Method and device for intelligent monitoring of pericardial effusion based on echocardiogram

Abstract

The application discloses a kind of based on video area information's echocardiogram pericardial effusion intelligent monitoring method and device, comprising: intercepting standard section arbitrary length echocardiogram video data;With pericardial effusion judgment model, the echocardiogram video data is identified, output whether there is pericardial effusion positive in the echocardiogram video data, and high-dimensional feature map;If there is pericardial effusion positive in the echocardiogram video data, then remove non-pericardial effusion area in echocardiogram video data, and map to high-dimensional feature map;With the activation information of high-dimensional feature map, output the responsibility frame and area of echocardiogram video center pericardial effusion;With the mapping relationship of activation high-dimensional matrix and original image, output the responsibility area heat map in echocardiogram video data original drawing.The application can automatically identify pericardial effusion in echocardiogram, automatically identify and extract responsibility frame on responsibility section, assist doctor to make judgment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, and in particular to a method and device for intelligent monitoring of pericardial effusion based on echocardiogram. BACKGROUND

[0002] Pericardial effusion is defined as an increase in fluid within the pericardial space, which can manifest as transudate, exudate, hemopericardium or pyopericardium. Under normal circumstances, there is only a small amount of fluid in the pericardial space, and if more than 50 mL of effusion is accumulated, it is defined as abnormal. The causes are very diverse, in addition to diseases that selectively affect the pericardium, other common medical conditions such as tuberculosis, hypothyroidism, end-stage renal disease or viral infection, postoperative complications can all cause pericardial effusion, and pericardial effusion can provide a lot of evidence for etiological analysis of various diseases.

[0003] The effect of pericardial effusion volume on cardiac hemodynamics is relatively complex, but generally once the effusion volume exceeds a certain threshold, a small change in fluid volume will cause a large change in pericardial space pressure, and in severe cases, it can lead to pericardial tamponade and even threaten life. On the other hand, asymptomatic pericardial effusion is usually discovered by chance through echocardiogram and CT when checking for unrelated diseases, such as a patient with acute pericarditis who complains of chest pain. Therefore, the identification of pericardial effusion is of great significance in both disease diagnosis and intraoperative monitoring.

[0004] Currently, echocardiogram is the preferred imaging method for evaluating the pericardium, which has the characteristics of easy to use, low cost, strong ability to evaluate cardiac hemodynamics, etc. In a typical ultrasound examination, the sonographer follows the clinical guidelines to manipulate the ultrasound probe to cover the patient's lesion site, and finally selects a few characteristic images to write the ultrasound report. The entire scanning video is replaced by a few representative responsibility frame images on the responsibility section. The responsibility frame image should reflect the unique morphological information of the patient's lesion site and can indicate the benign or malignant judgment. However, ultrasound diagnosis is highly dependent on the clinical experience of the sonographer. Currently, primary hospitals, community hospitals, township clinics and other primary medical institutions lack experienced sonographers, and the selection of responsibility sections relies on the personal experience and subjective judgment of the sonographer, lacking objective and quantitative standards. In clinical practice, there are many cases of missed important responsibility sections leading to misdiagnosis and missed diagnosis. SUMMARY

[0005] To solve the above technical problems, the present application provides a method and device for intelligent monitoring of pericardial effusion based on echocardiogram video region information, which does not need to label and outline the effusion in the video, saves time and effort, and can automatically identify the pericardial effusion region.

[0006] In a first aspect of the present application, a method for intelligent monitoring of pericardial effusion based on echocardiogram video region information is provided, comprising:

[0007] ultrasound echocardiogram video data of any length of time of a standard cross section;

[0008] identifying the ultrasound echocardiogram video data by using a pericardial effusion judgment model, outputting whether there is a pericardial effusion positive in the ultrasound echocardiogram video data, and a high-dimensional feature map;

[0009] If there is a pericardial effusion positive in the ultrasound echocardiogram video data, removing the non-pericardial effusion area in the ultrasound echocardiogram video data and mapping to the high-dimensional feature map to help the monitoring module more accurately locate the target position;

[0010] outputting the responsibility frame and area of the pericardial effusion in the ultrasound echocardiogram video by using the activation information of the high-dimensional feature map; and outputting the responsibility area heat map in the original image of the ultrasound echocardiogram video data by using the mapping relationship between the activated high-dimensional matrix and the original image.

[0011] In an optional embodiment, the ultrasound echocardiogram video data of any length of time of a standard cross section includes:

[0012] taking videos of the parasternal window PSAX cross section, the apical window A4C cross section, the parasternal window PLAX cross section, and the subxiphoid window SC4C cross section, and unifying the frame number, width, height, and channel number of the videos.

[0013] In an optional embodiment, the training process of the pericardial effusion judgment model includes:

[0014] taking a frame image from the video as an input image at a preset interval, so that the sampled video contains at least one cardiac cycle;

[0015] performing data augmentation processing on the input image;

[0016] training the pericardial effusion judgment model by using a stochastic gradient descent method.

[0017] In an optional embodiment, the pericardial effusion judgment model includes a skeleton part for extracting image features and a fully connected layer part for classification;

[0018] The skeleton part is a neural network based on a 3D-transformer structure. After operating the feature matrix of the last layer of the skeleton part, a high-dimensional feature map of D1xT1x(W1xH1+1) size can be obtained, where T1 is the number of retained frame channels, D1 is the encoding length of each patch in the transformer structure, and W1 and H1 are the width and height of the scaled feature map, respectively.

[0019] In an optional embodiment, the final classification result is back-propagated to the high-dimensional feature map of the second-to-last layer to obtain the gradient of the high-dimensional feature map D2xT2x(W2xH2+1) with respect to the final classification result score; wherein T2 is the time dimension, W2 and H2 are the spatial dimensions, and D2 is the abstract high-order semantic information.

[0020] After removing the dimension used for classification, the gradient is weighted and averaged in the dimension D, and the weight is the full connection layer weight corresponding to the pericardial effusion positive during the final classification, so as to obtain the average activation information in space and time.

[0021] In an optional embodiment, the removing of the non-pericardial effusion region in the echocardiogram video data comprises: after judging the section type and the corresponding anatomical structure from the average output of N frames of images, setting the region formed by connecting the centers of the detected atrium and ventricle as an image mask.

[0022] In a second aspect of the present application, an echocardiogram pericardial effusion intelligent monitoring device based on video region information is provided, comprising:

[0023] The acquisition module is configured to intercept echocardiogram video data of a standard section for any length of time.

[0024] The recognition module is configured to recognize the echocardiogram video data by using a pericardial effusion judgment model, and output whether there is pericardial effusion positive in the echocardiogram video data and a high-dimensional feature map.

[0025] The processing module is configured to remove the non-pericardial effusion region in the echocardiogram video data and map it to a high-dimensional feature map if there is pericardial effusion positive in the echocardiogram video data, to help the monitoring module to more accurately locate the target position.

[0026] The monitoring module is configured to output the responsible frame and region of the echocardiogram video pericardial effusion by using the activation information of the high-dimensional feature map, and output the responsible region heat map in the echocardiogram video data original image by using the mapping relationship between the activated high-dimensional matrix and the original image.

[0027] In an optional embodiment, the echocardiogram video data comprises:

[0028] The video of the parasternal acoustic window PSAX section, the apical acoustic window A4C section, the parasternal acoustic window PLAX section, and the subxiphoid acoustic window SC4C section is intercepted.

[0029] In a third aspect of the present application, an electronic device is provided, comprising:

[0030] At least one processor; and at least one memory connected with the processor in communication, wherein: the memory has program instructions which can be executed by the processor, the processor calling the program instructions can execute the method as described in the first aspect of the embodiment of the application.

[0031] The fourth aspect of the application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a computer to perform the method as described in the first aspect of the embodiment of the application.

[0032] The application can automatically identify pericardial effusion in echocardiogram, automatically identify and extract the responsible frame on the responsible section, and assist doctors in making judgments; and fills the technical blank of automatically identifying pericardial effusion in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 The technical roadmap of the method for monitoring pericardial effusion in echocardiogram based on video region information in the embodiment of the application.

[0034] Figure 2 The schematic diagram of the method for monitoring pericardial effusion in echocardiogram based on video region information in the embodiment of the application.

[0035] Figure 3 The schematic diagram of removing irrelevant regions according to known regions and mapping to a high-dimensional feature matrix in the embodiment of the application.

[0036] Figure 4 The schematic diagram of selection of an activation matrix and an activated region in the embodiment of the application.

[0037] Figure 5 The module diagram of the device for monitoring pericardial effusion in echocardiogram based on video region information in the embodiment of the application. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the application.

[0039] It is to be noted that the relative terms such as first and second and the like are used herein only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between such entities or operations. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by an "includes a" statement does not exclude the existence of additional elements of the same nature as those included in the process, method, article, or apparatus.

[0040] Referring to Figure 1 , the technical route of the present application is disclosed. The echocardiogram video of the original data is preprocessed to obtain a uniform format video, and then input into a pericardial effusion classification model. The uniform format video is used for training the pericardial effusion classification model after data augmentation by video frame extraction. The pericardial effusion classification model judges whether there is pericardial effusion in the echocardiogram video, and if so, an abnormal warning is issued to remind the pericardial effusion abnormality; and the responsibility frame output and the responsibility area heat map output are output to the user. The ventricular atrial detection model is also used in this technology to remove irrelevant areas in the responsibility frame and only display the pericardial effusion area.

[0041] Specifically, referring to Figure 2 , the present application provides a method for intelligent monitoring of pericardial effusion of echocardiogram based on video area information, comprising the following steps:

[0042] Step 210: intercepting echocardiogram video data of any length of standard cross section.

[0043] In this step, the video of parasternal acoustic window PSAX cross section, apical acoustic window A4C cross section, parasternal acoustic window PLAX cross section and subxiphoid acoustic window SC4C cross section is intercepted, and the frame number, width, height and channel number of the video are unified.

[0044] Step 220: using a pericardial effusion judgment model to identify the echocardiogram video data, outputting whether there is pericardial effusion positive in the echocardiogram video data, and outputting a high-dimensional feature map.

[0045] The training process of the pericardial effusion judgment model comprises:

[0046] A frame image is taken from the video as an input image at a preset interval, so that the sampled video contains at least one cardiac cycle; the input image is subjected to data augmentation processing; and the pericardial effusion judgment model is trained by using a stochastic gradient descent method.

[0047] Specifically, every x frames of images are taken from the video as input images, so that the sampled video contains 1-2 cardiac cycles; in order to increase the generalization ability of the neural network model and simulate the image displacement, scaling, deflection that occurs when the doctor scans the nodule in the actual clinical operation, random rotation, translation, flip and Gaussian noise are used on the basis of the original image. The model is trained using the stochastic gradient descent method, and the training process is to update the weight parameters and bias parameters by iterative training using the stochastic gradient descent method. In the training process, the features related to pericardial effusion in the image are learned.

[0048] Further, the pericardial effusion judgment model learns from the data whether the standard section video of the heart ultrasound contains pericardial effusion. The neural network model can be divided into a skeleton part for extracting image features and a fully connected layer part for classification. The skeleton part selects a neural network based on a 3D-transformer structure. For the feature matrix obtained by the last layer of the skeleton part, the operation is performed to obtain a high-dimensional feature map of D1xT1x(W1xH1+1) size, where T1 is the number of retained frame channels, D1 is the encoding length of each patch in the transformer structure, and W1 and H1 are the width and height of the scaled feature map, which are generally 16 or 32 times the original width and height. Because of the patch and attention structure design of the transformer, the final feature layer of each patch is obtained after processing and operation, while the position and time information of the original image is retained, so the selection of the responsibility frame and the responsibility area depends on the final high-dimensional feature map.

[0049] Step 230: If the echocardiogram video data contains pericardial effusion positive, remove the non-pericardial effusion area in the echocardiogram video data and map to a high-dimensional feature map.

[0050] In this step, if the pericardial effusion judgment model identifies that the echocardiogram video data contains pericardial effusion positive, in order to output the area of pericardial effusion in the image, the area of atrium and ventricle in the center of the image needs to be removed.

[0051] Specifically, the removal of the non-pericardial effusion area in the echocardiogram video data includes: after determining the section type and corresponding anatomical structure from the average output of the N frames of images of the input echocardiogram video, the area formed by connecting the centers of the detected atrium and ventricle is set as the image mask.

[0052] Step 240: output the responsibility frame and area of pericardial effusion in the echocardiogram video using the activation information of the high-dimensional feature map after removing the atrium and ventricle related area; output the responsibility area heat map in the echocardiogram video data original image using the mapping relationship between the activated high-dimensional matrix and the original image.

[0053] If the pericardial effusion judgment model outputs a pericardial effusion positive result, the activation information of the high-dimensional feature map is used to output the responsibility frame and area of the pericardial effusion video. The high-dimensional feature map is a high-dimensional feature map of the feature matrix extracted by the feature extraction model built in step 220, and each patch obtains the final feature layer after processing and operation, while retaining the position and time information of the original image.

[0054] There is an urgent need in the prior art for a method of automatically identifying pericardial effusion in echocardiogram, and extracting the responsibility section and responsibility frame for pericardial effusion. The present application can realize intelligent identification of pericardial effusion, and aims to fill this gap.

[0055] Further, in the present application, the pericardial effusion judgment model includes a skeleton part for extracting image features and a fully connected layer part for classification.

[0056] The skeleton part is a neural network based on a 3D-transformer structure. After operating the feature matrix of the last layer of the skeleton part, a high-dimensional feature map with a size of D1xT1x(W1xH1+1) can be obtained, where T1 is the number of retained frame channels, D1 is the encoding length of each patch in the transformer structure, W1 and H1 are the scaled feature map width and height, respectively, and are generally 16 or 32 times the original width and height.

[0057] Due to the patch and attention structure design of the transformer, each patch obtains the final feature layer after processing and operation, while retaining the position and time information of the original image, so the selection of the subsequent responsibility frame and responsibility area depends on the final high-dimensional feature map.

[0058] Pericardial effusion appears in the periphery of the image on all echocardiogram sections, and the central area is generally dominated by the left atrium, left ventricle, right atrium, right ventricle, or various valves and large blood vessels, so this information can be used when searching for the responsibility area. The pericardial effusion judgment model in the present application focuses on the parasternal long axis, short axis, apical four-chamber, subxiphoid four-chamber, and other standard echocardiogram standard windows. The preprocessed video in step 210 is read into N frames of images and input into the above-mentioned model. After comprehensively judging the section type and corresponding anatomical structure from the average output of the N frames of images, the area formed by connecting the detected atrial and ventricular centers is set as an image mask, i.e. the area unrelated to pericardial effusion in the image is removed, and the myocardial outer wall and pericardium target area are retained; if the model judges that the video is a non-standard section, this step can be skipped.

[0059] Further, the final classification result output by the classification model in the pericardial effusion judgment model is back propagated to the high-dimensional feature map of the second-to-last layer to obtain the gradient of the high-dimensional feature map D2xT2x(W2xH2+1) with respect to the score; wherein T2 is the time dimension, W2 and H2 are the spatial dimensions, and D2 is the abstract high-order semantic information; after removing the dimensions used for classification, the gradient is weighted and averaged in the dimension D, and the weight is the full connection layer weight corresponding to the pericardial effusion positive in the final classification, so as to obtain the average activation information in space and time.

[0060] For example, the gradient in the dimension D is weighted and averaged, the weight is the full connection layer weight corresponding to the pericardial effusion positive in the final classification, so as to obtain the average activation information in space and time, and after removing the dimensions used for classification, the dimensions of the activation high-dimensional matrix are T2x(W2xH2).

[0061] Since the transformer structure is based on the operation between patches, the mapping relationship between the activation high-dimensional matrix and the original image can be utilized. The T-layer activation high-dimensional matrix is linearly mapped to the corresponding frames of the original video, for example, T=8, and the original input frame is 16 frames, then the 1-8 layers of the activation high-dimensional matrix correspond to the 1, 3, 5…15 frames of the original video respectively, and W and H are the same. The region unrelated to the pericardial effusion has been obtained in step 4, according to the mapping, the corresponding gradient of the activation high-dimensional matrix is set to 0, and the irrelevant region is removed, as shown in Figure 3 According to the average activation value in the dimension T of the activation high-dimensional matrix, the responsibility frame sorting is obtained and the maximum responsibility frame is selected; and after the activation map in the dimension T of the activation matrix is transformed into the size of the original image by difference, the responsibility region heat map in the original image is output, as shown in Figure 4 .

[0062] As shown in Figure 5 The application provides a device for intelligent monitoring of pericardial effusion in echocardiogram based on video region information, characterized in that it comprises:

[0063] The acquisition module 51 is configured to intercept the echocardiogram video data of the standard cross section for any length of time;

[0064] The identification module 52 is configured to identify the echocardiogram video data by using the pericardial effusion judgment model, output whether there is pericardial effusion positive in the echocardiogram video data, and output the high-dimensional feature map;

[0065] The processing module 53 is configured to remove the non-pericardial effusion region in the echocardiogram video data and map it to the high-dimensional feature map if there is pericardial effusion positive in the echocardiogram video data, so as to help the monitoring module to accurately locate the target position;

[0066] The monitoring module 54 is used for outputting the responsibility frame and area of the pericardial effusion in the echocardiogram video by using the activation information of the high-dimensional feature map of the atrial ventricular related area removed; and outputting the responsibility area heat map in the echocardiogram video data original map by using the mapping relationship between the activation high-dimensional matrix and the original image.

[0067] The echocardiogram video data comprises the video of the parasternal acoustic window PSAX section, the apical acoustic window A4C section, the parasternal acoustic window PLAX section and the subxiphoid acoustic window SC4C section.

[0068] The application further provides an electronic device, comprising:

[0069] at least one processor; and at least one memory connected with the processor, wherein: the memory stores program instructions executable by the processor, and the processor calling the program instructions can execute the method for intelligent monitoring of pericardial effusion in echocardiogram based on video area information.

[0070] The application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method for intelligent monitoring of pericardial effusion in echocardiogram based on video area information.

[0071] It can be understood that the computer readable storage medium can include any entity or device capable of carrying the computer program, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM) and software distribution medium, etc. The computer program includes computer program code. The computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer readable storage medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM) and software distribution medium, etc.

[0072] In some embodiments of the present application, the device can comprise a controller, which is a single-chip microcontroller integrating a processor, memory, communication module, etc. The processor can refer to the processor contained in the controller. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.

[0073] Any process or method described in flow charts or otherwise described herein can be understood as representing code modules, segments, or portions of code that include one or more executable instructions for implementing specific logic functions (or steps) of the process, and the various embodiments of the present application can include additional or fewer steps performing the same or equivalent functions, in combination with or in place of, the steps described in this specification, as would be understood by those of ordinary skill in the art. The various embodiments of the present application can also modify the order of the steps described in this specification. Thus, a process or method that "comprises" one or more steps corresponds to having one or more of the steps included in the process or method, and vice versa.

[0074] Those of ordinary skill in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the above description in a general manner. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0075] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for intelligent monitoring of pericardial effusion using echocardiography based on video region information, characterized in that, include: Extract echocardiographic video data of arbitrary duration from a standard section; The pericardial effusion detection model is used to identify the echocardiographic video data, and outputs whether there is positive pericardial effusion in the echocardiographic video data, as well as a high-dimensional feature map; If pericardial effusion is positive in the echocardiogram video data, the non-pericardial effusion region in the echocardiogram video data is removed and mapped to a high-dimensional feature map; The activation information of the high-dimensional feature map is used to output the responsible frame and region of pericardial effusion in the echocardiographic video; The mapping relationship between the activation high-dimensional matrix and the original image is used to output the heat map of the responsible region in the original echocardiogram video data. The training process of the pericardial effusion diagnosis model includes: Take one frame from the video at preset intervals as the input image, so that the sampled video contains at least one cardiac cycle; Perform data augmentation processing on the input image; The pericardial effusion diagnosis model was trained using stochastic gradient descent. The pericardial effusion assessment model includes a skeleton part for extracting image features and a fully connected layer part for classification. The skeleton part is a neural network based on a 3D-transformer structure. After operating on the feature matrix of the last layer of the skeleton part, a high-dimensional feature map of size D1×T1×(W1×H1+1) can be obtained, where T1 is the number of frame channels retained, D1 is the encoding length of each patch in the transformer structure, and W1 and H1 are the width and height of the scaled feature map, respectively. The final classification result is backpropagated to the high-dimensional feature map of the penultimate layer to obtain the gradient of the high-dimensional feature map D2×T2×(W2×H2+1) with respect to the final classification result score; where T2 is the time dimension, W2 and H2 are the spatial dimensions, and D2 is the abstract high-order semantic information; After removing the extra dimension used for classification, the gradient is weighted and averaged in dimension D, with the weights being the weights of the fully connected layer corresponding to positive pericardial effusion during the final classification, thus obtaining the average activation information in space and time.

2. The method for intelligent monitoring of pericardial effusion based on video region information using echocardiography according to claim 1, characterized in that, The echocardiographic video data of arbitrary duration extracted from the standard section includes: Videos from the parasternal acoustic window PSAX section, apical acoustic window A4C section, parasternal acoustic window PLAX section, and subxiphoid acoustic window SC4C section are extracted, and the frame number, width, height, and number of channels of the video are unified.

3. The method for intelligent monitoring of pericardial effusion based on video region information using echocardiography according to claim 1, characterized in that, The process of removing non-pericardial effusion areas from echocardiogram video data includes: after comprehensively determining the section type and corresponding anatomical structure from the average output of N frames of images, setting the area formed by connecting the centers of the detected atria and ventricles as an image mask.

4. A device for intelligent monitoring of pericardial effusion based on video region information using echocardiography, characterized in that, include: The acquisition module is used to extract echocardiographic video data of arbitrary duration from a standard cross-section; The identification module is used to identify the echocardiogram video data using a pericardial effusion judgment model, and output whether there is a positive result for pericardial effusion in the echocardiogram video data, as well as a high-dimensional feature map; The processing module is used to remove the non-pericardial effusion area from the echocardiogram video data if pericardial effusion is positive in the echocardiogram video data, and map it to a high-dimensional feature map to help the monitoring module more accurately locate the target location. The monitoring module is used to output the responsible frame and region of pericardial effusion in the echocardiogram video using the activation information of the high-dimensional feature map; and to output the heat map of the responsible region in the original echocardiogram video data using the mapping relationship between the activation high-dimensional matrix and the original image. The training process of the pericardial effusion diagnosis model includes: Take one frame from the video at preset intervals as the input image, so that the sampled video contains at least one cardiac cycle; Perform data augmentation processing on the input image; The pericardial effusion diagnosis model was trained using stochastic gradient descent. The pericardial effusion assessment model includes a skeleton part for extracting image features and a fully connected layer part for classification. The skeleton part is a neural network based on a 3D-transformer structure. After operating on the feature matrix of the last layer of the skeleton part, a high-dimensional feature map of size D1×T1×(W1×H1+1) can be obtained, where T1 is the number of frame channels retained, D1 is the encoding length of each patch in the transformer structure, and W1 and H1 are the width and height of the scaled feature map, respectively. The final classification result is backpropagated to the high-dimensional feature map of the penultimate layer to obtain the gradient of the high-dimensional feature map D2×T2×(W2×H2+1) with respect to the final classification result score; where T2 is the time dimension, W2 and H2 are the spatial dimensions, and D2 is the abstract high-order semantic information; After removing the extra dimension used for classification, the gradient is weighted and averaged in dimension D, with the weights being the weights of the fully connected layer corresponding to positive pericardial effusion during the final classification, thus obtaining the average activation information in space and time.

5. The device for intelligent monitoring of pericardial effusion based on video region information using echocardiography according to claim 4, characterized in that, Echocardiographic video data includes: Videos of the parasternal acoustic window PSAX section, the apical acoustic window A4C section, the parasternal acoustic window PLAX section, and the subxiphoid acoustic window SC4C section were extracted.

6. An electronic device, characterized in that, include: At least one processor; And at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor invokes the program instructions to perform the method as described in any one of claims 1 to 3.

7. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a computer, performs the method as described in any one of claims 1 to 3.

Images