Vascular stenosis rate prediction method, prediction model and prediction system

Abstract

Provided in the present invention are a vascular stenosis rate prediction method, prediction model and prediction system. The method comprises: acquiring an angiographic image, and performing vascular segmentation processing on the angiographic image to obtain a vascular segmentation map; and then using the vascular segmentation map as an input of a pre-trained deep learning-based vascular stenosis rate prediction model, and outputting a vascular stenosis rate distribution map after performing prediction by means of the prediction model, the vascular stenosis rate distribution map comprising a plurality of first pixels. By combining deep learning technology with clinical medicine, the prediction model, method and system accurately calculate the conditions of vascular stenosis by means of the deep learning technology, can be adapted to medical images of varying quality, have fast prediction speed and accurate and consistent prediction results, conform to diagnostic norms and standards of doctors, and solve the disadvantage of poor interpretability of deep learning; and the vascular stenosis rate prediction method has high accuracy.

Description

Methods, models, and systems for predicting vascular stenosis rates

[0001] This application claims priority to Chinese Patent Application No. 202411688251.3, filed on November 22, 2024, entitled "Method, Prediction Model and Prediction System for Predicting Vascular Stenosis Rate". Technical Field

[0002] This invention relates to the field of medical image processing technology, and in particular to a method and system for predicting vascular stenosis rate, as well as a deep learning-based vascular stenosis rate prediction model upon which it depends. Background Technology

[0003] Magnetic resonance angiography (MRA) and computed tomography angiography (CTA) are widely used in vascular stenosis identification, and various related techniques are available. Automated identification and assessment of vascular stenosis in MRA / CTA images, providing clinicians with computer-aided diagnostic methods, can improve the diagnostic accuracy of intracranial and coronary artery stenosis while reducing the workload of physicians.

[0004] Current mainstream techniques for identifying vascular stenosis include manual identification, midline-based methods, and vessel tracking methods. Manual identification involves clinicians using tools to measure and interpret MRA and / or CTA images to determine vascular stenosis. Midline-based methods for identifying vascular stenosis, such as the method proposed in the paper "Automatic Identification Method for Vascular Stenosis Based on Coronary Angiography Image Vessel Segmentation," consist of two parts: vessel segmentation and stenosis identification. In the vessel segmentation part, image enhancement is first performed using an improved Frangi Hessian-based model, followed by segmentation of the vessel region using a statistical region fusion method. In the stenosis identification part, the segmentation results are first refined using a level set algorithm to obtain the vascular skeleton, then the vessel edges are extracted for diameter measurement, and finally, the local minimum point method is used to calculate the percentage of stenosis in the entire image, locating and classifying the stenotic segments. Another example of a midline-based method is the vascular midline extraction framework proposed in the paper "DeepCenterline: A Multi-task Fully Convolutional Network for Centerline Extraction," which combines a deep learning fully convolutional network (FCN) with a minimum path extractor. FCN can simultaneously predict the minimum distance map of blood vessels and detect all vessel endpoints. The minimum distance map is a mapping of the minimum distance from points on the vessel to its boundary, where points on the line should be local maxima. Then, a minimum path algorithm is used to extract and serialize the vessel midline points from the given starting root point, list of vessel branch endpoints, and predicted minimum distance map. Finally, the vessel diameter and stenosis rate are calculated sequentially along the serialized midline to determine the location and grade of stenosis. An example of a vessel tracking method is the approach proposed in the paper "Coronary Artery Centerline Extraction in Cardiac CT Angiography Using a CNN-Based Orientation Classifier." This method involves specifying a starting position on the original image or vessel segmentation map, extracting a small region at that position, and using a deep learning model to predict the vessel radius and orientation in the current region. Therefore, the model can move a certain distance along this orientation (to reach the next location of the vessel), extract and predict, and repeat this process until the vessel reaches its terminal point.

[0005] Extensive practical experience has shown that the judgment of vascular stenosis using manual identification methods is influenced by the knowledge, experience, and subjective factors of the clinician, as well as by measurement angles and tool errors, frequently resulting in inconsistent judgments among different clinicians. The midline-based method for identifying vascular stenosis suffers from several drawbacks. Adhesion during vascular segmentation can lead to errors in midline extraction, resulting in inaccurate identification of the stenosis location. Furthermore, the slow extraction speed and the potential for the midline to be off-center due to bulges in the vascular segmentation image cause significant errors in calculating the stenosis rate, leading to low accuracy. The disordered extraction of the midline also makes it impossible to directly calculate the stenosis rate, further hindering accurate identification. Improving the recognition rate would require adding a midline serialization step, increasing computational costs. Vascular tracking methods require specifying a starting point and cannot be parallelized. This results in tracking on the original image being affected by image quality, and tracking on the vascular segmentation image being affected by vascular adhesions and abnormal bulges, thus failing to accurately identify vascular stenosis.

[0006] Based on the shortcomings of the existing technology, it is necessary to propose a technical solution that can quickly and accurately determine and identify the rate of vascular stenosis. Summary of the Invention

[0007] To address the aforementioned problems in the prior art, this invention provides a method for predicting vascular stenosis rate, along with the vascular stenosis rate prediction model and system based thereon, based on magnetic resonance angiography (MRA) and computed tomography angiography (CTA).

[0008] The first aspect of the present invention provides a method for predicting vascular stenosis rate, comprising:

[0009] Acquire angiographic images; wherein, the angiographic images include magnetic resonance angiography and / or computed tomography angiography;

[0010] The angiography images are segmented to obtain a segmented vascular map;

[0011] The blood vessel segmentation map is used as input to a pre-trained blood vessel stenosis rate prediction model, and the output is a blood vessel stenosis rate distribution map; wherein, the blood vessel stenosis rate distribution map includes multiple first pixels.

[0012] Preferably, the method for predicting vascular stenosis rate further includes:

[0013] The first vascular stenosis rate is generated by calculating the pixel value V of each first pixel in the vascular stenosis rate distribution map; where V > 0.

[0014] More preferably, the calculation process based on the pixel value V of each first pixel in the vascular stenosis rate distribution map specifically includes:

[0015] According to the preset calculation rules, the pixel value V of each first pixel is calculated and processed to obtain the vascular stenosis rate at the position corresponding to the first pixel.

[0016] Preferably, the prediction calculation rule is to perform a square root operation on the pixel value V to obtain the vascular stenosis rate at the position corresponding to the first pixel.

[0017] Preferably, the method for predicting vascular stenosis rate further includes:

[0018] Based on the pixel values ​​V of multiple first pixels, stenotic and occluded vessels are detected in the vascular stenosis rate distribution map;

[0019] Based on the correspondence between the vascular stenosis rate distribution map and the vascular segmentation map or angiography image, the stenotic and occluded vessels are mapped to the vascular segmentation map and / or angiography image to generate a visual detection vascular segmentation map and / or a visual detection angiography image.

[0020] Output the visualized blood vessel segmentation map and / or visualized blood vessel angiography image to the display device.

[0021] More preferably, the method for predicting vascular stenosis rate further includes: classifying the stenotic vessel according to the degree of stenosis based on the pixel value V of the first pixel corresponding to the stenotic vessel and a preset vascular stenosis rate classification threshold; wherein, the preset vascular stenosis rate classification threshold is multiple.

[0022] More preferably, the number of preset vascular stenosis rate grading thresholds is 3, and the vascular stenosis rate prediction method further includes:

[0023] Based on the pixel value V of the first pixel corresponding to the stenotic vessel and three preset vascular stenosis rate grading thresholds, the stenotic vessels are classified into mild stenotic vessels, moderate stenotic vessels, and severe stenotic vessels.

[0024] Different marking methods were used to mark the mildly stenotic vessels, the moderately stenotic vessels, and the severely stenotic vessels.

[0025] The second aspect of this invention provides a deep learning-based vascular stenosis rate prediction model, including a vascular segmentation map and a matching first vascular stenosis rate distribution map, and employing a 3D U-net convolutional neural network model as the backbone network. The specific method for training the 3D U-net convolutional neural network model using a training set to obtain the vascular stenosis rate prediction model includes:

[0026] Data preprocessing: The blood vessel segmentation map and the matching first blood vessel stenosis rate distribution map are subjected to sliding segmentation using a data cutting method to obtain a blood vessel segmentation sub-map and a blood vessel stenosis rate distribution sub-map with a model input size of N×N×N; where N>0 and N is an integer;

[0027] Encoding process: The vessel segmentation submap and the vessel stenosis rate distribution submap are used as inputs to the 3D U-net convolutional neural network model;

[0028] Instance normalization is performed using M×M×M 3D convolutional kernels, and correction processing is performed using modified linear units with leakage. The stride is set to 2 to achieve downsampling, and the number of feature maps doubles with the number of downsampling times. The network depth is set to 5. M > 0, and M is an integer.

[0029] Decoding process: Upsampling is performed using the nearest neighbor upsampling method, and it is concatenated at the same level as the encoding process. The convolution results of the encoding and decoding processes are then concatenated.

[0030] Output process: The model input size is N×N×N, and a linear activation function is used to generate a stenosis rate prediction map that matches the vessel segmentation map.

[0031] Preferably, N = 64 and M = 3.

[0032] Preferably, the method for generating the first vascular stenosis rate distribution map includes:

[0033] The vessel midline is extracted from the segmented vessel map to generate a vessel midline map.

[0034] The midline image of blood vessels is optimized by performing midline processing to generate an optimized midline blood vessel image;

[0035] The optimized midline vascular map is processed by midline serialization to generate a midline serialized vascular map;

[0036] The midline sequenced vascular map is subjected to simulated diameter change processing to generate a simulated diameter change vascular map; wherein, the simulated diameter change vascular map includes multiple first pixels;

[0037] The pixel values ​​of the first pixel are set to V to obtain the first vascular stenosis rate distribution map.

[0038] A third aspect of the present invention provides a vascular stenosis rate prediction system, which includes a data acquisition module for acquiring angiographic images;

[0039] The data processing module is used to perform vascular segmentation processing on angiography images to obtain vascular segmentation maps;

[0040] The prediction module includes a pre-trained vascular stenosis rate prediction model, which takes a vascular segmentation map as input to the pre-trained vascular stenosis rate prediction model and outputs a vascular stenosis rate distribution map; wherein the vascular stenosis rate distribution map includes multiple first pixels;

[0041] The display module is used to receive image data output to any of the vascular stenosis rate prediction methods in claims 1-7, and to display the image data.

[0042] A fourth aspect of the present invention provides a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the above-described method.

[0043] This invention provides a method and system for predicting vascular stenosis rate, along with a deep learning-based vascular stenosis rate prediction model. This method combines deep learning technology with clinical medicine, efficiently and accurately calculating vascular stenosis using deep learning, adaptable to medical images of varying quality. The deep learning-based vascular stenosis prediction model offers fast prediction speed, accurate and consistent results, maintaining the standardization and norms of physician diagnosis, and overcoming the drawback of poor interpretability in deep learning. The vascular stenosis rate prediction method provided by this invention, through standardized and normalized vascular stenosis detection and localization, can provide a valuable reference for the diagnosis and treatment of vascular diseases and ischemic acute stroke. Attached Figure Description

[0044] Figure 1 is a CTA image of an exemplary human brain tissue provided by the present invention;

[0045] Figure 2 is an exemplary blood vessel segmentation diagram provided by the present invention;

[0046] Figure 3 is a flowchart of a processing method for generating a first vascular stenosis rate distribution map based on a vascular segmentation map provided by the present invention;

[0047] Figure 4 is a schematic diagram of a portion of an exemplary vascular centerline diagram provided by the present invention;

[0048] Figure 5 is a schematic diagram of a portion of another exemplary vascular centerline diagram provided by the present invention;

[0049] Figure 6 is an example of the blood vessel centerline diagram processing provided by the present invention;

[0050] Figure 7 is a schematic diagram of a portion of the first vascular stenosis rate distribution map provided by the present invention;

[0051] Figure 8 is a schematic diagram of a simulated blood vessel provided by the present invention;

[0052] Figure 9 is a schematic diagram of the training process of a vascular stenosis rate prediction model provided in an embodiment of the present invention;

[0053] Figure 10 is a flowchart of a method for predicting vascular stenosis rate provided in an embodiment of the present invention;

[0054] Figure 11 is a flowchart of another method for predicting vascular stenosis rate provided by an embodiment of the present invention;

[0055] Figure 12 is a flowchart of another method for predicting vascular stenosis rate provided in an embodiment of the present invention;

[0056] Figure 13 is a flowchart of another method for predicting vascular stenosis rate provided in an embodiment of the present invention;

[0057] Figure 14 is an overall flowchart of a training model for predicting vascular stenosis rate provided in an embodiment of the present invention;

[0058] Figure 15 is an overall flowchart of a method for predicting vascular stenosis rate using the vascular stenosis rate prediction model provided by the present invention, according to an embodiment of the present invention.

[0059] Figure 16 is a system block diagram of a vascular stenosis rate prediction system provided in an embodiment of the present invention. Detailed Implementation

[0060] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0061] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0062] To more clearly introduce the method and apparatus for predicting vascular stenosis rate provided by the present invention, we will first introduce the deep learning-based vascular stenosis rate prediction model provided by the present invention on which it depends.

[0063] The vascular stenosis rate prediction model provided by this invention uses a 3D U-NET as the backbone network, which is pre-trained using the training set provided by this invention. The training set of the vascular stenosis rate prediction model provided by this invention includes: a vascular segmentation map and a first vascular stenosis rate distribution map matched thereto.

[0064] In one example of an embodiment of the present invention, Figure 1 shows an exemplary CTA image of human brain tissue provided by the present invention, and Figure 2 shows an exemplary blood vessel segmentation map provided by the present invention. As shown in Figure 1, based on the characteristics of MRA / CTA images, the bright areas appearing in the brain tissue are the cross-sectional effects of blood vessels at that level. Using traditional image processing methods or deep learning models, blood vessels are segmented or extracted, removing all non-vascular tissue and noise, resulting in a blood vessel segmentation map after 3D reconstruction, as shown in Figure 2. In Figure 2, vessels indicated by 2200 are intracranial vessels, and vessels indicated by 2100 and 2300 are extracranial vessels. In this invention, the process of processing MRA / CTA to obtain the blood vessel segmentation map is generally referred to as blood vessel segmentation processing.

[0065] In clinical practice, physicians can diagnose vascular stenosis and occlusion based on the vascular segmentation diagram shown in Figure 2. However, this requires continuously rotating the image and observing the vessel's course sequentially to identify any stenotic or occluded vessels. The presence and degree of stenosis can be determined by physicians based on experience or by using measuring tools in image reading software. Therefore, both methods introduce significant errors due to subjective factors or the measurement process, affecting the calculation of vascular stenosis rates and the classification of stenosis degrees.

[0066] In an optional embodiment provided by the present invention, a large number of vascular segmentation maps as shown in Figure 2 are used as the training set for training the vascular stenosis rate prediction model provided by the present invention. To train the vascular stenosis rate prediction model provided by the present invention, the vascular segmentation maps also need to be processed to obtain a first vascular stenosis rate distribution map as the training set.

[0067] Figure 3 is a flowchart of a processing method for generating a first vascular stenosis rate distribution map based on a vascular segmentation map provided by the present invention. As shown in Figure 3, the steps for processing the vascular segmentation map to obtain the first vascular stenosis rate distribution map include:

[0068] Step 110: Extract the blood vessel midline from the blood vessel segmentation image to generate a blood vessel midline image.

[0069] Specifically, a pre-defined vascular centerline extraction method is used to extract the midline from the vascular segmentation map to obtain a vascular centerline map. The pre-defined vascular centerline extraction method can be selected from existing vascular centerline extraction methods. For example, in this embodiment of the invention, a skeleton algorithm is selected as the pre-defined vascular centerline extraction method.

[0070] In a specific example of an embodiment of the present invention, for instance, midline extraction processing can be performed on the vascular segmentation map shown in Figure 2 provided by the present invention to obtain a vascular midline map. Figure 4 is a schematic diagram of a portion of an exemplary vascular midline map provided by the present invention. As shown in the figure, the midline extracted from the vascular segmentation map obtained from MRA / CTA images is essentially composed of points.

[0071] Step 120: Optimize the midline of the blood vessel map to generate an optimized midline blood vessel map.

[0072] Specifically, ideally, a portion of the vascular centerline map obtained through step 110 should be neatly arranged, as shown in Figure 4. However, in practice, the accuracy and stability of the vascular segmentation stage are crucial for vascular centerline extraction. Adhesions, protrusions, and other issues can severely affect the centerline extraction results. In some complex locations, the centerline may be severely misaligned or experience frequent oscillations. Therefore, simply performing vascular centerline extraction through step 110 often fails to directly achieve the effect shown in Figure 4. Instead, due to algorithm limitations, centerline oscillations and offsets (misalignment) may occur, as shown in Figures 5 and 6. Therefore, if the vascular centerline map obtained through centerline extraction (as shown in Figure 5) deviates from the calculated vascular radius and stenosis rate along the current centerline, it cannot meet the requirements for use as training data.

[0073] Therefore, to improve the midline oscillation problem in vascular midline maps, this invention employs a preset midline optimization processing method to optimize the midline of the vascular midline map, resulting in an optimized midline vascular map. In this embodiment, the preset midline optimization processing method can be selected from existing midline optimization processing methods. For example, as shown in Figure 6, after optimizing the original midline of the vascular midline map (Figure 6) using the preset midline optimization processing method, an optimized midline vascular map (Figure 6) is obtained.

[0074] Step 130: Perform midline serialization processing on the optimized midline vascular map to generate a midline serialized vascular map.

[0075] Specifically, a pre-defined midline serialization method is used to perform midline serialization processing on the optimized midline vascular map to obtain a midline serialized vascular map. The pre-defined midline serialization method is selected from existing midline serialization methods.

[0076] The purpose of midline serialization is to provide a basis for calculating the stenosis rate of blood vessels. This is because, in order to calculate the stenosis rate, it is necessary to calculate the radius of the blood vessel at each point and the ratio of that radius to the radius of the blood vessels at all midline points within a certain distance before and after it. This requires comparison with its neighboring points. Therefore, midline serialization is needed to determine which midline points are adjacent to the current point.

[0077] Step 140: Perform simulated diameter change processing on the midline sequenced vascular map to generate a simulated diameter change vascular map.

[0078] Specifically, the midline sequenced vascular map obtained after processing in steps 120 and 130 solves the problem of midline oscillation and makes the midline smoother, but it does not improve the degree of midline deviation, and even aggravates the degree of midline deviation in some local positions.

[0079] Therefore, this invention employs a preset simulated diameter-changing method to simulate diameter-changing in midline serialized vascular images, thereby obtaining simulated diameter-changing vascular images with improved skewness. The preset simulated diameter-changing method can be selected from existing simulated diameter-changing methods. For example, in this invention, strict centering is used as the criterion to simulate diameter-changing vessels in the midline serialized vascular image, resulting in a simulated diameter-changing vascular image, as shown in the rightmost schematic diagram in Figure 6. This simulated diameter-changing vascular image includes multiple first pixels.

[0080] Step 150: Set the pixel value of each first pixel in the simulated variable diameter blood vessel map to V to obtain the first blood vessel stenosis rate distribution map.

[0081] Specifically, in this invention, the midline of the simulated variable-diameter vascular map processed through steps 110-140 is serialized, and the vascular stenosis rate at each midline point can be accurately calculated. The reference radius is used for the vascular cross-sectional radius at each midline point in the simulated variable-diameter vascular map, with a pixel value of 1. The pixel values ​​of the pixels on the cross-section are then set to V. Here, V can be the stenosis rate, the square of the stenosis rate, or the cube of the stenosis rate.

[0082] In one alternative embodiment of this invention, the value of V is set to the square of the narrowing rate, which can achieve the purpose of amplifying the difference. As shown in Figure 7, the pixel value of the points on the cross-section indicated by the dashed line is 0.9.

[0083] In the embodiments provided by the present invention, after processing the blood vessel segmentation map through steps 110-150, a first blood vessel stenosis rate distribution map can be obtained.

[0084] Figure 8 is a schematic diagram of a simulated blood vessel provided by the present invention. As shown in Figure 8, it simulates the first blood vessel stenosis rate distribution map. In the present invention, the blood vessel segmentation map obtained by image segmentation and 3D reconstruction of MRA / CTA images can be obtained. After the blood vessel segmentation map is processed by steps 110-150, the matching first blood vessel stenosis rate distribution map is obtained.

[0085] After the blood vessel segmentation map is processed through steps 110-150 provided in this embodiment of the invention, the resulting data includes the blood vessel segmentation map and a matching first blood vessel stenosis rate distribution map. Multiple blood vessel segmentation maps and their matching first blood vessel stenosis rate distribution maps can constitute a training set for training the blood vessel stenosis rate prediction model provided by this invention, and can be applied to the training of the blood vessel stenosis rate prediction model provided by this invention.

[0086] The vascular stenosis rate model provided in this embodiment of the invention is a dedicated vascular stenosis rate prediction model trained based on a deep learning model. The following describes in detail the method of training the vascular stenosis rate prediction model using the training set provided by this invention.

[0087] The vascular stenosis rate prediction model provided by one embodiment of the present invention uses a 3D U-net convolutional neural network model as the backbone network, and the training set used is the vascular segmentation map described above and the first vascular stenosis rate distribution map matched with it. Figure 9 is a schematic diagram of the training process of the vascular stenosis rate prediction model provided by one embodiment of the present invention. As shown in Figure 9, the specific process of training the deep learning-based vascular stenosis rate prediction model provided by the present invention includes:

[0088] Data preprocessing process S1: The blood vessel segmentation map and the matching first blood vessel stenosis rate distribution map are subjected to sliding cutting processing using the data cutting method to obtain a blood vessel segmentation sub-map and a blood vessel stenosis rate distribution sub-map with a model input size of N×N×N; where N>0 and N is an integer.

[0089] Specifically, in the data preprocessing process provided by this invention, since the first vascular stenosis rate distribution map in the training set has undergone simulated diameter change processing in step 140, its size is relatively large. Therefore, a data segmentation method is used to slide and segment the vascular segmentation map and the matching first vascular stenosis rate distribution map into vascular segmentation sub-maps and vascular stenosis rate distribution sub-maps with a model input size of N×N×N for model training. Wherein, N > 0, and N is an integer.

[0090] In one optional embodiment of the present invention, N=64, and the vascular segmentation map and the matching first vascular stenosis rate distribution map are slidably cut into vascular segmentation sub-maps and vascular stenosis rate distribution sub-maps with a model input size of 64×64×64 for model training.

[0091] Encoding process S2: The vessel segmentation sub-map and the vessel stenosis rate distribution sub-map are used as inputs to the 3D U-net convolutional neural network model. Instance normalization is performed using M×M×M 3D convolutional kernels, and correction processing is performed using modified linear units with leakage.

[0092] Specifically, in the encoding process provided by this invention, a 3D U-net convolutional neural network model is selected as the backbone network. The N×N×N vessel segmentation sub-maps and vessel stenosis rate distribution sub-maps obtained during data preprocessing are used as inputs to the 3D U-net convolutional neural network model. M×M×M 3D convolutional kernels are used for strength normalization, and leaky corrected linear units are used for correction. The stride is set to 2 for downsampling, and the number of feature maps doubles with each downsampling iteration. The network depth is set to 5; M > 0, and M is an integer. In the scheme provided by this invention, the leaky corrected linear unit can be selected from existing leaky corrected linear methods, such as the LeakyReLU correction function.

[0093] In a preferred embodiment of the present invention, M=3, meaning that a 3×3×3 3D convolutional kernel is used in the encoding stage for instance normalization, and LeakyReLU with leakage correction is used for correction. The stride is set to 2 to achieve downsampling, and the number of feature maps doubles with the number of downsampling iterations. The network depth is set to 5. In this embodiment of the present invention, excessive depth leads to an exponential increase in parameters, resulting in excessive computational demands. Therefore, in this embodiment of the present invention, the selection of each parameter fully considers computational speed and computational load, and is determined after balancing the corresponding training set.

[0094] In the embodiments provided by this invention, the core purpose of instance normalization is to adjust the pixel values ​​of image data to a fixed range, typically from 0 to 1, in order to improve the stability and accuracy of image processing algorithms. Instance normalization can eliminate differences in image intensity values ​​under different devices or acquisition conditions, accelerate model training, and improve model training stability.

[0095] In this embodiment of the invention, a method for instance normalization is selected from the prior art. In one optional example provided by this embodiment, the core network used is the 3D U-net convolutional neural network model. The specific steps and implementation methods for instance normalization include:

[0096] Generate random 3D image data: Use the NumPy library to generate random 3D image data.

[0097] Define a normalization function: convert 3D image data to float type and then normalize it to the range [0,1].

[0098] Applying a normalization function: Applying a normalization function to the generated 3D image data to obtain normalized image data.

[0099] Decoding process S3: The nearest neighbor upsampling method is used and concatenated with the encoding process at the same level to concatenate the convolution results of the encoding and decoding processes.

[0100] Specifically, the upsampling in the decoding process provided in this embodiment of the invention adopts the nearest neighbor upsampling method and is concatenated at the same level as the encoding stage, and the convolution results of the two are concatenated to perform region property analysis at multiple resolutions.

[0101] The upsampling process in the decoding process of this invention adopts the nearest neighbor upsampling method, which does not require training, effectively reducing the training parameters of the network and having better robustness and speed.

[0102] Output process S4: The input size of the restored model is N×N×N, and a linear activation function is used to generate a vascular stenosis rate prediction map that matches the vascular segmentation map.

[0103] Specifically, in the output process provided by this embodiment of the invention, the output layer is restored to the model input size N×N×N, and a final vascular stenosis rate prediction map is generated using a Linear activation function. Here, N is selected to be the same size as in the data preprocessing step. In an optional embodiment of the invention, N = 64 in the data preprocessing step; therefore, in the output process, the output layer is restored to the model input size of 64×64×64.

[0104] The foregoing described a deep learning-based vascular stenosis rate prediction model provided by the embodiments of the present invention. Next, based on this vascular stenosis rate prediction model, we will describe in detail the vascular stenosis rate prediction method provided by the present invention.

[0105] Figure 10 is a flowchart of a method for predicting vascular stenosis rate provided by an embodiment of the present invention. As shown in Figure 10, the method specifically includes the following steps:

[0106] Step 210: Obtain angiographic images.

[0107] Specifically, the angiographic images acquired by this invention include magnetic resonance angiography (MRA) and computed tomography angiography (CTA). The type of angiographic image acquired depends on the source of the image data.

[0108] Step 220: Perform vascular segmentation processing on the angiography image to obtain a vascular segmentation map.

[0109] Specifically, a preset vascular segmentation processing technique is used to segment the MRA or CTA images obtained in step 210 to obtain a vascular segmentation map. The preset vascular segmentation processing method can be a traditional image processing method or a deep learning model, which achieves vascular segmentation or extraction, removes all non-vascular tissues and noise, and obtains a vascular segmentation map after 3D vascular reconstruction.

[0110] Step 230: Use the blood vessel segmentation map as input to the pre-trained blood vessel stenosis rate prediction model and output the blood vessel stenosis rate distribution map.

[0111] Specifically, the blood vessel segmentation map is input into a deep learning-based blood vessel stenosis rate prediction model provided by this invention. The output after processing by the blood vessel stenosis rate prediction model is the blood vessel stenosis rate distribution map of this invention. The blood vessel stenosis rate distribution map includes multiple first pixels.

[0112] In a preferred embodiment of this solution, the vascular stenosis rate distribution map is also output to a display device connected to the system for display, and / or output to a printing device connected to the system for printing. The purpose of displaying and / or printing is to facilitate viewing and analysis by physicians or other relevant personnel.

[0113] In an optional example of the present invention, as shown in FIG7, in a schematic diagram of a portion of a vascular stenosis rate distribution map provided by an embodiment of the present invention, each of the plurality of first pixels in the vascular stenosis rate distribution map has its own pixel value V, which can reflect the vascular stenosis rate. The value of V is greater than 0. When the first pixel value V is equal to 0, it means that the location of the first pixel is the background, rather than inside the blood vessel.

[0114] Figure 11 is a flowchart of another method for predicting vascular stenosis rate provided by an embodiment of the present invention. As shown in Figure 11, in a preferred embodiment of the present invention, the method for predicting vascular stenosis rate provided by the present invention further includes:

[0115] Step 240: Calculate and process the pixel value V of each first pixel in the vascular stenosis rate distribution map to generate the first vascular stenosis rate at the position corresponding to the first pixel.

[0116] Specifically, during the training of the vascular stenosis rate prediction model provided in this embodiment of the invention, the first vascular stenosis rate distribution map in the training set is generated. In step 150, each pixel value of the first pixel is set to V, and V can be the stenosis rate, the square of the stenosis rate, or the cube of the stenosis rate. Therefore, the preset calculation rule here should be the inverse operation of the assignment in step 150. For example, if the V assigned to the first pixel in step 150 is the square of the vascular stenosis rate, then the preset calculation rule here is to take the square root and take a positive value; if the V assigned to the first pixel in step 150 is the cube of the vascular stenosis rate, then the preset calculation rule here is to take the cube root and take a positive value.

[0117] In the optional solution provided by the embodiments of the present invention, the pixel value V of each first pixel is calculated and processed according to a preset calculation rule to obtain the vascular stenosis rate at the position corresponding to the first pixel.

[0118] In an optional embodiment of the present invention, the prediction calculation rule is to perform a square root operation on the pixel value V to obtain the vascular stenosis rate at the position corresponding to the first pixel.

[0119] In a preferred embodiment of the present invention, the vascular stenosis rate is displayed according to a preset output rule. For example, it is displayed as a curve graph, list, or other similar format according to a certain arrangement rule. Furthermore, in a further embodiment, the values ​​in the list can correspond to each first pixel in the vascular stenosis rate distribution map, so that when a value in the list is selected, its corresponding first pixel is highlighted in the vascular stenosis rate distribution map.

[0120] Figure 12 is a flowchart of another method for predicting vascular stenosis rate provided by an embodiment of the present invention. As shown in Figure 12, in a preferred embodiment of the present invention, the method for predicting vascular stenosis rate further includes the following steps:

[0121] Step 250: Detect stenotic and occluded vessels in the vascular stenosis rate distribution map based on the pixel values ​​V of multiple first pixels.

[0122] Specifically, in this embodiment of the invention, the method is to determine whether the range of V value is within a first preset range. For example, if the V value is greater than or equal to 0.5 and less than or equal to 0.9, the blood vessel at the position corresponding to the first pixel is determined to be a narrow blood vessel; if the V value is greater than 0.9 and less than 1, the blood vessel at the position corresponding to the first pixel is determined to be an occluded blood vessel.

[0123] Step 260: According to the correspondence between the vascular stenosis rate distribution map and the vascular segmentation map or angiography image, the stenotic and occluded vessels are mapped to the vascular segmentation map and / or angiography image to generate a visual detection vascular segmentation map and / or a visual detection angiography image.

[0124] Specifically, according to the vascular stenosis rate prediction model provided by this invention, the input is a vascular segmentation map, and the output is a vascular stenosis rate prediction model corresponding to the vascular segmentation map. The vascular segmentation map is obtained by segmenting angiographic images, and there is a corresponding relationship between the vascular segmentation map and the angiographic images. Therefore, the stenotic and occluded vessels in the vascular stenosis rate distribution map output by this invention can be mapped to the vascular segmentation map and / or angiographic images according to the corresponding relationship, thereby obtaining a visualized detection vascular segmentation map and / or a visualized detection angiographic image. Furthermore, the stenotic and occluded vessels are displayed by using different display marking methods to facilitate intuitive observation.

[0125] Step 270: Output the visualized blood vessel segmentation map and / or the visualized blood vessel angiography image to the display device.

[0126] Specifically, the visualization of blood vessel segmentation and / or visualization of angiography images are displayed through a display device that communicates with the processor.

[0127] In an optional embodiment of the present invention, the visualized detection vessel segmentation map and / or visualized detection angiography image can also be printed by a printing device that is communicatively connected to the processor.

[0128] In a specific example of this invention, stenosis and occlusion can be directly detected using a deep learning detection model. A series of bounding boxes are predicted on the original image or vessel segmentation map, each representing a detected stenotic or occluded vessel. This original image with added bounding boxes becomes the final visual detection angiography image output by this invention, and the vessel segmentation map with added bounding boxes becomes the final visual detection vessel segmentation map output by this invention.

[0129] This invention also provides another method for predicting vascular stenosis rate. This method can classify the degree of vascular stenosis according to the cardiovascular stenosis rate value. Figure 13 is a flowchart of another method for predicting vascular stenosis rate provided by this invention. As shown in Figure 13, after completing step 250, the method further includes:

[0130] Step 2501: Classify the stenotic blood vessels according to the degree of stenosis based on the pixel value V of the first pixel corresponding to the stenotic blood vessel and the preset stenosis rate classification threshold.

[0131] Specifically, in the solution provided by the embodiments of the present invention, the narrow blood vessels are classified according to the pixel value V of the first pixel corresponding to the narrow blood vessels and the preset blood vessel stenosis rate classification threshold according to the degree of stenosis. That is to say, according to the size of the pixel value V of each first pixel in the blood vessel stenosis rate distribution map, the degree of blood vessel stenosis is determined. For example, the narrow blood vessels are divided into mildly stenosed blood vessels, moderately stenosed blood vessels, and severely stenosed blood vessels.

[0132] In an optional solution provided by the embodiments of the present invention, the number of preset blood vessel stenosis rate classification thresholds is 3. For example: the first preset blood vessel stenosis rate classification threshold V1, the second preset blood vessel stenosis rate classification threshold V2, and the third preset blood vessel stenosis rate classification threshold V3. Among them, V1>0, V2>V1, and V3>V2.

[0133] Classifying the narrow blood vessels according to the pixel value V of the first pixel corresponding to the narrow blood vessels and the preset blood vessel stenosis rate classification threshold according to the degree of stenosis specifically includes judging the size of the pixel value V of each first pixel to determine the degree of blood vessel stenosis. Such as:

[0134] When V<V1, it is determined that the narrow blood vessel at the position corresponding to the first pixel is a mildly stenosed blood vessel.

[0135] When V≥V1 and V<V2, it is determined that the narrow blood vessel at the position corresponding to the first pixel is a moderately stenosed blood vessel.

[0136] When V≥V2 and <V3, it is determined that the narrow blood vessel at the position corresponding to the first pixel is a severely stenosed blood vessel.

[0137] In the embodiments provided by the present invention, V1, V2, and V3 are preset values in the system. For example: the values of V1, V2, and V3 are 30%, 50%, and 70% respectively. In this way, the narrow blood vessels detected in step 250 can be divided into mildly stenosed blood vessels, moderately stenosed blood vessels, and severely stenosed blood vessels according to the size of the V value.

[0138] Furthermore, in the vascular stenosis rate prediction method provided in this embodiment, the number of preset vascular stenosis rate grading thresholds can be increased. The vascular stenosis rate can be further divided into more levels according to the pixel value V of each first pixel in the vascular stenosis rate distribution map. For example, four preset vascular stenosis rate grading thresholds can be set: a first preset vascular stenosis rate grading threshold V1, a second preset vascular stenosis rate grading threshold V2, a third preset vascular stenosis rate grading threshold V3, and a fourth preset vascular stenosis rate grading threshold V4, with their values ​​set to presets of 50%, 75%, 90%, and 95%, respectively. Therefore, stenotic vessels can be divided into four different levels according to the degree of stenosis, and the detected stenotic vessels can be classified as mild stenosis vessels, moderate stenosis vessels, severe stenosis vessels, and extremely severe stenosis vessels. This invention achieves the vascular stenosis level classification based on the pixel value V of each first pixel in the vascular stenosis rate distribution map.

[0139] In this embodiment of the invention, detected stenotic vessels are classified according to the degree of stenosis, and different marking methods are used to mark vessels of different grades. Optional marking methods include using different colors to mark the classified stenotic and occluded vessels, which can more intuitively display the condition of the vessels for diagnostic reference. For example, green lines are used to mark mildly stenotic vessels, yellow lines to mark moderately stenotic vessels, orange lines to mark severely stenotic vessels, purple lines to mark extremely stenotic vessels, and dark red lines to mark occluded vessels, etc. This example in the invention is merely to illustrate the use of different markings to represent vessels of different stenosis grades, and does not limit the marking method to only color lines; other methods can also be used to mark vessels of different stenosis grades.

[0140] In a further example of the vascular stenosis rate prediction method provided in this embodiment of the invention, the method further includes: according to the correspondence between the vascular stenosis rate distribution map and the vascular segmentation map or angiography image, mapping mildly stenotic vessels, moderately stenotic vessels, severely stenotic vessels, and occluded vessels to the vascular segmentation map or angiography image, generating a visualized stenosis grading vascular segmentation map and / or a visualized stenosis grading angiography image; and then outputting the visualized stenosis grading vascular segmentation map and / or the visualized stenosis grading angiography image to a display device. The principle of mapping mildly stenotic vessels, moderately stenotic vessels, severely stenotic vessels, and occluded vessels to the vascular segmentation map or angiography image and the method of display output are the same as steps 260 and 270, and will not be repeated here.

[0141] The vascular stenosis rate prediction model and method provided by this invention have been described. Figure 14 is a flowchart illustrating the overall process of training the vascular stenosis rate prediction model according to an embodiment of this invention. Referring to Figure 14 can help to more intuitively understand the training method of the vascular stenosis rate prediction model of this invention. Figure 15 is a flowchart illustrating the overall process of implementing the vascular stenosis rate prediction method using the vascular stenosis rate prediction model provided by this invention according to an embodiment of this invention. Referring to Figure 15 can help to more intuitively understand the implementation of the vascular stenosis rate prediction method provided by this invention using the vascular stenosis rate prediction model provided by this invention.

[0142] The following is a detailed description of a vascular stenosis rate prediction system provided by an embodiment of the present invention. Figure 16 is a system block diagram of a vascular stenosis rate prediction system provided by an embodiment of the present invention. As shown in the figure, the vascular stenosis rate device 1000 includes: a data acquisition module 1501, a data processing module 1502, a prediction module 1503, and a display module 1504.

[0143] In one embodiment of the present invention, the data acquisition module 1501, the data processing module 1502, the prediction module 1503 and the display module 1504 are connected to each other through wired or wireless communication methods, and can transmit data to each other.

[0144] The data acquisition module 1501 is used to acquire angiographic images, including magnetic resonance angiography and / or computed tomography angiography. The specific type of angiographic image acquired depends on the data source.

[0145] The data processing module 1502 is used to segment or extract blood vessels based on the image characteristics of MRA / CTA and by using traditional image processing methods or deep learning models. It removes all non-vascular tissues and noise, and realizes the blood vessel segmentation map after 3D reconstruction of blood vessels.

[0146] The prediction module 1503 contains a pre-trained vascular stenosis rate prediction model. This model is a deep learning-based vascular stenosis rate prediction model provided in this embodiment of the invention. The prediction module uses the vascular segmentation map obtained after processing by the data processing module 1502 as input to the pre-trained vascular stenosis rate prediction model. After computation by the vascular stenosis rate prediction model, it directly outputs a vascular stenosis rate distribution map. The vascular stenosis rate distribution map includes multiple first pixels.

[0147] The display module 1504 is used to receive all image data, vascular stenosis rate data, and vascular markers output to the vascular stenosis prediction method provided by the present invention, and to display the received data. The data received by the display module can come from one or more of the data acquisition module 1501, the data processing module 1502, and the prediction module 1503.

[0148] This invention provides a method and system for predicting vascular stenosis rate, along with a deep learning-based vascular stenosis rate prediction model. This method combines deep learning technology with clinical medicine, efficiently and accurately calculating vascular stenosis using deep learning, adaptable to medical images of varying quality. The deep learning-based vascular stenosis prediction model offers fast prediction speed, accurate and consistent results, maintaining the standardization and norms of physician diagnosis, and overcoming the drawback of poor interpretability in deep learning. The vascular stenosis rate prediction method provided by this invention, through standardized and normalized vascular stenosis detection and localization, can provide a valuable reference for the diagnosis and treatment of vascular diseases and ischemic acute stroke.

[0149] In some possible implementations, the vascular stenosis rate prediction method and system provided by the present invention, and the deep learning-based vascular stenosis rate prediction model upon which they depend, can be implemented by a processing component calling computer-readable instructions stored in memory. In one example, the processing component includes, but is not limited to, a separate processor, discrete components, or a combination of processors and discrete components. The processor may include a controller in an electronic device with the function of executing instructions. The processor may be implemented in any suitable manner, for example, by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components. Within the processor, executable instructions can be executed by hardware circuits such as logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers.

[0150] The use of "first" and "second" in the following text is merely for distinction and has no other meaning.

[0151] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0152] This disclosure also proposes a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the methods described above. The computer-readable storage medium can be a non-volatile computer-readable storage medium. It can also be a tangible device capable of holding and storing instructions used by an instruction execution device. For example, a computer-readable storage medium can be (but is not limited to) an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusion structures storing instructions thereon, and any suitable combinations thereof. The computer-readable storage medium used herein is not to be interpreted as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0153] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages.

[0154] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network (LAN), wide area network (WAN), and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to computer-readable storage media in the respective computing / processing device. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuits, such as programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), are personalized by utilizing state information of computer-readable program instructions. These electronic circuits can execute computer-readable program instructions to implement various aspects of this disclosure.

[0155] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0156] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0157] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0158] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0159] The computer program product can be implemented specifically through hardware, software, or a combination thereof. In one alternative embodiment, the computer program product is specifically embodied in a computer storage medium; in another alternative embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0160] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention. The terminology used herein is chosen to best explain the principles, practical applications, or improvements to the technology in the market, or to enable other those skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for predicting vascular stenosis rate, characterized in that, Methods for predicting vascular stenosis rate include: Acquire angiographic images; wherein the angiographic images include magnetic resonance angiography and / or computed tomography angiography; The angiography images are segmented to obtain a segmented vascular map; The blood vessel segmentation map is used as input to a pre-trained blood vessel stenosis rate prediction model, and the output is a blood vessel stenosis rate distribution map; wherein, the blood vessel stenosis rate distribution map includes multiple first pixels.

2. The method for predicting vascular stenosis rate according to claim 1, characterized in that, The method for predicting vascular stenosis rate also includes: The first vascular stenosis rate is generated by calculating the pixel value V of each first pixel in the vascular stenosis rate distribution map; where V > 0.

3. The method for predicting vascular stenosis rate according to claim 2, characterized in that, The specific calculation process based on the pixel value V of each first pixel in the vascular stenosis rate distribution map is as follows: According to the preset calculation rules, the pixel value V of each first pixel is calculated and processed to obtain the vascular stenosis rate at the position corresponding to the first pixel.

4. The method for predicting vascular stenosis rate according to claim 1, characterized in that, The prediction calculation rule is to perform a square root operation on the pixel value V to obtain the vascular stenosis rate at the position corresponding to the first pixel.

5. The method for predicting vascular stenosis rate according to claim 1, characterized in that, The method for predicting vascular stenosis rate also includes: Based on the pixel values ​​V of multiple first pixels, stenotic and occluded vessels are detected in the vascular stenosis rate distribution map; Based on the correspondence between the vascular stenosis rate distribution map and the vascular segmentation map or angiography image, the stenotic and occluded vessels are mapped to the vascular segmentation map and / or angiography image to generate a visual detection vascular segmentation map and / or a visual detection angiography image. Output the visualized blood vessel segmentation map and / or visualized blood vessel angiography image to the display device.

6. The method for predicting vascular stenosis rate according to claim 5, characterized in that, The method for predicting vascular stenosis rate further includes: classifying the stenotic vessels according to the degree of stenosis based on the pixel value V of the first pixel corresponding to the stenotic vessel and a preset vascular stenosis rate classification threshold; wherein, there are multiple preset vascular stenosis rate classification thresholds.

7. The method for predicting vascular stenosis rate according to claim 6, characterized in that, The number of preset vascular stenosis rate grading thresholds is 3, and the vascular stenosis rate prediction method further includes: Based on the pixel value V of the first pixel corresponding to the stenotic vessel and three preset vascular stenosis rate grading thresholds, the stenotic vessels are classified into mild stenotic vessels, moderate stenotic vessels, and severe stenotic vessels. Different marking methods were used to mark the mildly stenotic vessels, the moderately stenotic vessels, and the severely stenotic vessels.

8. A model for predicting vascular stenosis rate, characterized in that, The training set for the vascular stenosis rate prediction model includes a vascular segmentation map and a matching first vascular stenosis rate distribution map. A 3D U-net convolutional neural network model is used as the backbone network. The specific method for training the 3D U-net convolutional neural network model using the training set to obtain the vascular stenosis rate prediction model includes: Data preprocessing: The blood vessel segmentation map and the matching first blood vessel stenosis rate distribution map are subjected to sliding segmentation using a data cutting method to obtain a blood vessel segmentation sub-map and a blood vessel stenosis rate distribution sub-map with a model input size of N×N×N; where N>0 and N is an integer; Encoding process: The vessel segmentation submap and the vessel stenosis rate distribution submap are used as inputs to the 3D U-net convolutional neural network model; Instance normalization is performed using M×M×M 3D convolutional kernels, and correction processing is performed using modified linear units with leakage. The stride is set to 2 to achieve downsampling, and the number of feature maps doubles with the number of downsampling times. The network depth is set to 5. M > 0, and M is an integer. Decoding process: Upsampling is performed using the nearest neighbor upsampling method, and it is concatenated at the same level as the encoding process. The convolution results of the encoding and decoding processes are then concatenated. Output process: The model input size is N×N×N, and a linear activation function is used to generate a stenosis rate prediction map that matches the vessel segmentation map.

9. The vascular stenosis rate prediction model according to claim 8, characterized in that, Methods for generating the first vessel stenosis rate distribution map include: The vessel midline is extracted from the segmented vessel map to generate a vessel midline map. The midline image of blood vessels is optimized by performing midline processing to generate an optimized midline blood vessel image; The optimized midline vascular map is processed by midline serialization to generate a midline serialized vascular map; The midline sequenced vascular map is subjected to simulated diameter change processing to generate a simulated diameter change vascular map; wherein, the simulated diameter change vascular map includes multiple first pixels; The pixel values ​​of the first pixel are set to V to obtain the first vascular stenosis rate distribution map.

10. A system for predicting vascular stenosis rate, characterized in that, The device includes: The data acquisition module is used to acquire angiographic images; The data processing module is used to perform vascular segmentation processing on angiography images to obtain vascular segmentation maps; The prediction module includes a pre-trained vascular stenosis rate prediction model, which takes a vascular segmentation map as input to the pre-trained vascular stenosis rate prediction model and outputs a vascular stenosis rate distribution map; wherein the vascular stenosis rate distribution map includes multiple first pixels; The display module is used to receive image data output to any of the vascular stenosis rate prediction methods in claims 1-7, and to display the image data.

Images