An early myocardial infarction detection device, method and medium based on echocardiogram

Abstract

The present invention relates to an early myocardial infarction detection device and method based on echocardiogram. An intelligent detection model is deployed in the early myocardial infarction detection device. The intelligent detection model includes a data acquisition module, a deep contour localization network module, and a classification processing module. The data acquisition module is used to acquire ultrasonic video data of an echocardiogram. The deep contour localization network module is used to take the ultrasonic video data as input and locate the endocardium in each frame of the ultrasonic video data. The classification processing module is used to extract motion features based on the located endocardium in each frame of the image, and classify whether the detection result of the current echocardiogram belongs to early myocardial infarction based on the extracted motion features. An intelligent detection model for detecting early myocardial infarction can be deployed in the device, and this intelligent detection model can improve robustness and detection speed.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and medical image processing, and particularly to an early myocardial infarction detection device, method and medium based on echocardiogram. Background Art

[0002] Myocardial infarction (MI) is a heart disease with a large patient population and a high fatality rate. Early diagnosis of it is crucial for preventing further damage to heart tissue and reducing the mortality rate. Echocardiogram can non-invasively and inexpensively display abnormal movements of local ventricular walls, and thus can be applied to the rapid screening of early myocardial infarction.

[0003] However, the inventors of the present application found in their research that the existing deep learning algorithms usually process echocardiograms based on computer vision processing methods of semantic segmentation, which rely on relatively clear echocardiograms. In some environments with poor image quality and unclear myocardial imaging, the robustness of the deep learning algorithms is relatively poor. And due to the high cost of labeled samples in the prior art, the existing algorithm models are trained based on a small number of samples with low accuracy, so the detection accuracy of the obtained algorithm models is not high. At the same time, the semantic segmentation method of the existing non-deep learning method uses an active contour algorithm to extract the endocardium of the ventricle, which adopts an iterative optimization calculation method, with slow algorithm speed and a long time period for video processing. Summary of the Invention

[0004] Aiming at the above problems, the purpose of the present invention is to provide an early myocardial infarction detection device based on echocardiogram, in which an intelligent detection model for detecting early myocardial infarction can be deployed, and the intelligent detection model can improve the robustness and detection speed.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present application provides an early myocardial infarction detection device based on echocardiogram. An intelligent detection model is deployed in the early myocardial infarction detection device, and the intelligent detection model includes a data acquisition module, a deep contour localization network module and a classification processing module;

[0007] The data acquisition module is used to acquire the ultrasonic video data of the echocardiogram;

[0008] The deep contour localization network module is used to take the ultrasonic video data as input and locate the endocardium of each frame of image in the ultrasonic video data;

[0009] The classification processing module is used to extract motion features based on the endocardium in each located frame image, and classify whether the detection result of the current echocardiogram belongs to early myocardial infarction based on the extracted motion features.

[0010] In an implementation manner of the present application, the depth contour localization network module includes:

[0011] The backbone network is used to extract image features of each frame image in the ultrasonic video data;

[0012] The key point detection network is used to output the coordinates of the first preset number of key points for characterizing the endocardium according to the image features of each frame image;

[0013] The contour correction network is used to take the coordinates of the first preset number of key points as input, calculate the offset for correction, and output the coordinates of the second preset number of localization points, and represent the endocardium with the broken line segment of the second preset number of localization points, where the second preset number is greater than the first preset number.

[0014] In an implementation manner of the present application, the classification processing module is used to divide the located endocardium into predefined segments, and extract motion features according to the maximum offset of each segment of the endocardium relative to the initial frame in different frame images;

[0015] The classification processing module is further used to obtain a myocardial infarction classification result by using a machine learning algorithm based on the pre-stored case data set and the extracted motion features.

[0016] In an implementation manner of the present application, the machine learning algorithm includes a support vector machine algorithm or a random forest algorithm.

[0017] In an implementation manner of the present application, the intelligent detection model further includes a first training module for training the depth contour localization network module.

[0018] In an implementation manner of the present application, the training method for training the depth contour localization network module includes: making the depth contour localization network module minimize the L1 norm loss between the endocardium localization point coordinates output by the labeled sample and the labeled coordinates.

[0019] In an implementation manner of the present application, the intelligent detection model further includes a labeled sample acquisition module for interactively processing the collected sample video according to the pre-trained motion estimation network to obtain a labeled sample for labeling the endocardium of each frame image.

[0020] In an implementation manner of the present application, the intelligent detection model further includes a second training module for pre-training the motion estimation network.

[0021] In one implementation of the present application, the pre-training includes:

[0022] According to the collected sample video, use the phase correlation algorithm to calculate the motion field of the ultrasound video;

[0023] Process the sample video according to the calculated motion field to obtain a sequence of pseudo-sample pairs;

[0024] Use the sequence of pseudo-sample pairs as the input of the motion estimation network, and train the motion estimation network in a way that minimizes the mean square error between the predicted motion field output by the motion estimation network and the calculated motion field.

[0025] The second aspect of the present application provides a method for early myocardial infarction detection based on echocardiogram, including:

[0026] Obtain the ultrasound video data of the echocardiogram;

[0027] Use the ultrasound video data as the input to locate the endocardium in each frame of the ultrasound video data;

[0028] Extract motion features according to the located endocardium in each frame of the image, and classify whether the detection result of the current echocardiogram belongs to early myocardial infarction based on the extracted motion features.

[0029] The third aspect of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the following steps:

[0030] Obtain the ultrasound video data of the echocardiogram;

[0031] Use the ultrasound video data as the input to locate the endocardium in each frame of the ultrasound video data;

[0032] Extract motion features according to the located endocardium in each frame of the image, and classify whether the detection result of the current echocardiogram belongs to early myocardial infarction based on the extracted motion features.

[0033] Due to the above technical solutions adopted by the present invention, it has the following advantages: In the early myocardial infarction detection device in the solution of the present invention application, an intelligent detection model can be deployed. The intelligent detection model includes a data acquisition module, a deep contour localization network module, and a classification processing module. The ultrasonic video data of the echocardiogram is acquired through the data acquisition module, and then through the deep contour localization network module, the deep learning algorithm is used to localize the endocardium of each frame of image. Then through the classification processing module, the motion characteristics of the endocardium are extracted, and based on the motion characteristics, the detection results of early myocardial infarction are accurately classified. Compared with the existing image semantic segmentation algorithm, it does not need to be processed based on high-definition video, which improves the accuracy and robustness of the model and also improves the speed of calculation and processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 FIG. is a schematic diagram of the functional module structure of an early myocardial infarction detection device based on echocardiogram provided by an embodiment of the present invention;

[0035] Figure 2 FIG. is a schematic diagram of the network architecture adopted by the deep contour localization network module in an embodiment of the present invention;

[0036] Figure 3 FIG. is a schematic diagram of the data flow of the classification processing module for classifying early myocardial infarction in an embodiment of the present invention;

[0037] Figure 4 FIG. is a schematic diagram of the functional module structure of another early myocardial infarction detection device based on echocardiogram provided by an embodiment of the present invention;

[0038] Figure 5 FIG. is a schematic diagram of the data flow for pre-training the motion estimation network in an embodiment of the present invention;

[0039] Figure 6 FIG. is an application scenario diagram of obtaining labeled samples through interactive processing according to the motion estimation network in an embodiment of the present invention;

[0040] Figure 7 FIG. is a comparison schematic diagram of the result of endocardium localization of the original image and the existing semantic segmentation in an embodiment of the present invention;

[0041] Figure 8 FIG. is a schematic diagram of the structure of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions of the embodiments of the present invention in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present invention.

[0043] In view of the fact that the existing deep learning algorithms for echocardiogram processing usually rely on computer vision processing methods based on semantic segmentation, which depend on relatively clear echocardiograms. In some environments with poor image quality and unclear myocardial imaging, the robustness of the deep learning algorithms is relatively poor. Moreover, due to the high cost of labeled samples in the existing technology, the present application provides an early myocardial infarction detection device and medium based on echocardiogram. An intelligent detection model is deployed in the early myocardial infarction detection device, and the intelligent detection model includes a data acquisition module, a deep contour localization network module, and a classification processing module. The data acquisition module is used to acquire the ultrasonic video data of the echocardiogram. The deep contour localization network module is used to take the ultrasonic video data as input and locate the endocardium of each frame of the ultrasonic video data. The classification processing module is used to extract the motion features based on the located endocardium of each frame of the image, and classify whether the detection result of the current echocardiogram belongs to early myocardial infarction based on the extracted motion features.

[0044] See Figure 1 , in an embodiment of the present application, an early myocardial infarction detection device based on echocardiogram is provided. An intelligent detection model for detecting early myocardial infarction can be deployed in the device, and the intelligent detection model can improve the robustness and detection speed.

[0045] In the embodiments of the present application, an early myocardial infarction detection device based on echocardiogram can be, but is not limited to, a computer device, and the detection of early myocardial infarction can be implemented in a hardware or software manner. For example, the computer device has a data interface and has the ability to receive and transmit data. The computer device may also include a memory that stores various programs or instructions for processing the received data, such as deep learning algorithms, neural network models, etc. Of course, the computer also includes a processor with computing functions that can process data or respond to instructions, etc.

[0046] In the embodiments of the present application, an intelligent detection model is deployed in the early myocardial infarction detection device, and the intelligent detection model includes a data acquisition module, a deep contour localization network module, and a classification processing module.

[0047] Among them, the data acquisition module is used to acquire the ultrasonic video data of the echocardiogram.

[0048] A deep contour localization network module, which is used to take ultrasonic video data as input and locate the endocardium of each frame of image in the ultrasonic video data.

[0049] A classification processing module, which is used to extract motion features based on the endocardium in each located frame of image, and classify whether the detection result of the current echocardiogram belongs to early myocardial infarction based on the extracted motion features.

[0050] In the early myocardial infarction detection device in the solution of the present invention application, an intelligent detection model can be deployed. The intelligent detection model includes a data acquisition module, a deep contour localization network module, and a classification processing module. The ultrasonic video data of the echocardiogram is obtained through the data acquisition module, and then through the deep contour localization network module, a deep learning algorithm is used to locate the endocardium of each frame of image. Then, through the classification processing module, the motion features of the endocardium are extracted, and the detection results of early myocardial infarction are accurately classified based on the motion features. Compared with the existing image semantic segmentation algorithm, it does not need to be processed based on high-definition video, which improves the accuracy and robustness of the model and also improves the speed of calculation and processing.

[0051] Next, in some more detailed embodiments of the present application, in combination with more drawings, the structure, working principle, and advantages of the above-mentioned early myocardial infarction detection device will be described.

[0052] Such as Figure 1 , in the early myocardial infarction detection device 100 based on echocardiogram in the embodiment of the present application, an intelligent detection model is deployed therein. The intelligent detection model includes a data acquisition module 101, a deep contour localization network module 102, and a classification processing module 103.

[0053] Among them, the data acquisition module 101 is used to obtain the ultrasonic video data of the echocardiogram.

[0054] Specifically, in the early myocardial infarction detection device in the embodiment of the present application, the data acquisition module 101 can be connected to an ultrasonic device. The ultrasonic device can collect the echocardiogram of a patient or a tester. The data acquisition module 101 can, but is not limited to, transmit the ultrasonic video data of the collected echocardiogram to the early myocardial infarction detection device through various existing serial or parallel interfaces.

[0055] The deep contour localization network module 102 is used to take the ultrasonic video data as input and locate the endocardium of each frame of image in the ultrasonic video data.

[0056] Specifically, in the embodiments of the present application, different from existing semantic segmentation algorithms, the deep contour localization network module 102 of the present application uses a deep contour localization network based on deep learning, and can also have a robust performance in the case of poor image quality and unclear myocardial imaging.

[0057] The deep contour localization network module in the embodiments of the present application includes:

[0058] A backbone network for extracting image features of each frame of image in the ultrasonic video data;

[0059] A key point detection network for outputting the coordinates of a first preset number of key points used to represent the endocardium of the ventricle according to the image features of each frame of image;

[0060] A contour correction network for taking the coordinates of a first preset number of key points as input, calculating and correcting the offset, and outputting the coordinates of a second preset number of positioning points, and representing the endocardium of the ventricle with the broken line segment of the second preset number of positioning points, wherein the second preset number is greater than the first preset number.

[0061] Specifically in Figure 2 In the example of, the deep contour localization network extracts image features with ResNet18 as the backbone network. A fully connected layer is added to the end of the backbone network, and through regression, the coordinates of three key points (the starting point, vertex, and ending point of the endocardium of the ventricle) are output. The three key points form an initial contour. Then, taking the features at the position of the initial contour (the triangular contour connected by the three key points) as input, the contour correction network continuously outputs the offset to correct the initial contour, and finally outputs the endocardium localization connected by 32 coordinate points. The contour correction network is constructed by circular convolution to realize the fusion of features on the contour.

[0062] A classification processing module 103 for extracting motion features according to the endocardium in each frame of the located image, and classifying whether the detection result of the current echocardiogram belongs to early myocardial infarction based on the extracted motion features.

[0063] Specifically, the classification processing module 103 is used to divide the located endocardium into predefined segments, extract motion features according to the maximum offset of each segment of the endocardium in different frames of images relative to the initial frame. The classification processing module 103 is also used to obtain the myocardial infarction classification result by using a machine learning algorithm based on the pre-stored case data set according to the extracted motion features.

[0064] For example, the pre-stored case data set can be, but is not limited to, the ultrasonic echocardiogram video data of 130 patients. Each patient includes video data of the apical four-chamber view (A4C) and the apical two-chamber view (A2C), that is, 260 segments of videos. Each patient includes the annotation of whether it is myocardial infarction.

[0065] The classification processing module 103 obtains the endocardial localization in the ventricle processed by the deep contour localization network, as Figure 2 shown in the schematic output.

[0066] The classification processing module 103 further processes the located endocardium in the ventricle to extract the features therein, specifically as Figure 3 shown. First, the endocardium in the ventricle is segmented. The starting point to the vertex is equally divided into 7 segments by length. The 1-2 segments are numbered 1, the 3-4 segments are numbered 2, and the 5-6 segments are numbered 3. Similarly, the end point to the vertex is equally divided into 7 segments by length. The 1-2 segments are numbered 6, the 3-4 segments are numbered 5, and the 5-6 segments are numbered 4. Both the A4C and A2C views are segmented in this way. Each view contains 6 segments of the endocardium in the ventricle. Calculate the offset of the endocardium in the ventricle in each frame of the image relative to the first frame, and take the maximum value of each segment offset as the feature to form a 12-dimensional feature vector. To eliminate the influence of the scale of image magnification and reduction on the features, the offset is normalized by dividing the distance from the starting point to the end point of the endocardium in the ventricle.

[0067] Furthermore, the classification processing module 103 obtains the myocardial infarction classification result (that is, classified as "belonging to early myocardial infarction" or "not belonging to early myocardial infarction") based on the extracted features and the pre-stored case data set by using a machine learning algorithm. The machine learning algorithm can be a support vector machine algorithm (SVM) or a random forest algorithm (RF). When the deep contour localization network in the embodiment of the present application is Figure 2 the network structure shown in the figure, when using the pre-stored case data set for verification, the performance of SVM classification is better than that of RF.

[0068] The accuracy of the detection of early myocardial infarction by the device in the embodiment of the present application depends on the performance of the deep contour localization network and the classification processing module. In the embodiment of the present application, the deep contour localization network can be pre-trained, or according to actual needs, sample data can be collected in real time to re-train the deep contour localization network.

[0069] Please refer to Figures 4 to 6 together for the training of the network model in the embodiment of the present application.

[0070] Specifically, in Figure 4 a device 200 for early myocardial infarction detection based on echocardiogram is provided. An intelligent detection model is deployed in the device 200 for early myocardial infarction detection. The intelligent detection model includes: a data acquisition module 101, a deep contour localization network module 102, a classification processing module 103, a first training module 104, a labeled sample acquisition module 105, a motion estimation network 106, and a second training module 107.

[0071] Among them, for the working principles and processes of the data acquisition module 101, the deep contour localization network module 102, and the classification processing module 103, reference can be made to the foregoing embodiments. In this embodiment, the training of the model will be mainly described, and thus will not be repeated here.

[0072] The first training module 104 is used to train the deep contour localization network module. The training method includes: minimizing the L1 norm loss between the coordinates of the endocardial localization points output by the deep contour localization network module with the labeled samples and the labeled coordinates.

[0073] The labeled sample acquisition module 105 is used to perform interactive processing on the collected sample video according to the pre-trained motion estimation network to obtain labeled samples for labeling the endocardium of each frame of the image.

[0074] The second training module 107 is used to pre-train the motion estimation network 106.

[0075] Specifically, the process of pre-training includes:

[0076] Calculating the motion field of the ultrasound video using the phase correlation algorithm according to the collected sample video;

[0077] Processing the sample video according to the calculated motion field to obtain a sequence of pseudo-sample pairs;

[0078] Using the sequence of pseudo-sample pairs as the input of the motion estimation network, and training the motion estimation network in a way that minimizes the mean square error between the predicted motion field output by the motion estimation network and the calculated motion field.

[0079] Next, in combination with Figure 5 and Figure 6 , the above training process and the process of obtaining labeled samples will be described.

[0080] Specifically, the second training module 107 is used to pre-train the motion estimation network 106. Before training the motion estimation network, it is necessary to collect a training set and a test set for training through an ultrasound device. For example, 12 echocardiogram videos of 5 patients can be collected using an ultrasound device for the training and testing of the motion estimation algorithm. The training set of the motion estimation network uses the video data of 1 patient, including 8 videos and a total of 560 frames of images; the test set uses the video data of 4 patients, including 4 videos and a total of 257 frames. Each patient includes video data of the apical four-chamber view (A4C) and the apical two-chamber view (A2C).

[0081] The embodiment of the present application adopts the method of generating pseudo-sample pairs to generate image pairs that conform to the actual cardiac motion characteristics, and uses the generated image pairs to train the motion estimation network. Specifically, asFigure 5 As shown, given the Nth frame image I of a segment of ultrasound video N and the (N + 1)th frame image I N+1 , the predicted motion field F can be obtained using the traditional phase cross - correlation method PC (The so - called motion field records the motion vectors of each pixel point in a graphical manner). Using F PC to warp I N+1 :

[0082] I′ N = warp -1 (I N+1 , F PC )

[0083] Then, using the obtained I′ N and I N+1 as the input of the motion estimation network, the motion estimation network is trained in a way that minimizes the mean square error between the predicted motion field output by the motion estimation network and the calculated motion field, denoted as:

[0084] min MSEF PC , Φ(I′ N , I N+1 ))

[0085] where Φ represents the motion estimation network

[0086] The above is the process of training the motion estimation network. When the trained motion estimation network receives consecutive ultrasound video image frames subsequently, for every two adjacent images, the corresponding predicted motion field can be output

[0087] Furthermore, based on the trained motion estimation network, through interactive processing, a large number of low - cost and more accurate labeled samples compared with the prior art can be obtained. In traditional technologies, some algorithms for locating the endocardium are trained based on a small number of samples. The accuracy of these small - number samples is not high. If a large number of accurately labeled samples are required, it needs long - time and careful manual discrimination and labeling, with high costs. Therefore, the detection accuracy of traditional algorithms trained based on a small number of accurate samples is naturally limited

[0088] Such as Figure 6As shown, the solution of the present application can obtain a large number of accurately labeled samples at low cost. For a video, a user (such as a doctor) only needs to label the position of the endocardium in the first frame of the image. As described above, when the trained motion estimation network subsequently receives consecutive ultrasound video image frames, the corresponding predicted motion field can be output for every two adjacent images. Based on the predicted motion field, the labels of the remaining frames can be automatically obtained through the motion estimation network. In addition, if a user such as a doctor is not satisfied with the labeling results of some frames, only simple interaction adjustment needs to be performed on the corresponding frames, and the subsequent labeling is automatically executed by the motion estimation network. The labeled data can be used for the training of the deep contour localization network. For example, in an embodiment of the present application, the positions of the endocardium in 100 video segments can be labeled, including 50 A4C video segments and 50 A2C video segments.

[0089] Furthermore, the labeled samples are used to train the deep contour localization network. The labeled samples can be divided into a training set and a validation set. For example, the training set includes 50 video segments, and the test set contains 50 video segments. The training method for training the deep contour localization network module includes: minimizing the L1 norm loss between the coordinates of the endocardium localization points output by the deep contour localization network module and the labeled coordinates.

[0090] Next, the effect of the present application is described on a specific software and hardware platform. For example, the parameters of the experimental platform of the present application include: Intel(R) Core(TM) i9-9820X CPU @ 3.30GHz, 64GB RAM, GeForce RTX2080Ti. The trained intelligent detection model described above is deployed on this platform.

[0091] The present application first measures the motion estimation network in terms of tracking accuracy. Given the position of the endocardium in the first frame of the cardiac cycle, the error of the points on the contour (Endpoint Error, EPE) is calculated based on the difference between the tracking result of the subsequent frames and the true marked position of the endocardium. Compared with the existing phase correlation algorithm, the average EPE based on the phase cross-correlation algorithm is reduced from 5.23px to 3.97px (px represents pixel).

[0092] Second, for the labeling of samples, compared with complete manual labeling, the labeling time per frame can be reduced from about 20s to less than 5s.

[0093] Third, for the positioning accuracy of the deep contour localization network, the endocardium localization error (EPE) can reach 4.86px. Compared with the traditional technology using semantic segmentation algorithms, the present application can also have a robust performance in the case of unclear myocardial imaging, such as Figure 7As shown, for images with poor imaging quality, the semantic segmentation algorithm cannot segment continuous myocardium, while the network of the present application directly outputs the coordinate positions of continuous endocardium, which has stronger robustness. And compared with traditional algorithms, the present application has greatly improved the detection speed of the endocardium. For a video of two views of a patient, the traditional algorithm takes 59 s, while the present application only takes less than 0.5 s of processing time.

[0094] Fourth, for the classification effect of the classification processing module, five-fold cross-validation is performed on the pre-stored data described above. The accuracy rate of SVM reaches 83.08%, and the recall rate and precision rate reach 89.92% and 85.60% respectively, realizing relatively accurate screening of early myocardial infarction.

[0095] In another aspect of the embodiments of the present application, an early myocardial infarction detection method based on echocardiogram is further provided, including:

[0096] Obtaining ultrasonic video data of an echocardiogram;

[0097] Using the ultrasonic video data as input, positioning the endocardium of each frame of image in the ultrasonic video data;

[0098] Extracting motion features according to the endocardium in each located frame of image, and classifying whether the detection result of the current echocardiogram belongs to early myocardial infarction based on the extracted motion features.

[0099] The early myocardial infarction detection method in the above embodiments can be implemented by, but not limited to, the early myocardial infarction detection device in the foregoing embodiments. The specific implementation process can refer to the description of the embodiments in the device part.

[0100] At the same time, in another aspect of the present application, a computer-readable storage medium can also be provided.

[0101] A computer-readable storage medium provided by an embodiment of the present application, the storage medium includes a stored program, wherein when the program runs, it controls the device where the storage medium is located to execute the above-mentioned early myocardial infarction detection method. The specific implementation process is not repeated here.

[0102] In another aspect of the embodiments of the present application, a computer device is further provided, which can store the foregoing computer-readable storage medium, and when executing the computer program in the computer-readable storage medium, realizes the detection of early myocardial infarction.

[0103] A computer device 500 provided by an embodiment of the present application, such as Figure 8As shown. The computer device 500 of this embodiment includes: a processor 501, a memory 502, and a computer program 503 stored in the memory and executable on the processor 501. When the processor 501 executes the computer program 503, it implements the early myocardial infarction detection method in the embodiment. To avoid repetition, it will not be elaborated here one by one. Alternatively, when the computer program is executed by the processor 501, it implements the functions of each model / unit in the early myocardial infarction detection device in the embodiment. To avoid repetition, it will not be elaborated here one by one.

[0104] The computer device 500 can be a computing device such as a desktop computer, a notebook, a palm computer, a server, and a cloud server. The computer device may include, but is not limited to, a processor 501 and a memory 502. Those skilled in the art can understand that Figure 8 These are merely examples of the computer device 500 and do not constitute a limitation on the computer device 500. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, the computer device may also include input / output devices, network access devices, a bus, etc.

[0105] The so-called processor 501 may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0106] The memory 502 may be an internal storage unit of the computer device 500, such as the hard disk or memory of the computer device 500. The memory 502 may also be an external storage device of the computer device 300, such as a plug-in hard disk equipped on the computer device 500, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the memory 502 may also include both the internal storage unit and the external storage device of the computer device 500. The memory 502 is used to store the computer program and other programs and data required by the computer device. The memory 502 may also be used to temporarily store data that has been output or will be output.

[0107] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0108] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the above division of units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces, and the indirect coupling or communication connection of devices or units can be in electrical, mechanical, or other forms.

[0109] The integrated units implemented in the form of software function units can be stored in a computer-readable storage medium. The above software function units stored in a storage medium include several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute some steps of the methods in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0110] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An early myocardial infarction detection device based on echocardiogram, characterized in that, an intelligent detection model is deployed in the early myocardial infarction detection device, and the intelligent detection model includes a data acquisition module, a deep contour localization network module, and a classification processing module; the data acquisition module is used to acquire ultrasonic video data of the echocardiogram; the deep contour localization network module is used to take the ultrasonic video data as input and locate the endocardium of each frame of image in the ultrasonic video data; the classification processing module is used to extract motion features based on the located endocardium of each frame of image, and classify whether the detection result of the current echocardiogram belongs to early myocardial infarction based on the extracted motion features; the intelligent detection model further includes a labeled sample acquisition module, which is used to perform interactive processing on the collected sample video according to a pre-trained motion estimation network to obtain labeled samples for labeling the endocardium of each frame of image; the intelligent detection model further includes a second training module for pre-training the motion estimation network; the pre-training includes: According to the collected sample video, use the phase correlation algorithm to calculate the motion field of the ultrasonic video; Process the sample video according to the calculated motion field to obtain a sequence of pseudo-sample pairs; Use the sequence of pseudo-sample pairs as the input of the motion estimation network, and train the motion estimation network in a way that minimizes the mean square error between the predicted motion field output by the motion estimation network and the calculated motion field.

2. The early myocardial infarction detection device according to claim 1, characterized in that, the deep contour localization network module includes: a backbone network for extracting image features of each frame of image in the ultrasonic video data; a key point detection network for outputting the coordinates of a first preset number of key points used to represent the endocardium according to the image features of each frame of image; a contour correction network for taking the coordinates of a first preset number of key points as input, calculating an offset for correction, and outputting the coordinates of a second preset number of positioning points, and representing the endocardium by a broken line segment of the second preset number of positioning points, where the second preset number is greater than the first preset number.

3. The early myocardial infarction detection device according to claim 2, characterized in that, the classification processing module is used to divide the located endocardium into predefined segments, and extract motion features according to the maximum offset of each segment of the endocardium relative to the initial frame in different frames of images; the classification processing module is further used to obtain a myocardial infarction classification result based on the pre-stored case data set and the extracted motion features by using a machine learning algorithm.

4. The early myocardial infarction detection device as claimed in claim 3, characterized in that, the machine learning algorithm includes a support vector machine algorithm or a random forest algorithm.

5. The early myocardial infarction detection device according to claim 1, characterized in that, the intelligent detection model further includes a first training module for training the deep contour localization network module.

6. The early myocardial infarction detection device according to claim 5, characterized in that, The training method for training the depth contour localization network module includes minimizing the L1 norm loss between the coordinates of the endocardial localization points output by the depth contour localization network module with the labeled samples and the labeled coordinates.

Images