A blood vessel contour extraction method and device, electronic equipment and storage medium

Abstract

The application provides a blood vessel contour extraction method and device, electronic equipment and storage medium, including: inputting a blood vessel cross-section image set of an acquired 3D blood vessel image to be identified into a pre-trained contour point recognition model, determining a first target polar coordinate point set of a blood vessel inner diameter and a second target polar coordinate point set of a blood vessel outer diameter in each blood vessel cross-section image to be identified; respectively performing coordinate transformation processing on the first target polar coordinate point set and the second target polar coordinate point set of each blood vessel cross-section image to be identified, determining a first target rectangular coordinate point set of the blood vessel inner diameter and a second target rectangular coordinate point set of the blood vessel outer diameter in each blood vessel cross-section image to be identified; and sequentially performing curve fitting processing on each target rectangular coordinate point set, determining a blood vessel inner diameter contour curve and a blood vessel outer diameter contour curve of the 3D blood vessel image to be identified. In this way, the speed and accuracy of blood vessel inner and outer diameter contour extraction can be effectively improved by the technical solution of the application.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, and in particular to a blood vessel contour extraction method and device, electronic equipment and a storage medium. BACKGROUND

[0002] Globally, cardiovascular and cerebrovascular diseases have become one of the major diseases threatening human health, so in medical imaging, accurately segmenting blood vessels such as coronary artery blood vessels and extracting blood vessel contours is of great significance for determining coronary artery stenosis, lesions, etc.

[0003] In the prior art, intravenous ultrasound (IVUS) technology is usually used for intramural and adventitial edge detection, wherein intravenous ultrasound (IVUS) refers to a medical imaging technology that combines non-invasive ultrasound technology and invasive catheter technology, using a special catheter with an ultrasonic probe at the end.

[0004] However, since the IVUS image twists with the beating of the heart, it is not conducive to the extraction of the intima contour of the lumen, resulting in slow speed and low accuracy of the extraction of the intima contour of the lumen. Therefore, how to improve the accuracy of blood vessel contour extraction is a technical problem to be solved by those skilled in the art. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a blood vessel contour extraction method, device, electronic equipment and storage medium, which can effectively improve the speed and accuracy of the extraction of the intramural and adventitial contour.

[0006] The present application provides a blood vessel contour extraction method, which comprises:

[0007] Obtaining a blood vessel cross-sectional image set of a to-be-recognized 3D blood vessel image; the blood vessel cross-sectional image set comprises a plurality of to-be-recognized blood vessel cross-sectional images, and the to-be-recognized blood vessel cross-sectional image is an image obtained according to a centerline point on a blood vessel centerline of the to-be-recognized 3D blood vessel image;

[0008] Inputting the blood vessel cross-sectional image set into a pre-trained contour point recognition model to determine a first target polar coordinate point set of an intramural diameter and a second target polar coordinate point set of an extramural diameter in each to-be-recognized blood vessel cross-sectional image in the blood vessel cross-sectional image set;

[0009] perform coordinate transformation processing on a first target polar coordinate point set of the inner diameter of the blood vessel and a second target polar coordinate point set of the outer diameter of the blood vessel in each to-be-identified blood vessel cross-section image, to determine a first target rectangular coordinate point set of the inner diameter of the blood vessel and a second target rectangular coordinate point set of the outer diameter of the blood vessel in each to-be-identified blood vessel cross-section image;

[0010] perform curve fitting processing on the first target rectangular coordinate point set of the inner diameter of the blood vessel and the second target rectangular coordinate point set of the outer diameter of the blood vessel in each to-be-identified blood vessel cross-section image in sequence, to determine an inner diameter profile curve and an outer diameter profile curve of the to-be-identified 3D blood vessel image.

[0011] Optionally, the profile point identification model is constructed through the following steps:

[0012] obtain a sample blood vessel cross-section image set of a plurality of sample 3D blood vessel images; each sample blood vessel cross-section image in the sample blood vessel cross-section image set corresponds to a first real polar coordinate point set of the inner diameter of the blood vessel and a second real polar coordinate point set of the outer diameter of the blood vessel;

[0013] input the sample blood vessel cross-section image set of each sample 3D blood vessel image into the initial profile point identification neural network in sequence, to predict a first predicted polar coordinate point set of the inner diameter of the blood vessel and a second predicted polar coordinate point set of the outer diameter of the blood vessel in each sample blood vessel cross-section image;

[0014] determine a target loss function value based on the first real polar coordinate point set, the second real polar coordinate point set, the first predicted polar coordinate point set, and the second predicted polar coordinate point set of each sample blood vessel cross-section image;

[0015] based on the target loss function value, iteratively train the initial profile point identification neural network and update the network parameters of the initial profile point identification neural network until the target loss function value converges, stop training, and obtain the profile point identification model.

[0016] Optionally, the first real polar coordinate point set of the inner diameter of the blood vessel and the second real polar coordinate point set of the outer diameter of the blood vessel of each sample blood vessel cross-section image in the sample blood vessel cross-section image set are determined through the following steps:

[0017] perform center line extraction processing on the sample 3D blood vessel image to determine a blood vessel center line of the sample 3D blood vessel image; the blood vessel center line includes a plurality of center line points;

[0018] for each center line point, intercept a preset size plane in a direction perpendicular to a tangent of the blood vessel center line and including the center line point, to determine a sample blood vessel cross-section image corresponding to the center line point;

[0019] For each sample blood vessel cross-section image, based on the inner and outer diameter contour labeling results on the sample blood vessel cross-section image, contour coordinate point extraction processing is performed to determine the first true rectangular coordinate point set of the inner diameter of the blood vessel and the second true rectangular coordinate point set of the outer diameter of the blood vessel in the sample blood vessel cross-section image;

[0020] For each sample blood vessel cross-section image, based on the inner and outer diameter contour labeling results on the sample blood vessel cross-section image, contour coordinate point extraction processing is performed to determine the first true rectangular coordinate point set of the inner diameter of the blood vessel and the second true rectangular coordinate point set of the outer diameter of the blood vessel in the sample blood vessel cross-section image;

[0021] Optionally, the target loss function value is determined based on the first true polar coordinate point set, the second true polar coordinate point set, the first predicted polar coordinate point set and the second predicted polar coordinate point set of each sample blood vessel cross-section image, comprising:

[0022] Based on the first true polar coordinate point set and the second true polar coordinate point set, a true interval set of the inner diameter of the blood vessel and the outer diameter of the blood vessel is determined;

[0023] Based on the first predicted polar coordinate point set and the second predicted polar coordinate point set, a predicted interval set of the inner diameter of the blood vessel and the outer diameter of the blood vessel is determined;

[0024] The first loss function value is determined by comparing the first predicted polar coordinate point set with the first true polar coordinate point set;

[0025] The second loss function value is determined by comparing the second predicted polar coordinate point set with the second true polar coordinate point set;

[0026] The third loss function value is determined by comparing the predicted interval set with the true interval set;

[0027] The target loss function value is determined based on the first loss function value, the second loss function value and the third loss function value.

[0028] Optionally, the target loss function value is determined based on the first loss function value, the second loss function value and the third loss function value, comprising:

[0029] The first loss function value, the second loss function value and the third loss function value are respectively multiplied by their corresponding weight coefficients, and the three loss function values multiplied by the weight coefficients are added to determine the target loss function value.

[0030] Optionally, before sequentially performing curve fitting processing on the first target rectangular coordinate point set of the inner diameter of the blood vessel and the second target rectangular coordinate point set of the outer diameter of the blood vessel in each to-be-identified blood vessel cross-section image, the extraction method further comprises:

[0031] respectively, the first target rectangular coordinate point set and the second target rectangular coordinate point set are subjected to interpolation processing, so as to increase the number of coordinate points in the first target rectangular coordinate point set and the second target rectangular coordinate point set.

[0032] Optionally, the contour point recognition model is a single-task recognition model or a multi-task recognition model.

[0033] The embodiment of the present application further provides an extraction device for a blood vessel contour, and the extraction device comprises:

[0034] An acquisition module is configured to acquire a blood vessel cross-section image set of a 3D blood vessel image to be recognized, wherein the blood vessel cross-section image set comprises a plurality of blood vessel cross-section images to be recognized, and each blood vessel cross-section image to be recognized is an image acquired according to a center line point on a center line of the 3D blood vessel image to be recognized;

[0035] A first determination module is configured to input the blood vessel cross-section image set into a pre-trained contour point recognition model, and determine a first target polar coordinate point set of an inner diameter of a blood vessel and a second target polar coordinate point set of an outer diameter of the blood vessel in each blood vessel cross-section image to be recognized in the blood vessel cross-section image set.

[0036] A coordinate conversion module is configured to respectively perform coordinate conversion processing on the first target polar coordinate point set of the inner diameter of the blood vessel and the second target polar coordinate point set of the outer diameter of the blood vessel in each blood vessel cross-section image to be recognized, and determine a first target rectangular coordinate point set of the inner diameter of the blood vessel and a second target rectangular coordinate point set of the outer diameter of the blood vessel in each blood vessel cross-section image to be recognized.

[0037] A fitting processing module is configured to sequentially perform curve fitting processing on the first target rectangular coordinate point set of the inner diameter of the blood vessel and the second target rectangular coordinate point set of the outer diameter of the blood vessel in each blood vessel cross-section image to be recognized, and determine an inner diameter contour curve of the blood vessel and an outer diameter contour curve of the blood vessel of the 3D blood vessel image to be recognized.

[0038] Optionally, the extraction device further comprises a model construction module, and the model construction module is configured to:

[0039] A plurality of sample blood vessel cross-section image sets of sample 3D blood vessel images are acquired, wherein each sample blood vessel cross-section image in the sample blood vessel cross-section image set corresponds to a first real polar coordinate point set of an inner diameter of a blood vessel and a second real polar coordinate point set of an outer diameter of the blood vessel.

[0040] The sample blood vessel cross-section image set of each sample 3D blood vessel image is sequentially input into an initial contour point recognition neural network, and a first predicted polar coordinate point set of an inner diameter of a blood vessel and a second predicted polar coordinate point set of an outer diameter of the blood vessel in each sample blood vessel cross-section image are predicted.

[0041] determine a target loss function value based on the first real polar coordinate point set, the second real polar coordinate point set, the first predicted polar coordinate point set and the second predicted polar coordinate point set of each sample vessel cross-section image;

[0042] perform iterative training on the initial contour point recognition neural network and update network parameters of the initial contour point recognition neural network based on the target loss function value until the target loss function value converges, stop training, and obtain the contour point recognition model.

[0043] Optionally, the extraction device further comprises a second determination module, which is configured to:

[0044] perform center line extraction processing on the sample 3D vessel image to determine a vessel center line of the sample 3D vessel image; wherein the vessel center line comprises a plurality of center line points;

[0045] For each center line point, a preset size plane in the direction of the perpendicular vessel center line tangent line including the center line point is intercepted to determine a sample vessel cross-section image corresponding to the center line point;

[0046] For each sample vessel cross-section image, based on the inner and outer diameter contour annotation results on the sample vessel cross-section image, contour coordinate point extraction processing is performed to determine a first real rectangular coordinate point set of the vessel inner diameter and a second real rectangular coordinate point set of the vessel outer diameter in the sample vessel cross-section image;

[0047] For each real polar coordinate point set in each sample vessel cross-section image, coordinate conversion processing is performed based on the center line point in the sample vessel cross-section image to determine a first real polar coordinate point set of the vessel inner diameter and a second real polar coordinate point set of the vessel outer diameter in the sample vessel cross-section image.

[0048] Optionally, when the model construction module is used to determine a target loss function value based on the first real polar coordinate point set, the second real polar coordinate point set, the first predicted polar coordinate point set and the second predicted polar coordinate point set of each sample vessel cross-section image, the model construction module is configured to:

[0049] determine a real distance set of the vessel inner diameter and the vessel outer diameter based on the first real polar coordinate point set and the second real polar coordinate point set;

[0050] determine a predicted distance set of the vessel inner diameter and the vessel outer diameter based on the first predicted polar coordinate point set and the second predicted polar coordinate point set;

[0051] use the first predicted polar coordinate point set and the first real polar coordinate point set for comparison to determine a first loss function value;

[0052] determining a second loss function value by comparing the second predicted polar coordinate point set and the second real polar coordinate point set;

[0053] determining a third loss function value by comparing the predicted distance set and the real distance set;

[0054] determining the target loss function value based on the first loss function value, the second loss function value and the third loss function value.

[0055] Optionally, when the model construction module is used to determine the target loss function value based on the first loss function value, the second loss function value and the third loss function value, the model construction module is used to:

[0056] multiplying the first loss function value, the second loss function value and the third loss function value by respective weight coefficients respectively, and adding the three loss function values multiplied by the weight coefficients to determine the target loss function value.

[0057] Optionally, the extraction device further comprises an interpolation module, which is used to perform interpolation processing on the first target rectangular coordinate point set of the first target inner diameter of the blood vessel and the second target rectangular coordinate point set of the second target outer diameter of the blood vessel respectively before sequentially performing curve fitting processing on the first target rectangular coordinate point set and the second target rectangular coordinate point set in each of the to-be-identified blood vessel cross-sectional images, so as to increase the number of coordinate points in the first target rectangular coordinate point set and the second target rectangular coordinate point set.

[0058] Optionally, the contour point recognition model is a single-task recognition model or a multi-task recognition model.

[0059] Embodiments of the present application also provide an electronic device, which comprises a processor, a memory and a bus, the memory stores machine readable instructions executable by the processor, when the electronic device is running, the processor and the memory communicate through the bus, and the machine readable instructions are executed by the processor to perform the steps of the extraction method as described above.

[0060] Embodiments of the present application also provide a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to perform the steps of the extraction method as described above.

[0061] The embodiment of the application provides a blood vessel contour extraction method and device, electronic equipment and storage medium, which comprises the following steps: acquiring a blood vessel cross-section image set of a to-be-identified 3D blood vessel image; the blood vessel cross-section image set comprises a plurality of to-be-identified blood vessel cross-section images, and the to-be-identified blood vessel cross-section image is an image obtained by cutting a center line point on a blood vessel center line of the to-be-identified 3D blood vessel image; inputting the blood vessel cross-section image set into a pre-trained contour point identification model, determining a first target polar coordinate point set of a blood vessel inner diameter and a second target polar coordinate point set of a blood vessel outer diameter in each to-be-identified blood vessel cross-section image in the blood vessel cross-section image set; performing coordinate conversion processing on the first target polar coordinate point set of the blood vessel inner diameter and the second target polar coordinate point set of the blood vessel outer diameter in each to-be-identified blood vessel cross-section image respectively, to determine a first target rectangular coordinate point set of the blood vessel inner diameter and a second target rectangular coordinate point set of the blood vessel outer diameter in each to-be-identified blood vessel cross-section image; and sequentially performing curve fitting processing on the first target rectangular coordinate point set of the blood vessel inner diameter and the second target rectangular coordinate point set of the blood vessel outer diameter in each to-be-identified blood vessel cross-section image, to determine a blood vessel inner diameter contour curve and a blood vessel outer diameter contour curve of the to-be-identified 3D blood vessel image.

[0062] In this way, by converting the rectangular coordinates into polar coordinates, background interference is reduced, model learning is simplified, and model learning speed is improved; by calculating a loss function based on the distance between the inner diameter and the outer diameter, blood vessel wall thickness learning is enhanced, so that the model recognition accuracy is improved, and the accuracy of extracting the inner and outer diameter contours of the blood vessel is improved.

[0063] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the following preferred embodiments are specifically described below, and the accompanying drawings are described in detail as follows. BRIEF DESCRIPTION OF DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0065] Figure 1 The flow chart of the blood vessel contour extraction method provided by the embodiments of the application;

[0066] Figure 2 The structural schematic diagram of the blood vessel center line provided by the application;

[0067] Figure 3 The structural schematic diagram of the blood vessel inner and outer diameter contour annotation result provided by the application;

[0068] Figure 4A schematic diagram of a coordinate conversion principle provided for the present application;

[0069] Figure 5 A structural schematic diagram of a blood vessel inner diameter profile provided for the present application;

[0070] Figure 6 A structural schematic diagram of a blood vessel profile extraction device provided for an embodiment of the present application.

[0071] Figure 7 A structural schematic diagram of a blood vessel profile extraction device provided for an embodiment of the present application.

[0072] Figure 8 A structural schematic diagram of an electronic device provided for an embodiment of the present application. DETAILED DESCRIPTION

[0073] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the following will be combined with the accompanying drawings for the embodiments of the present application to make a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only a 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 accompanying drawings can be arranged and designed in various different configurations. 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 claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, every other embodiment obtained by a person skilled in the art without creative work falls within the scope of protection of the present application.

[0074] Globally, cardiovascular and cerebrovascular diseases have become one of the major diseases threatening human health, so in medical imaging, it is of great significance to accurately segment blood vessels such as coronary artery vessels and extract blood vessel profiles for determining the stenosis, lesions, etc. of coronary arteries.

[0075] In the prior art, intravenous ultrasound (IVUS) technology is usually used for intramural and adventitial edge detection, wherein intravenous ultrasound (IVUS) refers to a medical imaging technology that combines non-invasive ultrasound technology and invasive catheter technology, using a special catheter with an ultrasonic probe at the end.

[0076] However, since the IVUS image twists with the beating of the heart, it is not conducive to the extraction of the intimal profile of the lumen, resulting in a slow speed and low accuracy of the extraction of the intimal profile of the lumen. Therefore, how to improve the accuracy of the extraction of the blood vessel profile is a technical problem to be solved by those skilled in the art.

[0077] Based on this, the application embodiment provides a blood vessel contour extraction method and device, electronic equipment and storage medium, which can effectively improve the speed and accuracy of blood vessel inner and outer diameter contour extraction.

[0078] Please refer to Figure 1 , Figure 1 The flowchart of the blood vessel contour extraction method provided by the application embodiment is shown in the following figure. Figure 1 As shown in the figure, the extraction method provided by the application embodiment comprises the following steps.

[0079] S101, acquiring a blood vessel cross-section image set of a to-be-identified 3D blood vessel image.

[0080] Here, the blood vessel cross-section image set comprises a plurality of to-be-identified blood vessel cross-section images, which are images taken according to the centerline points on the blood vessel centerline of the to-be-identified 3D blood vessel image.

[0081] The to-be-identified blood vessel cross-section image has the following characteristics: a plane perpendicular to the centerline tangent direction of the centerline point, and a certain size (including the blood vessel range) of the original medical image.

[0082] The to-be-identified 3D blood vessel image can be a CT, MRI or other medical image, and specifically can be a 3D blood vessel image at the coronary position.

[0083] The blood vessel centerline is composed of a series of continuous points (pixel space). For example, please refer to Figure 2 , Figure 2 The structure diagram of the blood vessel centerline provided by the application is shown in the following figure. The centerline points can be a plurality of points extracted from the blood vessel centerline at equal intervals, or a plurality of points extracted from the blood vessel centerline according to certain rules. The to-be-identified blood vessel cross-section is a blood vessel cross-section taken by the centerline point, and each centerline point can take a to-be-identified blood vessel cross-section. The number of images in the blood vessel cross-section image set is determined by the number of selected centerline points.

[0084] S102, inputting the blood vessel cross-section image set to a pre-trained contour point recognition model to determine a first target polar coordinate point set of the blood vessel inner diameter and a second target polar coordinate point set of the blood vessel outer diameter in each to-be-identified blood vessel cross-section image in the blood vessel cross-section image set.

[0085] Here, the contour point recognition model can be a single-task recognition model or a multi-task recognition model.

[0086] When the contour point recognition model is a single-task recognition model, a group of vectors after combination of the inner diameter contour points and the outer diameter contour points is output, and the first target polar coordinate point set of the inner diameter of the blood vessel and the second target polar coordinate point set of the outer diameter of the blood vessel can be obtained by segmenting and decoding the combined vectors.

[0087] When the contour point recognition model is a multi-task recognition model, two groups of vectors of the inner diameter contour points and the outer diameter contour points are output, and the first target polar coordinate point set of the inner diameter of the blood vessel and the second target polar coordinate point set of the outer diameter of the blood vessel can be obtained by decoding the two groups of vectors.

[0088] The single-task recognition model has the advantage of high efficiency compared with the multi-task recognition model, and the multi-task recognition model has the advantage of high accuracy compared with the single-task recognition model.

[0089] In an implementation, the contour point recognition model is constructed by the following steps:

[0090] S201, a sample blood vessel cross-section image set of a plurality of sample 3D blood vessel images is obtained.

[0091] Here, each sample blood vessel cross-section image in the sample blood vessel cross-section image set corresponds to a first real polar coordinate point set of the inner diameter of the blood vessel and a second real polar coordinate point set of the outer diameter of the blood vessel.

[0092] The first real polar coordinate point set and the second real polar coordinate point set can be a point set composed of a preset number of coordinate points subjected to discrete sampling.

[0093] The sample blood vessel cross-section image set is obtained in the same way as the sample blood vessel cross-section image set in step S101, and details are not repeated here.

[0094] S202, the sample blood vessel cross-section image set of each sample 3D blood vessel image is sequentially input into an initial contour point recognition neural network to predict a first predicted polar coordinate point set of the inner diameter of the blood vessel and a second predicted polar coordinate point set of the outer diameter of the blood vessel in each sample blood vessel cross-section image.

[0095] Here, the initial contour point recognition neural network can use a classic CNN network or a more advanced specific network, which is not limited here.

[0096] During model training, independent training of the inner and outer diameter contour points can be performed, and the training task can use MultiClass, MultiTask, etc.; or combined training of the inner and outer diameter contour points can be performed, the number of inner and outer diameter contour points is fixed respectively, the inner diameter and the outer diameter 2 vectors are combined into one 1 vector, and a single-task vector regression method can be specifically used.

[0097] S203, determine a target loss function value based on the first real polar coordinate point set, the second real polar coordinate point set, the first predicted polar coordinate point set and the second predicted polar coordinate point set of each sample blood vessel cross-section image.

[0098] Here, the target loss function value is determined according to the error between the predicted value and the real value.

[0099] S204, iteratively train the initial contour point recognition neural network and update the network parameters of the initial contour point recognition neural network based on the target loss function value until the target loss function value converges, stop training, and obtain the contour point recognition model.

[0100] In an embodiment, for the polar coordinate point set in step S201, the first real polar coordinate point set of the inner diameter and the second real polar coordinate point set of the outer diameter of each sample blood vessel cross-section image in the sample blood vessel cross-section image set can be determined by the following steps:

[0101] S301, perform center line extraction processing on the sample 3D blood vessel image to determine the blood vessel center line of the sample 3D blood vessel image.

[0102] Here, the center line extraction processing is performed on each sample 3D blood vessel image to determine the blood vessel center line corresponding to each sample 3D blood vessel image. The blood vessel center line is composed of a series of continuous points (pixel space), and the blood vessel center line includes a plurality of center line points, which are selected from a plurality of target points on the blood vessel center line.

[0103] S302, for each center line point, intercept a preset size plane in the direction of the vertical blood vessel center line tangent line including the center line point to determine the sample blood vessel cross-section image corresponding to the center line point.

[0104] Here, the preset size can be dynamically changed according to the thickness of the blood vessel, or it can be a pre-set fixed value. The number of sample blood vessel cross-section images is the same as the number of selected center line points.

[0105] S303, for each sample blood vessel cross-section image, based on the inner and outer diameter contour annotation result on the sample blood vessel cross-section image, perform contour coordinate point extraction processing to determine the first real rectangular coordinate point set of the inner diameter and the second real rectangular coordinate point set of the outer diameter in the sample blood vessel cross-section image.

[0106] Here, the inner and outer diameter contour annotation result on the sample blood vessel cross-section image is a manual annotation according to prior knowledge and / or gradient map after the sample blood vessel cross-section image is determined. The contour annotation result can be two closed curves, one for the inner diameter and one for the outer diameter.

[0107] For example, refer to Figure 3 , Figure 3 A structural diagram of a blood vessel inner and outer diameter contour labeling result provided by the present application. Figure 3 (a) in FIG. 1 is a schematic diagram of an inner and outer diameter contour labeling result on a sample blood vessel cross-section image based on IVUS (intravascular ultrasound) technology using prior knowledge, Figure 3 (b) in FIG. 1 is an inner and outer diameter contour labeling result on a sample blood vessel cross-section image based on a gradient map.

[0108] In this way, according to the labeling result, a first real rectangular coordinate point set of the inner diameter of the blood vessel and a second real rectangular coordinate point set of the outer diameter of the blood vessel in the sample blood vessel cross-section image can be directly extracted. The rectangular coordinate point set is also a Cartesian coordinate point set.

[0109] S304, for each real polar coordinate point set in the sample blood vessel cross-section image, based on the center line point in the sample blood vessel cross-section image, a coordinate conversion process is performed to determine a first real polar coordinate point set of the inner diameter of the blood vessel and a second real polar coordinate point set of the outer diameter of the blood vessel in the sample blood vessel cross-section image.

[0110] Here, the coordinate conversion process is to convert the Cartesian coordinate point into a polar coordinate point.

[0111] It should be noted that the reason for performing the coordinate conversion process is that there are the following difficulties in profile prediction based on the Cartesian coordinate system: 1. The given learning information is chaotic and the background occupies a large proportion: due to the different sizes of blood vessels, the large size of the cross-section image needs to be considered, which will increase a lot of background interference; the profile center information of the cross-section image may not be consistent: the brightness of the angiographic agent image, the near and far sections and the false shadow are inconsistent; 2. The learning target is multiple, including the blood vessel profile center point and the profile point set composed of the blood vessel profile.

[0112] And converting the rectangular coordinate (Cartesian coordinate) into polar coordinate simplifies the model training task, because the learned coordinate point (x, y) is converted into a learned radius r; so that the input information of learning has no excessive interference: the background area is reduced; and the learning target is reduced: from learning (x, y) 2 features, to learning radius r, and can be reduced to a simple regression task.

[0113] For example, refer to Figure 4 , Figure 4 A schematic diagram of a coordinate conversion principle provided by the present application. As shown in Figure 4As shown, the left figure is the representation of the profile points in the rectangular coordinate system, and the right figure is the representation of the profile points in the polar coordinate system. Here, only the representation of the profile points of one of the inner diameter or the outer diameter is shown. Here, the spatial relationship in the Cartesian coordinate system is converted into the relationship of the polar coordinate with respect to the rotation angle θ and the radius r; the rotation angle θ of the polar coordinate is unfolded in sequence, and the radius r is described as a continuous curve. In this way, only one parameter of the radius r needs to be learned during model training, simplifying the task.

[0114] In an embodiment, the first real polar coordinate point set, the second real polar coordinate point set, the first predicted polar coordinate point set, and the second predicted polar coordinate point set based on each sample blood vessel cross-sectional image in step S203 are used to determine the target loss function value, including:

[0115] S2031, based on the first real polar coordinate point set and the second real polar coordinate point set, the real interval set of the inner diameter and the outer diameter of the blood vessel is determined.

[0116] It should be noted that the number of coordinate points in the first real polar coordinate point set and the second real polar coordinate point set can be the same or different. When they are different, the number of points is mainly determined by the smaller number of inner and outer diameter intervals.

[0117] S2032, based on the first predicted polar coordinate point set and the second predicted polar coordinate point set, a predicted interval set of the inner diameter and the outer diameter of the blood vessel is determined.

[0118] The determination method is the same as step S2031, which will not be described here.

[0119] S2033, the first loss function value is determined by comparing the first predicted polar coordinate point set with the first real polar coordinate point set.

[0120] S2034, the second loss function value is determined by comparing the second predicted polar coordinate point set with the second real polar coordinate point set.

[0121] S2035, the third loss function value is determined by comparing the predicted interval set with the real interval set.

[0122] S2036, the target loss function value is determined based on the first loss function value, the second loss function value, and the third loss function value.

[0123] In an embodiment, the determining the target loss function value based on the first loss function value, the second loss function value and the third loss function value in step S2036 comprises: multiplying the first loss function value, the second loss function value and the third loss function value by respective weight coefficients respectively, and adding the three loss function values multiplied by the weight coefficients to determine the target loss function value.

[0124] The first loss function value, the second loss function value and the third loss function value and the respective weight coefficients are pre-set. Here, the relative distance between the inner diameter and the outer diameter is calculated to determine the loss function value, which enhances the learning of the blood vessel wall thickness.

[0125] S103, respectively, the first target polar coordinate point set of the inner diameter of the blood vessel and the second target polar coordinate point set of the outer diameter of the blood vessel in each to-be-identified blood vessel cross-section image are subjected to coordinate transformation processing to determine the first target rectangular coordinate point set of the inner diameter of the blood vessel and the second target rectangular coordinate point set of the outer diameter of the blood vessel in each to-be-identified blood vessel cross-section image.

[0126] Here, the included angle value between two adjacent points in the target polar coordinate point set is fixed. For example, assuming that the first target polar coordinate point set corresponding to each to-be-identified blood vessel cross-section image includes 60 points, the included angle between two adjacent points is 6°.

[0127] S104, the first target rectangular coordinate point set of the inner diameter of the blood vessel and the second target rectangular coordinate point set of the outer diameter of the blood vessel in each to-be-identified blood vessel cross-section image are subjected to curve fitting processing in sequence to determine the inner diameter profile curve and the outer diameter profile curve of the to-be-identified 3D blood vessel image.

[0128] In an embodiment, before the first target rectangular coordinate point set of the inner diameter of the blood vessel and the second target rectangular coordinate point set of the outer diameter of the blood vessel in each to-be-identified blood vessel cross-section image are subjected to curve fitting processing in sequence, the extraction method further comprises: respectively, the first target rectangular coordinate point set and the second target rectangular coordinate point set are subjected to interpolation processing to increase the number of coordinate points in the first target rectangular coordinate point set and the second target rectangular coordinate point set.

[0129] Here, the interpolation processing is performed on the profile point set to be closer to the true shape of the profile and to increase the profile points to enrich the profile details.

[0130] For example, please refer to 5, Figure 5 A structural schematic diagram of a blood vessel inner diameter profile line is provided in the present application. Here, only the inner diameter profile line is displayed for better observation, and the outer diameter profile line is similar to the inner diameter profile line.

[0131] In addition, after the inner diameter profile line and the outer diameter profile line of the blood vessel are determined, the inner diameter profile surface and the outer diameter profile surface can also be determined through curved surface smoothing processing.

[0132] In this way, by converting the rectangular coordinates into polar coordinates, background interference is reduced, model learning is simplified, and model learning speed is improved; by calculating the loss function based on the distance between the inner diameter and the outer diameter, blood vessel wall thickness learning is enhanced, thereby improving model recognition accuracy and further improving the accuracy of extracting the inner and outer diameter profiles of the blood vessel.

[0133] Please refer to Figure 6 、 Figure 7 , Figure 6 Figure 1 is a structural schematic diagram of a blood vessel profile extraction device provided by an embodiment of the present application, Figure 7 Figure 2 is a structural schematic diagram of a blood vessel profile extraction device provided by another embodiment of the present application. As shown in Figure 6 the extraction device 600 comprises:

[0134] The acquisition module 610 is configured to acquire a set of blood vessel cross-section images of a 3D blood vessel image to be recognized; the set of blood vessel cross-section images comprises a plurality of blood vessel cross-section images to be recognized, and each blood vessel cross-section image to be recognized is an image taken according to a center line point on a center line of the 3D blood vessel image to be recognized;

[0135] The first determination module 620 is configured to input the set of blood vessel cross-section images to a pre-trained profile point recognition model, and determine a first target polar coordinate point set of an inner diameter of the blood vessel and a second target polar coordinate point set of an outer diameter of the blood vessel in each blood vessel cross-section image to be recognized in the set of blood vessel cross-section images;

[0136] The coordinate conversion module 630 is configured to perform coordinate conversion processing on the first target polar coordinate point set of the inner diameter of the blood vessel and the second target polar coordinate point set of the outer diameter of the blood vessel in each blood vessel cross-section image to be recognized, respectively, and determine a first target rectangular coordinate point set of the inner diameter of the blood vessel and a second target rectangular coordinate point set of the outer diameter of the blood vessel in each blood vessel cross-section image to be recognized.

[0137] The fitting processing module 640 is configured to sequentially perform curve fitting processing on the first target rectangular coordinate point set of the inner diameter of the blood vessel and the second target rectangular coordinate point set of the outer diameter of the blood vessel in each blood vessel cross-section image to be recognized, and determine an inner diameter profile curve and an outer diameter profile curve of the 3D blood vessel image to be recognized.

[0138] Optionally, as shown in Figure 7 the extraction device 600 further comprises a model construction module 650, and the model construction module 650 is configured to:

[0139] obtain a sample vessel cross-section image set of a plurality of sample 3D vessel images; wherein each sample vessel cross-section image in the sample vessel cross-section image set corresponds to a first real polar coordinate point set of a vessel inner diameter and a second real polar coordinate point set of a vessel outer diameter;

[0140] input the sample vessel cross-section image set of each sample 3D vessel image into the initial contour point recognition neural network in sequence, and predict a first predicted polar coordinate point set of the vessel inner diameter and a second predicted polar coordinate point set of the vessel outer diameter in each sample vessel cross-section image;

[0141] determine a target loss function value based on the first real polar coordinate point set, the second real polar coordinate point set, the first predicted polar coordinate point set, and the second predicted polar coordinate point set of each sample vessel cross-section image;

[0142] perform iterative training on the initial contour point recognition neural network and update network parameters of the initial contour point recognition neural network based on the target loss function value, until the target loss function value converges, stop training, and obtain the contour point recognition model.

[0143] Optionally, the extraction device 600 further includes a second determination module 660, which is configured to:

[0144] perform center line extraction processing on the sample 3D vessel image to determine a vessel center line of the sample 3D vessel image; wherein the vessel center line includes a plurality of center line points;

[0145] for each center line point, intercept a preset size plane in a direction perpendicular to a vessel center line tangent line of the center line point to determine a sample vessel cross-section image corresponding to the center line point;

[0146] for each sample vessel cross-section image, perform contour coordinate point extraction processing based on inner and outer diameter contour annotation results on the sample vessel cross-section image to determine a first real rectangular coordinate point set of a vessel inner diameter and a second real rectangular coordinate point set of a vessel outer diameter in the sample vessel cross-section image;

[0147] for each real polar coordinate point set in each sample vessel cross-section image, perform coordinate conversion processing based on a center line point in the sample vessel cross-section image to determine a first real polar coordinate point set of a vessel inner diameter and a second real polar coordinate point set of a vessel outer diameter in the sample vessel cross-section image.

[0148] Optionally, when the model construction module 650 is used to determine a target loss function value based on the first real polar coordinate point set, the second real polar coordinate point set, the first predicted polar coordinate point set, and the second predicted polar coordinate point set of each sample vessel cross-section image, the model construction module 650 is configured to:

[0149] determine a real interval set of the inner diameter and the outer diameter of the blood vessel based on the first real polar coordinate point set and the second real polar coordinate point set;

[0150] determine a predicted interval set of the inner diameter and the outer diameter of the blood vessel based on the first predicted polar coordinate point set and the second predicted polar coordinate point set;

[0151] determine a first loss function value by comparing the first predicted polar coordinate point set and the first real polar coordinate point set;

[0152] determine a second loss function value by comparing the second predicted polar coordinate point set and the second real polar coordinate point set;

[0153] determine a third loss function value by comparing the predicted interval set and the real interval set;

[0154] determine the target loss function value based on the first loss function value, the second loss function value and the third loss function value.

[0155] Optionally, when the model construction module 650 is used to determine the target loss function value based on the first loss function value, the second loss function value and the third loss function value, the model construction module 650 is configured to:

[0156] multiply the first loss function value, the second loss function value and the third loss function value by respective weight coefficients respectively, and add the three loss function values multiplied by the weight coefficients to determine the target loss function value.

[0157] Optionally, the extraction device 600 further comprises an interpolation module 670, which is configured to: before sequentially performing curve fitting processing on the first target rectangular coordinate point set of the inner diameter and the second target rectangular coordinate point set of the outer diameter of the blood vessel in each to-be-identified blood vessel cross-section image, respectively perform interpolation processing on the first target rectangular coordinate point set and the second target rectangular coordinate point set to increase the number of coordinate points in the first target rectangular coordinate point set and the second target rectangular coordinate point set.

[0158] Optionally, the contour point identification model is a single-task identification model or a multi-task identification model.

[0159] Please refer to Figure 8 , Figure 8 a structural schematic diagram of an electronic device provided by an embodiment of the present application. As shown in Figure 8 the electronic device 800 comprises a processor 810, a memory 820 and a bus 830.

[0160] The memory 820 stores machine readable instructions executable by the processor 810, when the electronic device 800 is running, the processor 810 and the memory 820 communicate through the bus 830, the machine readable instructions are executed by the processor 810, can execute the steps in the method embodiments as described above Figures 1 to 5 The specific implementation can be referred to the method embodiments, and details are not described here.

[0161] The embodiments of the application also provide a computer readable storage medium, the computer readable storage medium stores a computer program, the computer program is run by the processor, and the computer program can execute the steps in the method embodiments as described above Figures 1 to 5 The specific implementation can be referred to the method embodiments, and details are not described here.

[0162] The skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, and details are not described here.

[0163] In several embodiments provided in the application, it should be understood that the disclosed system, device and method can be implemented by other ways.The device embodiments described above are only schematic, for example, the division of the unit is only a logical function division, and actual implementation can have another division mode, and for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed.In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some communication interface, device or unit, which can be electrical, mechanical or other forms.

[0164] The unit described as a separate component can be or can not be physically separated, and the component shown as a unit can be or can not be a physical unit, that is, it can be located in one place, or it can be distributed on a plurality of network units.According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0165] In addition, the functional units in each embodiment of the application can be integrated in one processing unit, or each unit can be physically present, or two or more units can be integrated in one unit.

[0166] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a nonvolatile computer readable storage medium executable by a processor. Based on this understanding, the technical solutions of the present application essentially or the parts of the prior art that make contributions or parts of the technical solutions can be embodied in the form of software products. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0167] Finally, it should be noted that: the above-described embodiments are only specific embodiments of the present application, which are used to illustrate the technical solutions of the present application, but not to limit them. The protection scope of the present application is not limited thereto. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can make modifications or easily think of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed by the present application, or make equivalent replacements to some of the technical features. The modifications, changes or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of extraction of a blood vessel profile, characterized by, The extraction method comprises: obtaining a set of vessel cross-section images of a 3D vessel image to be recognized; the set of vessel cross-section images comprises a plurality of vessel cross-section images to be recognized, and each vessel cross-section image to be recognized is an image obtained according to a center line point on a vessel center line of the 3D vessel image to be recognized; inputting the set of vessel cross-section images into a pre-trained contour point recognition model to determine a first target polar coordinate point set of an inner diameter of each vessel cross-section image to be recognized in the set of vessel cross-section images and a second target polar coordinate point set of an outer diameter of each vessel cross-section image to be recognized in the set of vessel cross-section images; performing coordinate transformation processing on the first target polar coordinate point set of the inner diameter and the second target polar coordinate point set of the outer diameter in each vessel cross-section image to be recognized respectively to determine a first target rectangular coordinate point set of the inner diameter and a second target rectangular coordinate point set of the outer diameter in each vessel cross-section image to be recognized; performing curve fitting processing on the first target rectangular coordinate point set of the inner diameter and the second target rectangular coordinate point set of the outer diameter in each vessel cross-section image to be recognized in sequence to determine an inner diameter contour curve and an outer diameter contour curve of the 3D vessel image to be recognized; the contour point recognition model is constructed by the following steps: obtaining a set of sample vessel cross-section images of a plurality of sample 3D vessel images; each sample vessel cross-section image in the set of sample vessel cross-section images corresponds to a first real polar coordinate point set of an inner diameter and a second real polar coordinate point set of an outer diameter; inputting the set of sample vessel cross-section images of each sample 3D vessel image into an initial contour point recognition neural network in sequence to predict a first predicted polar coordinate point set of the inner diameter and a second predicted polar coordinate point set of the outer diameter in each sample vessel cross-section image; determining a target loss function value based on the first real polar coordinate point set, the second real polar coordinate point set, the first predicted polar coordinate point set and the second predicted polar coordinate point set of each sample vessel cross-section image; iteratively training the initial contour point recognition neural network and updating network parameters of the initial contour point recognition neural network based on the target loss function value until the target loss function value converges, stopping training, and obtaining the contour point recognition model; the first real polar coordinate point set of the inner diameter and the second real polar coordinate point set of the outer diameter of each sample vessel cross-section image in the set of sample vessel cross-section images are determined by the following steps: performing center line extraction processing on the sample 3D vessel image to determine a vessel center line of the sample 3D vessel image; the vessel center line comprises a plurality of center line points; for each center line point, a preset size plane perpendicular to a tangent direction of the vessel center line at the center line point is intercepted to determine a sample vessel cross-section image corresponding to the center line point; for each sample vessel cross-section image, based on inner and outer diameter contour annotation results on the sample vessel cross-section image, contour coordinate point extraction processing is performed to determine a first real rectangular coordinate point set of the inner diameter and a second real rectangular coordinate point set of the outer diameter in the sample vessel cross-section image. For each sample blood vessel cross-section image, the first real polar coordinate point set and the second real polar coordinate point set of the sample blood vessel cross-section image are determined based on the center line points in the sample blood vessel cross-section image.

2. The extraction method according to claim 1, characterized in that, The target loss function value is determined based on the first real polar coordinate point set, the second real polar coordinate point set, the first predicted polar coordinate point set and the second predicted polar coordinate point set of each sample blood vessel cross-section image, including: The real distance set between the inner diameter and the outer diameter of the blood vessel is determined based on the first real polar coordinate point set and the second real polar coordinate point set; The predicted distance set between the inner diameter and the outer diameter of the blood vessel is determined based on the first predicted polar coordinate point set and the second predicted polar coordinate point set; The first loss function value is determined by comparing the first predicted polar coordinate point set with the first real polar coordinate point set; The second loss function value is determined by comparing the second predicted polar coordinate point set with the second real polar coordinate point set; The third loss function value is determined by comparing the predicted distance set with the real distance set; The target loss function value is determined based on the first loss function value, the second loss function value and the third loss function value.

3. The extraction method according to claim 2, characterized in that, The target loss function value is determined based on the first loss function value, the second loss function value and the third loss function value, including: The first loss function value, the second loss function value and the third loss function value are multiplied by the respective weight coefficients, and the three loss function values multiplied by the weight coefficients are added to determine the target loss function value.

4. The extraction method of claim 1, wherein, Before sequentially performing the curve fitting processing on the first target rectangular coordinate point set of the inner diameter and the second target rectangular coordinate point set of the outer diameter of each to-be-identified blood vessel cross-section image, the extraction method further includes: The first target rectangular coordinate point set and the second target rectangular coordinate point set are respectively subjected to interpolation processing to increase the number of coordinate points in the first target rectangular coordinate point set and the second target rectangular coordinate point set.

5. The extraction method of claim 1, wherein, The contour point recognition model is a single-task recognition model or a multi-task recognition model.

6. An apparatus for extracting a blood vessel profile, characterized by comprising: The extraction device includes: An acquisition module is configured to acquire a set of blood vessel cross-section images of a to-be-identified 3D blood vessel image, wherein the set of blood vessel cross-section images includes a plurality of to-be-identified blood vessel cross-section images, and each to-be-identified blood vessel cross-section image is an image obtained by a center line point on a blood vessel center line of the to-be-identified 3D blood vessel image; A first determination module is configured to input the set of blood vessel cross-section images into a pre-trained contour point recognition model to determine a first target polar coordinate point set of an inner diameter and a second target polar coordinate point set of an outer diameter of each to-be-identified blood vessel cross-section image in the set of blood vessel cross-section images; A coordinate conversion module is configured to perform coordinate conversion processing on the first target polar coordinate point set of the inner diameter and the second target polar coordinate point set of the outer diameter of each to-be-identified blood vessel cross-section image to determine a first target rectangular coordinate point set of the inner diameter and a second target rectangular coordinate point set of the outer diameter of each to-be-identified blood vessel cross-section image. The fitting processing module is configured to sequentially perform curve fitting processing on a first target rectangular coordinate point set of the inner diameter of the blood vessel and a second target rectangular coordinate point set of the outer diameter of the blood vessel in each of the to-be-identified blood vessel cross-sectional images, and determine the inner diameter profile curve and the outer diameter profile curve of the to-be-identified 3D blood vessel image. The extraction device further comprises a model construction module, which is configured to: obtain a sample blood vessel cross-sectional image set of a plurality of sample 3D blood vessel images; wherein each sample blood vessel cross-sectional image in the sample blood vessel cross-sectional image set corresponds to a first real polar coordinate point set of the inner diameter of the blood vessel and a second real polar coordinate point set of the outer diameter of the blood vessel; sequentially input the sample blood vessel cross-sectional image set of each sample 3D blood vessel image into the initial profile point recognition neural network, and predict a first predicted polar coordinate point set of the inner diameter of the blood vessel and a second predicted polar coordinate point set of the outer diameter of the blood vessel in each sample blood vessel cross-sectional image; determine a target loss function value based on the first real polar coordinate point set, the second real polar coordinate point set, the first predicted polar coordinate point set, and the second predicted polar coordinate point set of each sample blood vessel cross-sectional image; based on the target loss function value, iteratively train the initial profile point recognition neural network and update the network parameters of the initial profile point recognition neural network until the target loss function value converges, stop training, and obtain the profile point recognition model Optionally, the extraction device further comprises a second determination module, which is configured to: perform center line extraction processing on the sample 3D blood vessel image to determine a blood vessel center line of the sample 3D blood vessel image; wherein the blood vessel center line comprises a plurality of center line points; for each center line point, intercept a preset size plane in a direction perpendicular to the tangent of the blood vessel center line including the center line point to determine a sample blood vessel cross-sectional image corresponding to the center line point; for each sample blood vessel cross-sectional image, based on the inner and outer diameter profile annotation results on the sample blood vessel cross-sectional image, perform profile coordinate point extraction processing to determine a first real rectangular coordinate point set of the inner diameter of the blood vessel and a second real rectangular coordinate point set of the outer diameter of the blood vessel in the sample blood vessel cross-sectional image; for each real polar coordinate point set in each sample blood vessel cross-sectional image, perform coordinate conversion processing based on the center line point in the sample blood vessel cross-sectional image to determine a first real polar coordinate point set of the inner diameter of the blood vessel and a second real polar coordinate point set of the outer diameter of the blood vessel in the sample blood vessel cross-sectional image.

7. An electronic device, comprising: comprise: a processor, a memory, and a bus, the memory storing machine-readable instructions executable by the processor, when the electronic device is running, the processor and the memory communicate through the bus, the machine-readable instructions are executed by the processor to perform the steps of the extraction method as claimed in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which is executed by the processor to perform the steps of the extraction method as claimed in any one of claims 1 to 5.

Images