduced during the enhancement process, and the enhanced data may deviate from the distribution of the original OCTA image data. Therefore, this application adopts the above-mentioned NURBS curve design for OCTA images to achieve style enhancement while retaining image details.

[0071] In order to maintain structural details and minimize style changes during the enhancement process, the curve transformation with reduced curvature is the key. NURBS curves allow the shape of the curve to be precisely controlled by the operation of control points and node vectors. To ensure that the curve transformation retains structural details and minimizes style changes, this application introduces a method to reduce curvature. The distribution and function of the control points are affected by designing the following node vectors:

[0072]

[0073] in, represents the node vector, represents a node in the second set of nodes, Indicates a node in the third group of nodes. As an example, the actual maximum normalized result value of the OCTA image pixel is 1, and the actual minimum normalized result value of the OCTA image pixel is 0. and is a randomly generated value between 0 and 1 that satisfies The knot vector divides the parameter space into three intervals with different knot densities, ensuring smooth transitions and consistent curvature of the curve. The random assignment of knot vectors ensures smooth transitions between intervals, thereby minimizing the overall change of the curve. The generation of control points is constrained by the pixel values ​​of the image, ensuring that the shape of the NURBS curve is consistent with the image features with minimal changes. The knot vector directly affects the effect of the control points on the shape of the curve, by ensuring , which can make the control point and , has a gentler effect on NURBS curves, which helps preserve the structural details of the image while enhancing the style.

[0074] By using NURBS curves with reduced curvature and guiding them through specific node vectors, an effective method for style enhancement of OCTA images is provided. This method maintains the key structural details of the image while enhancing the image style, ensuring that the enhanced data remains faithful to the original pixel value distribution. By leveraging the mathematical properties of NURBS, a smooth and controllable enhancement process is achieved, which is suitable for the unique characteristics of OCTA images.

[0075] (2) Multiple original OCTA images are divided into different retinal vascular layers, and each retinal vascular layer is then divided into different vascular branch regions; local elastic deformation is performed on each vascular branch region to obtain a local elastic deformation, and then the local elastic deformations of different vascular branch regions in the same retinal vascular layer are superimposed to obtain the final result of the local elastic deformation, corresponding to which multiple OCTA images after local elastic deformation are obtained as a structure-enhanced image group.

[0076] In order to simulate the subtle differences between different retinal vascular layers in OCTA images, fitting functions are constructed for the vascular branching regions in different retinal vascular layers. Usually, the deformation for natural images is global, not specifically for the local details of the image. In contrast, in OCTA images, the vascular structure may mutate in specific local areas. Therefore, the first i The vascular function of the vascular branch area , for i The vascular function of the vascular branch area , No. i Sine function of the branching area of ​​a vessel and random variables By adding, we get the fitting functions of different vascular branching regions, expressed as;

[0077]

[0078] in, It is i The vascular branch area, i The vascular function of the vascular branch area To capture the changing trend of the corresponding area. In order to realize the elastic deformation of different retinal vascular layers in each local area, the vascular function is combined with the sine function and the random variable is introduced. , in order to introduce uncertainty, thus simulating the changes between different retinal vascular layers and in different images. Each retinal vascular layer is fitted with different vascular functions from top to bottom, and combined with a positive sine function with uncertainty to generate a local elastic deformation field.

[0079] These local elastic deformation fields associated with different retinal vascular layers are then applied to the corresponding vascular branch regions. The final effect of local elastic deformation is obtained by applying the local elastic deformation fields to each vascular branch region and superimposing the results. This method can accurately model the changes in the retinal vascular layer, reflecting the natural vascular changes and local structural changes in OCTA images. By generating controllable local elastic deformation fields, these perturbations can induce local changes in the blood vessels while maintaining the integrity of the actual vascular structure.

[0080] By fine-tuning the fitting functions of different vascular branching regions, the generated local elastic deformation field can simulate the natural changes and local structural changes of the retinal vascular layer in OCTA images. It can effectively deal with the domain shift caused by local structural changes and ensure that the changes in the retinal vascular layer in the segmentation model are closely consistent with the natural vascular structure information observed in OCTA images.

[0081] (3) The style enhanced image group, the structure enhanced image group and a plurality of original OCTA images are input into a segmentation model, and the segmentation model is trained to obtain a trained segmentation model.

[0082] Mixing the style-enhanced image group, the structure-enhanced image group and the original OCTA image in proportion as the input of the segmentation model can prompt the segmentation model to simultaneously learn features under different enhancement methods.

[0083] Based on the above OCTA image blood vessel segmentation method, the present application also proposes an OCTA image quantitative analysis method.

[0084] like Figure 2 As shown, it is a schematic diagram of a process of a quantitative analysis method of OCTA images of the present application, which may include:

[0085] S201, using an OCTA image blood vessel segmentation method to obtain an OCTA image segmentation result.

[0086] The OCTA image is input into the segmentation model, which outputs a segmentation result, usually a two-dimensional matrix or image of the same size as the input image, in which each pixel value indicates whether the location belongs to a blood vessel (for example, binary representation: 1 for blood vessel, 0 for non-blood vessel).

[0087] In practical applications, the segmentation results can also be post-processed, such as removing small isolated areas and smoothing boundaries, to improve the accuracy and visual effect of the segmentation.

[0088] S202: Calculate at least one of a blood vessel curvature parameter, a blood vessel diameter parameter, and a blood vessel density parameter based on the OCTA image segmentation result.

[0089] The vascular images obtained based on segmentation can be used to calculate a variety of vascular parameters, which are very important for evaluating vascular health, diagnosing diseases, and formulating treatment plans. Among them, the vascular curvature parameter can be calculated by analyzing the direction of the vascular centerline. The vascular diameter parameter can be achieved by measuring the diameter of the cross section of each blood vessel in the segmentation result, and calculating the maximum distance or average distance between the vascular boundaries. In some cases, more complex algorithms can also be used to estimate the diameter distribution of blood vessels. The vascular density parameter is defined as the total length or total area of ​​the blood vessels per unit area. A simple vascular density estimate can be obtained by counting the number of vascular pixels in the segmentation result and dividing it by the total area of ​​the image. More complex vascular density parameters can also calculate the vascular density of different regions, analyze the connectivity of the vascular network, etc.

[0090] In practical applications, one or more of the above parameters can be selected for calculation based on research or clinical needs. These parameters can provide doctors with quantitative information about the patient's vascular condition, which helps to diagnose and treat related diseases more accurately.

[0091] The specific descriptions of the above three parameters are as follows:

[0092] The vascular curvature is an important parameter to measure the degree of curvature of the blood vessel in space, which directly reflects the morphological characteristics of the blood vessel. In order to accurately quantify the curvature of the blood vessel, the segmented vascular structure can be refined first, the vascular skeleton curve can be extracted, and the curve for calculating the curvature can be fitted using a high-order curve (B-spline) to ensure the accuracy and precision of the curvature calculation. The local curvature of the calculated curve can be calculated using the following formula:

[0093]

[0094] in, is the parameter along the curve, is the local curvature of the curve, and are the horizontal and vertical coordinate functions of the curve respectively, for The first derivative of for The second-order derivative of for The first derivative of for The second-order derivative of . Through the above formula, the curvature distribution of each blood vessel can be calculated in different areas of the OCTA image.

[0095] In practical applications, if the calculated local curvature is too large, it may indicate vascular lesions or abnormal morphology. Therefore, the statistical distribution of curvature can be used as a reference indicator for disease assessment.

[0096] The vessel curvature parameters can be calculated by:

[0097] (1) Perform necessary preprocessing on OCTA images to ensure data quality, including image denoising and smoothing to reduce noise interference. Then use the segmentation model to segment the blood vessels and obtain a binary vascular structure map as the segmentation result.

[0098] (2) Based on the segmentation results, morphological operations are used to extract the vascular skeleton, and the central axis of each segment of the vascular skeleton is obtained. In practical applications, this can be achieved through a refinement algorithm to ensure the accuracy of vascular skeleton extraction.

[0099] (3) For each segment of the vascular skeleton, a high-order curve fitting technique (such as B-spline) is used to fit the vascular skeleton to obtain a fitting curve to ensure the accuracy of the calculation. The curvature is calculated based on the geometric derivative of the curve, and the relationship between the second-order derivative of the curve and the tangential rate of change of the curve is used to calculate the local curvature of each point. In practical applications, in order to ensure the smoothness of the curvature, especially in the presence of noise or local anomalies, curvature upper and lower limit constraints can also be introduced to exclude abnormal data points and avoid drastic fluctuations in curvature values.

[0100] (4) By performing statistical analysis on the local curvature, key features such as the average curvature and maximum curvature of each blood vessel segment can be extracted, which can be used to reveal the overall morphological changes of the blood vessels and quantify the local geometric features.

[0101] The blood vessel diameter is used to measure the change in the width of the blood vessel. It can be calculated using the following method:

[0102] (1) After obtaining the segmented binary image of the blood vessels, edge detection is first performed to accurately extract the outer edge of the blood vessels. This can be achieved through edge detection algorithms (such as the Sobel operator) to ensure that the boundaries of the blood vessels can be accurately identified.

[0103] (2) By sampling the points on the vascular skeleton in the vertical direction and measuring the distance between the corresponding points on the outer edge of the blood vessel, the local blood vessel diameter is calculated. To ensure the accuracy of the results, a local sampling strategy based on the normal direction of the skeleton points is designed to ensure that the diameter measurement is perpendicular to the direction of the blood vessel at each skeleton point. The local blood vessel diameter can be calculated using the following formula:

[0104]

[0105] in, For the Segment of blood vessels The local diameter value of the sampling point, For the The total number of sampling points on the segmented blood vessel, For the The local blood vessel diameter of the segmented blood vessel. By counting the blood vessel diameters of the entire image, a distribution map of blood vessel diameters can be generated, thus providing a basis for evaluating the health status of the blood vessels.

[0106] (3) In order to eliminate the diameter fluctuation caused by local noise or anomalies, smoothing processing can also be introduced to ensure that the changes in the vascular diameter are more stable by smoothing the local diameter of each segment of the blood vessel. This process can be achieved through a local window smoothing algorithm (Gaussian smoothing), which effectively reduces the impact of outliers on diameter analysis.

[0107] (4) By performing statistical analysis on the local vascular diameter, features such as the average diameter, maximum diameter, and standard deviation can be extracted.

[0108] The vascular density reflects the distribution of blood vessels per unit area and is an important parameter for measuring local microcirculation. The following methods can be used:

[0109] (1) The OCTA image segmentation results are divided into multiple regions of equal size, such as 64×64 pixels or 128×128 pixels, to quantify the vascular density of the local area. We use the sliding window technique to gradually traverse the entire image to ensure that all regions can participate in the density calculation.

[0110] (2) For each divided area, the ratio of blood vessel pixels in the divided area to the total pixels is calculated as the blood vessel density of the divided area. The number of blood vessel pixels in the divided area is directly counted through the binary blood vessel image and compared with the total pixels in the divided area to obtain the blood vessel density of the divided area.

[0111] The vascular density of each divided area is defined as :

[0112]

[0113] in, is the area of ​​blood vessel pixels in the divided region, In this way, a vascular density heat map of the entire OCTA image can be generated, and potential lesion areas can be identified by further analyzing the uniformity of density distribution and local abnormal points.

[0114] (3) In order to intuitively display the density distribution of blood vessels, the calculated vascular density of each divided area can be mapped to the OCTA image segmentation result in a color-coded manner to generate a heat map of the density distribution.

[0115] This application has important innovations in OCTA image generalization segmentation, vascular quantitative analysis, data enhancement strategies, etc., which significantly improves the generalization, robustness and clinical practicality of OCTA image analysis, and promotes the process of OCTA intelligent detection towards accurate clinical application.

[0116] The present application also proposes a computer program product, the computer program product comprising instructions, the instructions being executed by a processor to implement the steps in the above-mentioned OCTA image blood vessel segmentation method embodiment. Alternatively, the processor implementing the steps in the above-mentioned OCTA image quantitative analysis method embodiment when executing the computer program.

[0117] The computer program may be divided into one or more modules / units, and the one or more modules / units are stored in a memory and executed by the processor to accomplish the present invention.

[0118] The carrier for implementing the above-mentioned computer program product may be a computer device, or the computer program product may be stored in a computer storage medium.

[0119] The computer device may be a desktop computer, a notebook computer, a palm computer, a cloud server, etc. The computer device may include, but is not limited to, a processor and a memory.

[0120] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0121] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0122] If the module / unit integrated in the computer device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0123] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for segmenting blood vessels in OCTA images, characterized in that: include: Acquire the acquired OCTA images; Input the acquired OCTA image into the trained segmentation model to obtain the OCTA image segmentation result; The method for obtaining the trained segmentation model comprises: Style enhancement is performed on multiple original OCTA images through NURBS curves to obtain a style-enhanced image group; the control point generation of the NURBS curve is constrained by the pixel value of the OCTA image, the node vector of the NURBS curve includes four groups of nodes, the first group of nodes in the four groups of nodes is equal to the actual minimum normalized result value of the pixels of the OCTA image, the fourth group of nodes is equal to the actual maximum normalized result value of the pixels of the OCTA image, the second group of nodes and the third group of nodes are randomly generated between the first group of nodes and the fourth group of nodes, and the second group of nodes is greater than or equal to the third group of nodes; The multiple original OCTA images are respectively divided into different retinal vascular layers, and each retinal vascular layer is then divided into different vascular branch regions; each vascular branch region is subjected to local elastic deformation to obtain a local elastic deformation field, and then the local elastic deformation fields of different vascular branch regions in the same retinal vascular layer are superimposed to obtain a final result of the local elastic deformation, and a plurality of OCTA images after local elastic deformation are correspondingly obtained as a structure enhanced image group; wherein the local elastic deformation of each vascular branch region includes: By segmenting the labels, vascular functions of different vascular branch regions are constructed; vascular functions of different vascular branch regions, sine functions of different vascular branch regions and random variables are added to obtain fitting functions of different vascular branch regions; fitting functions are applied to corresponding vascular branch regions respectively to obtain local elastic deformation fields of corresponding vascular branch regions; The style enhanced image group, the structure enhanced image group and a plurality of original OCTA images are input into a segmentation model, and the segmentation model is trained to obtain a trained segmentation model.

2. The OCTA image blood vessel segmentation method according to claim 1, characterized in that: The node vector of the NURBS curve includes: in, represents the node vector, represents a node in the second set of nodes, Represents a node in the third set of nodes.

3. The OCTA image blood vessel segmentation method according to claim 2, characterized in that: The method of performing style enhancement on multiple original OCTA images by using NURBS curves includes: Multiply the control points in the NURBS curve by the pixels of the original OCTA image.

4. A method for quantitative analysis of OCTA images, characterized in that: include: Using the OCTA image blood vessel segmentation method described in any one of claims 1 to 3 to obtain an OCTA image segmentation result; Based on the OCTA image segmentation result, at least one of a blood vessel curvature parameter, a blood vessel diameter parameter and a blood vessel density parameter is calculated.

5. The OCTA image quantitative analysis method according to claim 4, characterized in that: The method for calculating the blood vessel curvature parameter comprises: Morphological operations are used to extract the vascular skeleton from the OCTA image segmentation results, and the central axis of each segment of the vascular skeleton corresponding to the blood vessel is obtained; For each segment of the vascular skeleton, a high-order curve fitting technique is used to fit the central axis of the blood vessel corresponding to each segment of the vascular skeleton to obtain a fitting curve; Calculating the geometric derivative of the fitting curve; According to the relationship between the second-order derivative in the geometric derivative and the tangential change rate of the fitting curve, the local curvature of each point on the fitting curve is calculated to obtain the curvature value of each point on the fitting curve; By statistically analyzing the curvature values, the vascular curvature parameters corresponding to each segment of the vascular skeleton are extracted.

6. The OCTA image quantitative analysis method according to claim 5, characterized in that: After obtaining the curvature value of each point on the fitting curve, the method further includes: Set the curvature upper and lower constraints. If the curvature value exceeds the curvature upper or lower constraints, it will be deleted.

7. The OCTA image quantitative analysis method according to claim 6, characterized in that: The method for calculating the blood vessel diameter parameter comprises: Perform edge detection on the OCTA image segmentation results to extract the outer edge of the blood vessels and obtain the blood vessel skeleton; By sampling the points on the vascular skeleton in the vertical direction and measuring the distance between the corresponding points on the outer edge of the blood vessel, the local blood vessel diameter is obtained; Statistical analysis is performed on the local blood vessel diameter to obtain the blood vessel diameter parameters.

8. The OCTA image quantitative analysis method according to claim 7, characterized in that: The method for calculating the blood vessel density parameter comprises: The OCTA image segmentation result is divided into a plurality of regions of equal size to obtain a plurality of divided regions; For each divided area, the ratio of blood vessel pixels in the divided area to the total pixels in the divided area is calculated as the blood vessel density of the divided area; The vascular density of each divided area is mapped to the OCTA image segmentation result in a color-coded manner to obtain a heat map of the density distribution as the calculation result of the vascular density parameter.

9. A computer program product, characterized in that: The computer program product comprises instructions, which when executed by a processor implement an OCTA image blood vessel segmentation method according to any one of claims 1 to 3, or, when executed by a processor, implement an OCTA image quantitative analysis method according to any one of claims 4 to 8.

Images