Intravascular ultrasound image processing method based on paired contour coding, computer device and storage medium

Abstract

The application discloses an intravascular ultrasound image processing method based on paired contour coding, a computer device and a storage medium, and comprises the following steps: contouring an intravascular ultrasound image to obtain intimal contour sampling points and external elastic membrane contour resampling points; pairing and stacking the intimal contour sampling points and the external elastic membrane contour resampling points to obtain a paired contour sampling point matrix; using a regression network to obtain predicted contour features; decoding the predicted contour features to obtain predicted paired contour points; determining an error loss; and training the regression network according to the error loss. The application can make the generated contour have a reasonable anatomical shape and topological relationship by pairing and coding the intimal and external elastic membrane contours and directly predicting the position of the target contour through a convolutional neural network, so that the automatic contouring of the intimal and external elastic membrane contours in the intravascular ultrasound image can be performed. The application is widely applied in the technical field of image processing.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method, computer device, and storage medium for intravascular ultrasound image processing based on paired contour coding. Background Technology

[0002] Coronary atherosclerotic heart disease (CAD) is the most common cardiovascular disease. It is usually caused by atherosclerosis of the coronary arteries, leading to narrowing and blockage of the blood vessel lumen, reduced vascular elasticity, and insufficient blood and oxygen supply to the heart. It is characterized by high morbidity, high disability rate, and high mortality rate. Accurate diagnosis and quantitative analysis of CAD are crucial for developing treatment plans. Among various imaging techniques, coronary angiography is considered the gold standard for CAD diagnosis. It can visualize the contour of the blood vessel lumen, helping doctors determine whether stenosis exists, and the location and degree of stenosis. However, coronary angiography only displays a two-dimensional image of the lumen contour and cannot directly provide information on the lesions in the vessel wall or related clinical measurement parameters. Therefore, intravascular ultrasound has become an important auxiliary tool for the diagnosis and treatment of CAD. Intravascular ultrasound imaging sequences can provide a 360-degree tomographic cross-sectional view of the vessel segment, allowing doctors to determine the boundary between the intima and the external elastic lamina, thereby enabling quantitative analysis of the blood vessel and guiding percutaneous coronary intervention. However, an intravascular ultrasound sequence typically contains thousands of images, and manually delineating the boundaries of the intima and outer elastic lamina is extremely time-consuming and laborious. To improve the efficiency of diagnosis and treatment, a fast and accurate automated delineation method is essential. At the same time, intravascular ultrasound images contain various image artifacts and noise, as well as other intravascular tissues and anatomical structures. Therefore, achieving a method for automated delineation that maintains a certain level of accuracy while also possessing greater robustness remains a very challenging problem.

[0003] Traditional automatic contouring methods typically employ algorithms based on image features, probability distributions, or active contours. These algorithms rely on the design of contour-related features and energy functions, often requiring multiple iterations to obtain the target contour. Given the various artifacts and noise present in intravascular ultrasound images, the feature and energy function designs of these algorithms are often insufficient, and the repeated iterations still result in significant time consumption. In recent years, due to the development of deep learning, many models based on fully convolutional networks have been proposed and used for segmentation of intravascular tissue structures. These models utilize convolutional neural networks to directly extract depth features from the input intravascular ultrasound images and perform pixel-level classification to divide different structural regions within the blood vessel. However, these models do not consider the anatomical structure of the blood vessel itself, leading to discrepancies between the segmented contours of the intima and external elastic membrane and the actual shapes and topologies in clinical settings. Even with additional post-processing steps, these segmentation errors cannot be completely eliminated. Therefore, the key to deploying this type of research in clinical applications is to quickly and accurately delineate the contours of the intima and external elastic membrane from intravascular ultrasound images, and to ensure that the contours have reasonable anatomical shapes and topological relationships. This will help doctors to diagnose and treat patients in a timely and effective manner.

[0004] Terminology Explanation:

[0005] 1. MBConv (Mobile inverted bottlenect convolution): Mobile inverted bottlenect convolution;

[0006] 2. DWConv (Depth-wise separable convolution): Depthwise separable convolution;

[0007] 3. SE (Squeeze-and-Excitation): Squeeze-and-Excitation module;

[0008] 4. BN (Batch normalization): Batch normalization;

[0009] 5. HD (Hausdorff Distance): Hausdorff distance;

[0010] 6. DSC (Dice similarity coefficient): Dice similarity coefficient;

[0011] 7. JI (Jaccard Index): Jaccard coefficient. Summary of the Invention

[0012] While current related technologies can achieve rapid and automatic delineation of the intima and external elastic membrane, they neglect the anatomical shape of the vascular structure and the topological relationship between the structures, resulting in unreasonable anatomical shape and topological relationship of the automatically delineated intima and external elastic membrane contours. The purpose of this invention is to provide a method, computer device, and storage medium for intravascular ultrasound image processing based on paired contour coding.

[0013] On one hand, embodiments of the present invention include an intravascular ultrasound image delineation system based on paired contour coding, comprising:

[0014] Acquire intravascular ultrasound image data; delineate the intravascular ultrasound images to determine the intima contour and the outer elastic membrane contour;

[0015] The inner membrane contour is sampled to obtain inner membrane contour sampling points;

[0016] The external elastic membrane profile is resampled to obtain the resampled points of the external elastic membrane profile.

[0017] The inner membrane contour sampling points are paired and stacked with the outer elastic membrane contour resampling points to obtain a paired contour sampling point matrix.

[0018] Principal component analysis is performed on the paired contour sampling point matrix to obtain the decoding matrix; the intravascular ultrasound image of the sample and the corresponding paired contour sampling points are obtained.

[0019] Based on the intravascular ultrasound images of the samples, a regression network is used to obtain the predicted contour features;

[0020] The predicted contour features are decoded using a decoding matrix to obtain the predicted paired contour points.

[0021] The error loss is determined based on the predicted paired contour points;

[0022] The regression network is trained based on the error loss.

[0023] Further, sampling the intima contour to obtain intima contour sampling points includes:

[0024] Determine the rightmost intersection point of the intimal contour and the center horizontal extension line of the intravascular ultrasound image of the sample.

[0025] Starting from the rightmost intersection point, the inner membrane contour is sampled with equal arc length to obtain the inner membrane contour sampling points.

[0026] Further, the step of resampling the external elastic membrane profile to obtain resampling points of the external elastic membrane profile includes:

[0027] Determine the rightmost intersection point between the outline of the external elastic membrane and the center horizontal extension line of the intravascular ultrasound image of the sample.

[0028] Starting from the rightmost intersection point, the outer elastic membrane contour is sampled with equal arc length to obtain the sampling points of the outer elastic membrane contour;

[0029] The optimal arc length offset is determined based on the inner membrane contour sampling points and the outer elastic membrane contour sampling points.

[0030] The outer elastic membrane profile is resampled using the optimal arc length offset to obtain the resampled points of the outer elastic membrane profile.

[0031] Further, determining the optimal arc length offset based on the inner membrane contour sampling points and the outer elastic membrane contour sampling points includes:

[0032] The Δs that minimizes C(Δs) is determined as the optimal arc length offset; where,

[0033]

[0034] N is the number of sampling points, x L (s i ) and y L (s i ) represents the i-th arc length segment on the inner membrane contour sampling point, x M (s i ) and y M (s i ) represents the i-th arc length segment on the sampling point of the external elastic membrane profile, and λ is a coefficient.

[0035] Further, the step of performing principal component analysis on the paired contour sampling point matrix to obtain contour features includes:

[0036] Use formula Processing is performed; where b is the contour feature, x is the paired contour sampling point matrix, and P is the feature vector matrix in the paired contour sampling point matrix x. x is the arithmetic mean of the paired contour sampling point matrix.

[0037] Further, determining the error loss based on the predicted paired contour points includes:

[0038] According to the formula Determine the error loss; where, l c The error loss is defined as S, which represents the set consisting of the inner membrane profile and the outer elastic membrane profile, and s represents a point in set S. x represents the component corresponding to s in the predicted paired contour points. s This represents the component corresponding to s in the paired contour sampling point matrix.

[0039] Furthermore, the intravascular ultrasound image processing method based on paired contour coding also includes:

[0040] Acquire intravascular ultrasound images to be processed;

[0041] The intravascular ultrasound image to be processed is input into the regression network for processing;

[0042] Obtain the delineation results output by the regression network.

[0043] Furthermore, the intravascular ultrasound image processing method based on paired contour coding also includes:

[0044] Based on the intravascular ultrasound image to be processed and the delineation results, the anatomical rationality index value is determined;

[0045] The performance of the regression network is evaluated based on the anatomical rationality index value.

[0046] On the other hand, embodiments of the present invention also include a computer device including a memory and a processor, the memory being used to store at least one program, and the processor being used to load the at least one program to execute the intravascular ultrasound image processing method based on paired contour coding in the embodiments.

[0047] On the other hand, embodiments of the present invention also include a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to perform the intravascular ultrasound image processing method based on paired contour coding in the embodiments.

[0048] The beneficial effects of the present invention are as follows: The intravascular ultrasound image processing method based on paired contour encoding in the embodiments can train a regression network. The trained regression network can directly predict the position of the target contour by pairing and encoding the contours of the intima and the external elastic membrane. It can achieve the segmentation performance of a fully convolutional network model, while also enabling the generated contours to have reasonable anatomical shapes and topological relationships, thereby more accurately delineating the contours of the intima and the external elastic membrane in intravascular ultrasound images. Attached Figure Description

[0049] Figure 1 This is a flowchart of the intravascular ultrasound image processing method based on paired contour coding in the embodiments;

[0050] Figure 2This is a schematic diagram of the intravascular ultrasound image processing method based on paired contour coding in the embodiment.

[0051] Figure 3 This is a schematic diagram illustrating the structure and principle of the regression network used in the examples. Detailed Implementation

[0052] In this embodiment, refer to Figure 1 The intravascular ultrasound image processing method based on paired contour coding includes the following steps:

[0053] S1. Acquire intravascular ultrasound image data;

[0054] S2. Draw out the intravascular ultrasound image to determine the contours of the intima and the external elastic membrane;

[0055] S3. Sample the intima contour to obtain intima contour sampling points;

[0056] S4. Resample the external elastic membrane profile to obtain the resampled points of the external elastic membrane profile;

[0057] S5. Pair and stack the inner membrane contour sampling points with the outer elastic membrane contour resampling points to obtain a paired contour sampling point matrix;

[0058] S6. Perform principal component analysis on the paired contour sampling point matrix to obtain the decoding matrix;

[0059] S7. Obtain the intravascular ultrasound image of the sample and the corresponding paired contour sampling points;

[0060] S8. Based on the intravascular ultrasound images of the samples, a regression network is used to obtain the predicted contour features;

[0061] S9. Use the decoding matrix to deencode the predicted contour features to obtain the predicted paired contour points;

[0062] S10. Determine the error loss based on the predicted paired contour points;

[0063] S11. Train the regression network based on the error loss.

[0064] After completing steps S1-S11 to train the regression network, the following steps can be performed:

[0065] S12. Acquire the intravascular ultrasound image to be processed;

[0066] S13. Input the intravascular ultrasound image to be processed into the regression network for processing;

[0067] S14. Obtain the delineation results output by the regression network.

[0068] The principle of steps S1-S14 is as follows: Figure 2 As shown, the regression network is trained using intravascular ultrasound images, and the trained regression network is used to process the intravascular ultrasound images to be processed, outputting the delineation results of the intima and external elastic membrane in the intravascular ultrasound images to be processed.

[0069] When performing step S1, an intravascular ultrasound image dataset can be constructed by clinical practice or by collecting data from a database.

[0070] During step S2, two analysts can manually sketch the contours, and a highly qualified analyst can review all the sketching results and make appropriate corrections. For controversial contours, an interventional cardiologist with more than 10 years of experience can perform the final contour sketching, thereby sketching the intima and external elastic membrane contours in the intravascular ultrasound image of the sample.

[0071] After completing steps S1-S2, 90% of the intravascular ultrasound image dataset can be selected as the training set according to the patient ID, and the remaining 10% can be used as the test set. This ensures that the training set and the test set are independent and mutually exclusive, and there is no information leakage problem.

[0072] When performing step S3, which involves sampling the intima contour to obtain intima contour sampling points, the following steps can be performed:

[0073] S301. Determine the rightmost intersection point of the intimal contour and the center horizontal extension line of the intravascular ultrasound image;

[0074] S302. Starting from the rightmost intersection point determined in step S301, sample the inner membrane contour with equal arc length to obtain the inner membrane contour sampling points.

[0075] By executing steps S301-S302, the sampling of the intima contour is completed, and the intima contour sampling points are obtained.

[0076] When performing step S4, which involves resampling the external elastic membrane profile to obtain resampling points, the following steps can be performed:

[0077] S401. Determine the rightmost intersection point of the outer elastic membrane contour and the center horizontal extension line of the intravascular ultrasound image of the sample;

[0078] S402. Starting from the rightmost intersection point determined in step S401, sample the outer elastic membrane profile with equal arc length to obtain the sampling points of the outer elastic membrane profile;

[0079] S403. Determine the optimal arc length offset based on the sampling points of the inner membrane contour and the outer elastic membrane contour;

[0080] S404. Resample the external elastic membrane profile with the optimal arc length offset to obtain the resampled points of the external elastic membrane profile.

[0081] In step S402, an arc length Δs0 can be set, and the rightmost intersection point is determined as the first sampling point. Sampling begins every arc length Δs0 to obtain the external elastic membrane profile sampling points. The external elastic membrane profile sampling points can be a set of multiple sampled values. After completing steps S401-S402 and sampling the external elastic membrane profile, steps S403-S404 are executed to resample the external elastic membrane profile.

[0082] In step S403, the formula can be used. The optimal arc length offset is determined by finding the Δs that minimizes C(Δs). Here, N is the number of sampling points, which can be calculated based on the error between all sampled contours and the true contour; x L (s i ) and y L (s i ) represents the i-th arc length segment on the inner membrane contour sampling point, where x L (s i The x-axis represents the arc length of the arc segment, and the y-axis represents the arc length of the arc segment. L (s i () can represent the arc length value in the y direction of this arc segment; x M (s i ) and y M (s i ) represents the i-th arc length segment on the external elastic membrane profile sampling point, where x M (s i The x-axis represents the arc length of the arc segment, and the y-axis represents the arc length of the arc segment. M (s i ) can represent the arc length value in the y direction of the arc segment; λ is a coefficient, and λ can be set to an appropriate value.

[0083] After determining the optimal arc length offset Δs in step S403, step S404 can be executed to resample the external elastic membrane profile using the optimal arc length offset Δs, thereby obtaining resampled points for the external elastic membrane profile. Specifically, when performing resampling in step S404, the process of step S402 can be followed, with the rightmost intersection point determined in step S401 being designated as the first sampling point. Sampling begins every arc length Δs, thus obtaining resampled points for the external elastic membrane profile. The resampled points for the external elastic membrane profile can be a set of multiple sampled values.

[0084] After obtaining the inner membrane contour sampling points and the outer elastic membrane contour resampling points, step S5 is executed to pair and stack the inner membrane contour sampling points and the outer elastic membrane contour resampling points of all training sets, resulting in a paired contour sampling point matrix x. By sequentially pairing the inner membrane contour sampling points with the resampled outer elastic membrane contour points, the correlation between the inner and outer membrane contour points can be strengthened.

[0085] After obtaining the paired contour sampling point matrix x, step S6 is executed to perform principal component analysis on the paired contour sampling point matrix x to obtain the decoding matrix PD. 1 / 2 Where P is the eigenvector matrix of the paired contour sampling point matrix x, and D is a diagonal matrix containing the first d eigenvector values. That is, P can be obtained by eigenvalue decomposition of the paired contour sampling point matrix x, and the first d eigenvector values ​​D are selected from it. The specific value of d can be determined by calculating the reconstruction error between all sampled contours and the true contour in the training set. This is the arithmetic mean of the paired contour sampling point matrix x. The eigenvector matrix P, and the diagonal matrix D containing the first d eigenvector values ​​of P, are stored for use in subsequent decoding processes.

[0086] When performing step S7, sample intravascular ultrasound images can be read from the constructed intravascular ultrasound image data training set.

[0087] During step S8, theoretically, any type of regression network can be used as the backbone network for encoding. The most suitable backbone network structure is determined based on the training set. The paired contour encoding network used in this invention is as follows: Figure 3 As shown. The network takes three consecutive frames of intravascular ultrasound images as input and outputs a predicted contour feature vector b. The network consists of a backbone network and two linear layers. The backbone network uses the EfficientNet-B0 architecture and contains 18 layers, including two separate convolutional layers and two types of mobile inverted bottlenectconvolution (MBConv) modules. MBConv1 and MBConv6 modules are shown below. Figure 2As shown, the modules all employ depthwise separable convolution (DWConv) instead of traditional convolution, reducing the number of parameters required by the model and lowering its computational cost. Furthermore, the model uses the Swish function as the network activation function, which improves performance to some extent. SE (Squeeze-and-Excitation) is used for channel-oriented attention mechanisms. Two linear layers are used to regress contour features; the first linear layer has 2048 channels, and the second linear layer's channel count corresponds to the dimension of the regressed contour features. The appropriate dimension was determined experimentally; this invention uses a dimension of 128. Figure 3 In the diagram, K represents the kernel size, N represents the number of kernels, S represents the stride of the convolution, BN represents batch normalization, and Conv represents ordinary convolution.

[0088] After obtaining the contour feature b, step S9 is executed, using the decoding matrix to deencode the contour feature and obtain the predicted paired contour points. Specifically, the decoder executes the formula... Decoding is performed to output the predicted paired contour points.

[0089] The decoding process in step S9 only requires simple linear operations, which can avoid the large number of parameters required by the decoder based on the convolutional neural network, and at the same time improve the running speed of the decoder.

[0090] After performing step S9, the predicted paired contour points are obtained. Then, step S10 can be executed to determine the error loss based on the predicted paired contour points. Specifically, this can be done according to the formula... Determine the error loss; where, l c For error loss, S represents the set consisting of the inner membrane profile and the outer elastic membrane profile, and s represents a point in set S. x represents the component corresponding to s in the predicted paired contour points. s This represents the component corresponding to s in the paired contour sampling point matrix.

[0091] In step S11, the error loss can be calculated based on... c The size of the error loss is used to train the regression network. Specifically, if the error loss l c If the value exceeds a preset threshold, the parameters of the regression network can be changed, and steps S1-S11 can be executed again. Conversely, if the value is below a preset threshold, the training of the regression network can be terminated, the final parameters of the regression network can be saved, and the trained regression network can be obtained. The error loss is then calculated using the calculated error loss. cThe training process ensures that the paired contour points generated by the regression network and decoder are as close as possible to the contours of the real inner and outer elastic membranes.

[0092] The trained regression network and decoder have the ability to process intravascular ultrasound images, extract features of the contours of the intima and external elastic membrane in the intravascular ultrasound images, and thus delineate the contours of the intima and external elastic membrane in the intravascular ultrasound images.

[0093] After obtaining the trained regression network, the following steps can be performed:

[0094] S12. Acquire the intravascular ultrasound image to be processed;

[0095] S13. Input the intravascular ultrasound image to be processed into the regression network for processing;

[0096] S14. Obtain the delineation results output by the regression network.

[0097] In summary, the intravascular ultrasound image processing method based on paired contour encoding in this embodiment can directly predict and regress the contour coordinates of the intima and external elastic membrane in intravascular ultrasound images. By directly regressing and predicting the coordinates of the target contour, the multi-layer deencoder required by previous fully convolutional neural network-based models can be avoided, significantly reducing the training parameters required by the model and accelerating the inference time. Simultaneously, by using a regression network, the accuracy of other fully convolutional segmentation networks can be achieved, indicating that the method in this embodiment is quite effective for the automatic delineation of intravascular ultrasound images. Furthermore, it facilitates the clinical deployment of the model, enhancing its practicality.

[0098] Based on the execution of steps S1-S14, the following steps can also be performed:

[0099] S15. Determine the anatomical rationality index value based on the intravascular ultrasound image and delineation results to be processed;

[0100] S16. Evaluate the performance of the regression network based on the anatomical rationality index value.

[0101] In step S13, based on the characteristics of the intima and external elastic membrane in intravascular ultrasound images, while automatically plotting the routine indicators of the results on the test set, the corresponding anatomical rationality indicators were also statistically analyzed. The routine results include the following three types:

[0102] a. HD (Hausdorff Distance): Hausdorff distance. As shown in the following formula, HD is used to measure the predicted contour C. pred and actual outline C trueThe maximum value of the shortest distance between all contour points a and b. The larger the value, the greater the error between the two contours, and vice versa.

[0103]

[0104] b. DSC (Dice similarity coefficient): The Dice similarity coefficient. As shown in the following formula, it is used to measure the region M within the predicted contour. pred and the actual contour region M true The similarity between the regions ranges from 0 to 1. A value closer to 1 indicates greater similarity between the contour regions, while a value closer to 0 indicates greater difference.

[0105]

[0106] c. JI (Jaccard Index): Jaccard coefficient. As shown in the following formula, similar to DSC, it is used to measure the similarity between the predicted contour region and the actual contour region, with a value between 0 and 1. The closer to 1, the more similar the contour regions are; the closer to 0, the greater the difference.

[0107]

[0108] Of the three conventional indicators above, the predicted region M within the contour is... pred This can represent the outline regions of the inner and outer elastic membranes in the delineation result, with the actual outline region M being the outer elastic membrane region. true It can represent the contour regions of the intima and the outer elastic membrane in the intravascular ultrasound image to be processed.

[0109] In addition to using the three conventional indicators mentioned above—Hausdorff distance, Dice similarity coefficient, and Jaccard coefficient—as indicators of anatomical rationality, other anatomical rationality indicators can also be used.

[0110] Specifically, based on the anatomical characteristics of the intima and external elastic membrane in intracanal ultrasound images, including: 1) the intima and external elastic membrane have smooth, closed contours; and 2) the intima is always enclosed by the external elastic membrane, with no instances of the intima protruding beyond it. Therefore, the value calculated using the following formula can also be used as an anatomical rationality index to evaluate the anatomical rationality of the automatically drawn results:

[0111]

[0112] The errortype includes three of the most common anatomical error types: 1) the intima protrudes beyond the outer elastic membrane; 2) there is a discontinuous region; 3) there is a hole within the intima or the outer elastic membrane. It is an indicator function, C i This refers to the case of the i-th intratubular ultrasound image, specifically when the i-th intratubular ultrasound image exhibits at least one of the conditions represented by errortype. Take a certain value (e.g., 0) when the i-th intratubular ultrasound image does not contain any of the conditions represented by errortype. Alternatively, m can be set to the total number of intratubular ultrasound images (e.g., 1).

[0113] By executing steps S15-S16, a novel paired contour encoding method was implemented, avoiding anatomical inconsistencies present in previous fully convolutional network segmentation models. By combining the anatomical information of the intima and external elastic lamina with a convolutional neural network, the contour coordinates of the intima and external elastic lamina are directly predicted, ensuring that the predicted results avoid anatomical errors. More importantly, the generated results do not require further image post-processing to obtain reasonable contours, thus facilitating direct transfer to subsequent research in vascular reconstruction and hemodynamic analysis, demonstrating significant engineering application value.

[0114] A computer program for executing the intravascular ultrasound image processing method based on paired contour coding in this embodiment can be written into a computer device or storage medium. When the computer program is read out and run, the intravascular ultrasound image processing method based on paired contour coding in this embodiment is executed, thereby achieving the same technical effect as the intravascular ultrasound image processing method based on paired contour coding in the embodiment.

[0115] It should be noted that, unless otherwise specified, when a feature is referred to as "fixed" or "connected" to another feature, it can be directly fixed or connected to the other feature, or indirectly fixed or connected to the other feature. Furthermore, the descriptions of "upper," "lower," "left," and "right" used in this disclosure are only relative to the relative positional relationships of the various components of this disclosure in the accompanying drawings. The singular forms "a," "described," and "the" used in this disclosure are also intended to include the plural forms, unless the context clearly indicates otherwise. Moreover, unless otherwise defined, all technical and scientific terms used in this embodiment have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this embodiment specification is only for describing particular embodiments and is not intended to limit the invention. The term "and / or" as used in this embodiment includes any combination of one or more of the associated listed items.

[0116] It should be understood that although the terms first, second, third, etc., may be used to describe various elements in this disclosure, these elements should not be limited to these terms. These terms are only used to distinguish elements of the same type from each other. For example, a first element may also be referred to as a second element without departing from the scope of this disclosure, and similarly, a second element may also be referred to as a first element. The use of any and all instances or exemplary language (“e.g.,” “such as,” etc.) provided in this embodiment is intended only to better illustrate embodiments of the invention and, unless otherwise required, does not impose a limitation on the scope of the invention.

[0117] It should be recognized that embodiments of the present invention can be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium. The method can be implemented using standard programming techniques—including a non-transitory computer-readable storage medium configured with a computer program, wherein such a storage medium causes the computer to operate in a specific and predefined manner—according to the methods and drawings described in the specific embodiments. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit (ASIC).

[0118] Furthermore, the procedures described in this embodiment can be performed in any suitable order unless otherwise indicated by this embodiment or clearly contradicted by the context. The procedures (or variations and / or combinations thereof) described in this embodiment can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program includes a plurality of instructions executable by one or more processors.

[0119] Furthermore, the method can be implemented in any suitable type of computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices. Aspects of the invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it is readable by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein. Furthermore, the machine-readable code, or portions thereof, can be transmitted via wired or wireless networks. The invention described in this embodiment includes these and other different types of non-transitory computer-readable storage media when such media comprises instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. When programmed according to the methods and techniques described in the invention, the invention also includes the computer itself.

[0120] A computer program can be applied to input data to perform the functions described in this embodiment, thereby transforming the input data to generate output data stored in non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including specific visual depictions of physical and tangible objects generated on the display.

[0121] The above description is merely a preferred embodiment of the present invention. The present invention is not limited to the above-described embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention, as long as they achieve the technical effects of the present invention by the same means, should be included within the scope of protection of the present invention. Within the scope of protection of the present invention, the technical solutions and / or implementation methods can have various modifications and variations.

Claims

1. An intravascular ultrasound image processing method based on pairing contour coding, characterized by, The paired contour coding-based intravascular ultrasound image processing method comprises the following steps: acquiring intravascular ultrasound image data; contouring the intravascular ultrasound image to determine an intimal contour and an external elastic membrane contour; sampling the intimal contour to obtain intimal contour sampling points; resampling the external elastic membrane contour to obtain external elastic membrane contour resampling points; pairing and stacking the intimal contour sampling points and the external elastic membrane contour resampling points to obtain a paired contour sampling point matrix; performing principal component analysis on the paired contour sampling point matrix to obtain a decoding matrix; obtaining sample intravascular ultrasound images and corresponding paired contour sampling points; using a regression network to obtain predicted contour features according to the sample intravascular ultrasound images; using a decoding matrix, the prediction profile feature is decoded to obtain a prediction paired profile point; by executing the formula is decoded to output the prediction paired profile point ; wherein, is an arithmetic mean value of a paired profile sampling point matrix , the decoding matrix is , and the prediction profile feature is . determining an error loss according to the predicted paired contour points; training the regression network according to the error loss; the resampling of the external elastic membrane contour to obtain external elastic membrane contour resampling points comprises the following steps: determining the rightmost intersection point of the external elastic membrane contour and the horizontal extension line of the center of the sample intravascular ultrasound image; starting from the rightmost intersection point, sampling the external elastic membrane contour at equal arc lengths to obtain the external elastic membrane contour sampling points; determining an optimal arc length offset according to the intimal contour sampling points and the external elastic membrane contour sampling points; resampling the external elastic membrane contour by the optimal arc length offset to obtain the external elastic membrane contour resampling points.

2. The intravascular ultrasound image processing method based on paired contour coding according to claim 1, characterized in that, the sampling of the intimal contour to obtain intimal contour sampling points comprises the following steps: determining the rightmost intersection point of the intimal contour and the horizontal extension line of the center of the sample intravascular ultrasound image; starting from the rightmost intersection point, sampling the intimal contour at equal arc lengths to obtain the intimal contour sampling points.

3. The intravascular ultrasound image processing method based on paired contour coding according to claim 1, characterized in that, the determination of an optimal arc length offset according to the intimal contour sampling points and the external elastic membrane contour sampling points comprises the following steps: determining to cause minimizing as the optimal arc length offset; wherein, , is the number of sampling points, and denotes the arc length segment number of the inner membrane profile sampling point, and denotes the arc length segment number of the outer elastic membrane profile sampling point, is a coefficient.

4. The intravascular ultrasound image processing method based on paired contour coding according to claim 1, characterized in that, the determination of an error loss according to the predicted paired contour points comprises the following steps: According to the formula determining the error loss; wherein, is the error loss, denotes a set consisting of the inner membrane profile and the outer elastic membrane profile, denotes a set of points, denotes a component of the prediction pair profile points corresponding to, denotes a component of the pair profile sampling points matrix corresponding to.

5. The intravascular ultrasound image processing method based on matched contour coding according to any one of claims 1 to 4, characterized in that, the paired contour coding-based intravascular ultrasound image processing method further comprises the following steps: acquiring an intravascular ultrasound image to be processed; inputting the intravascular ultrasound image to be processed into the regression network for processing; obtaining a contouring result output by the regression network.

6. The intravascular ultrasound image processing method based on paired contour coding according to claim 5, characterized in that, the paired contour coding-based intravascular ultrasound image processing method further comprises the following steps: determining an anatomical rationality index value according to the intravascular ultrasound image to be processed and the contouring result; evaluating the performance of the regression network according to the anatomical rationality index value.

7. A computer apparatus, comprising: a device comprising a memory and a processor, the memory being configured to store at least one program, and the processor being configured to load the at least one program to execute the paired contour coding-based intravascular ultrasound image processing method according to any one of claims 1-6.

8. A computer readable storage medium having stored therein a program which is executable by a processor, characterized in that, the program executable by the processor, when executed by the processor, is configured to execute the paired contour coding-based intravascular ultrasound image processing method according to any one of claims 1-6.

Images