Configuration method and device of syncope prediction model and processing equipment

Abstract

The invention provides a configuration method and device of a syncope prediction model and processing equipment, which are used for creating the syncope prediction model aiming at the symptom of syncope, so that long-range electrocardiogram data in a normal time period can be used as input through the syncope prediction model, and the syncope prediction model can be used for predicting the syncope. According to the method, whether the tested user is syncope or not and the specific syncope type are accurately predicted, good data support is provided for syncope diagnosis and treatment in the aspect of decision making, and better diagnosis and treatment can be achieved. The method comprises the steps of obtaining sample electrocardiogram data; a corresponding labeling result is configured for the sample electrocardiogram data, the labeling result is used for indicating whether a user corresponding to the sample electrocardiogram data has a syncope condition or not, the labeling result is further used for indicating the syncope type of the user corresponding to the sample electrocardiogram data under the syncope condition, and the syncope type comprises a heart suppression type, a blood vessel suppression type and a mixed type; and training a syncope prediction model based on the labeled sample electrocardiogram data.

Description

Technical Field

[0001] The present application relates to the field of information, and in particular to a configuration method, device and processing equipment for a syncope prediction model. Background Art

[0002] Syncope is a common clinical condition, and the diagnosis or detection of syncope can usually be accomplished through an upright tilt test.

[0003] The upright tilt test is used to stimulate and diagnose vasovagal syncope by adjusting the angle of inclination to make the patient passively tilt. In the specific test process, because the sensitivity of the basic tilt test is low, sublingual nitroglycerin can also better stimulate the disease, and the electrocardiogram can be used to observe whether the patient has syncope. Therefore, this scheme is often used clinically to increase the diagnostic positive rate and provide objective evidence for unexplained syncope.

[0004] From the above content, it can be found that the syncope detection scheme based on the upright tilt test has complex operating costs, and the effective detection of syncope requires medical staff to observe the patient's syncope in real time. Obviously, there are obvious limitations in detection efficiency and accuracy. Summary of the invention

[0005] The present application provides a configuration method, device and processing equipment of a syncope prediction model, which is used to create a syncope prediction model for the disease of syncope. In this way, the syncope prediction model can use long-term electrocardiogram data in a normal time period as input to accurately predict whether the user being tested will experience syncope and the specific type of syncope, thereby providing good data support for syncope diagnosis and treatment in decision-making, and thus helping to achieve better diagnosis and treatment.

[0006] In a first aspect, the present application provides a method for configuring a syncope prediction model, the method comprising:

[0007] Acquire sample electrocardiogram data, wherein the sample electrocardiogram data is specifically long-term electrocardiogram data within a normal time period;

[0008] Configuring corresponding annotation results for the sample electrocardiogram data, wherein the annotation results are used to indicate whether the user corresponding to the sample electrocardiogram data has experienced syncope, and the annotation results are also used to indicate the syncope type when the user corresponding to the sample electrocardiogram data has experienced syncope, and the syncope types include cardiac inhibition type, vascular inhibition type, and mixed type;

[0009] A syncope prediction model is trained based on the labeled sample ECG data, wherein the syncope prediction model is used to predict whether the user will experience syncope within a preset time period in the future and the specific syncope type if syncope occurs based on the ECG data input into the model.

[0010] In combination with the first aspect of the present application, in a first possible implementation of the first aspect of the present application, in a one-dimensional model configuration scheme, the labeling result uses 0 to represent a normal type, 1 to represent a cardiac inhibition type, 2 to represent a vascular inhibition type, and 3 to represent a mixed type;

[0011] The syncope prediction result of the syncope prediction model is a one-dimensional array with a length of 4, and the 4 output numbers represent the probabilities of normal type, cardiac inhibition type, vascular inhibition type and mixed type respectively.

[0012] In combination with the first possible implementation of the first aspect of the present application, in a second possible implementation of the first aspect of the present application, the sample electrocardiogram data is specifically the electrocardiogram data measured by the second lead, and before training the syncope prediction model based on the labeled sample electrocardiogram data, the method further includes:

[0013] The first two minutes of sample ECG data are discarded to obtain the remaining data;

[0014] Based on the sliding window method, the remaining data is cut into multiple segments according to the cutting unit of 7200 data length corresponding to ten minutes. Moreover, for different types of sample ECG data, each normal user randomly retains 6 segments of data, each cardiac inhibition user retains all data, each vascular inhibition user and each mixed user randomly retains 24 segments of data, thus completing the update of the sample ECG data;

[0015] The syncope prediction model is specifically configured with input data of a tensor with a shape of [n, 1, 72000] and a label corresponding to each sequence segment, where n is the number of sequence segments.

[0016] In combination with the second possible implementation of the first aspect of the present application, in a third possible implementation of the first aspect of the present application, the syncope prediction model is specifically a one-dimensional variant network based on VGG-16, and the model architecture of the syncope prediction model has the following configuration content:

[0017] The number of convolution kernels in the first convolution layer is set to 16, and the number of convolution kernels doubles after each pooling;

[0018] After the two front-to-back convolutional layers, connect a pooling layer, then continue to connect two front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then continue to connect two front-to-back fully-connected layers configured with ReLU functions, and then continue with a fully-connected layer configured with Softmax function.

[0019] In a fourth possible implementation of the first aspect of the present application, in the two-dimensional model configuration scheme, in the annotation result, the label assigned to the normal type is [0,0], the label assigned to the cardiac inhibition type is [1,0], the label assigned to the vascular inhibition type is [0,1], and the label assigned to the mixed type is [1,1];

[0020] The syncope prediction result of the syncope prediction model includes two predicted probability values. If the first probability value is greater than 0.5, the output is 1, and if the second predicted value is greater than 0.5, the output is 1.

[0021] In combination with the fourth possible implementation of the first aspect of the present application, in a fifth possible implementation of the first aspect of the present application, the syncope prediction model is specifically a two-dimensional variant network based on VGG-16, and the model architecture of the syncope prediction model has the following configuration content:

[0022] After the two convolutional layers with 64 convolution kernels connected front and back, connect a pooling layer, then continue to connect two convolutional layers with 128 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 256 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 512 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 512 convolution kernels connected front and back, then continue to connect three fully connected layers configured with ReLU functions, then continue with a Sigmoid function, and all convolutional layers are configured with ReLU functions.

[0023] In combination with the first aspect of the present application, in a sixth possible implementation manner of the first aspect of the present application, the method further includes:

[0024] Get the current user's electrocardiogram data;

[0025] Inputting the electrocardiogram data of the current user into the syncope prediction model to perform corresponding syncope prediction processing;

[0026] extracting the syncope prediction processing result output by the syncope prediction model;

[0027] Based on the syncope prediction processing result, a syncope intervention measure reminder is output to the current user.

[0028] In a second aspect, the present application provides a device for configuring a syncope prediction model, the device comprising:

[0029] An acquisition unit, used for acquiring sample electrocardiogram data, wherein the sample electrocardiogram data is specifically long-term electrocardiogram data in a normal time period;

[0030] a configuration unit, configured to configure corresponding annotation results for the sample electrocardiogram data, wherein the annotation results are used to indicate whether the user corresponding to the sample electrocardiogram data has a syncope, and the annotation results are also used to indicate the syncope type when the user corresponding to the sample electrocardiogram data has a syncope, and the syncope types include cardiac inhibition type, vascular inhibition type and mixed type;

[0031] A training unit is used to train a syncope prediction model based on labeled sample ECG data, wherein the syncope prediction model is used to predict whether the user will experience syncope within a preset time period in the future according to the ECG data input by the model, as well as the specific syncope type if syncope occurs.

[0032] In combination with the second aspect of the present application, in a first possible implementation of the second aspect of the present application, in a one-dimensional model configuration scheme, the labeling result uses 0 to represent a normal type, 1 to represent a cardiac inhibition type, 2 to represent a vascular inhibition type, and 3 to represent a mixed type;

[0033] The syncope prediction result of the syncope prediction model is a one-dimensional array with a length of 4, and the 4 output numbers represent the probabilities of normal type, cardiac inhibition type, vascular inhibition type and mixed type respectively.

[0034] In combination with the first possible implementation of the second aspect of the present application, in a second possible implementation of the second aspect of the present application, the device further includes a preprocessing unit, which is used to:

[0035] The first two minutes of sample ECG data are discarded to obtain the remaining data;

[0036] Based on the sliding window method, the remaining data is cut into multiple segments according to the cutting unit of 7200 data length corresponding to ten minutes. Moreover, for different types of sample ECG data, each normal user randomly retains 6 segments of data, each cardiac inhibition user retains all data, each vascular inhibition user and each mixed user randomly retains 24 segments of data, thus completing the update of sample ECG data;

[0037] The syncope prediction model is specifically configured with input data of a tensor with a shape of [n, 1, 72000] and a label corresponding to each sequence segment, where n is the number of sequence segments.

[0038] In combination with the second possible implementation of the second aspect of the present application, in a third possible implementation of the second aspect of the present application, the syncope prediction model is specifically a one-dimensional variant network based on VGG-16, and the model architecture of the syncope prediction model has the following configuration content:

[0039] The number of convolution kernels in the first convolution layer is set to 16, and the number of convolution kernels doubles after each pooling;

[0040] After the two front-to-back convolutional layers, connect a pooling layer, then continue to connect two front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then continue to connect two front-to-back fully-connected layers configured with ReLU functions, and then continue with a fully-connected layer configured with Softmax function.

[0041] In a fourth possible implementation manner of the second aspect of the present application, in the two-dimensional model configuration scheme, in the annotation result, the label assigned to the normal type is [0,0], the label assigned to the cardiac inhibition type is [1,0], the label assigned to the vascular inhibition type is [0,1], and the label assigned to the mixed type is [1,1];

[0042] The syncope prediction result of the syncope prediction model includes two predicted probability values. If the first probability value is greater than 0.5, the output is 1, and if the second predicted value is greater than 0.5, the output is 1.

[0043] In combination with the fourth possible implementation of the second aspect of the present application, in a fifth possible implementation of the second aspect of the present application, the syncope prediction model is specifically a two-dimensional variant network based on VGG-16, and the model architecture of the syncope prediction model has the following configuration content:

[0044] After the two convolutional layers with 64 convolution kernels connected front and back, connect a pooling layer, then continue to connect two convolutional layers with 128 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 256 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 512 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 512 convolution kernels connected front and back, then continue to connect three fully connected layers configured with ReLU functions, then continue with a Sigmoid function, and all convolutional layers are configured with ReLU functions.

[0045] In combination with the second aspect of the present application, in a sixth possible implementation manner of the second aspect of the present application, the apparatus further includes an application unit, configured to:

[0046] Get the current user's electrocardiogram data;

[0047] Inputting the electrocardiogram data of the current user into the syncope prediction model to perform corresponding syncope prediction processing;

[0048] extracting the syncope prediction processing result output by the syncope prediction model;

[0049] Based on the syncope prediction processing result, a syncope intervention measure reminder is output to the current user.

[0050] In a third aspect, the present application provides a processing device, including a processor and a memory, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the method provided in the first aspect of the present application or any possible implementation method of the first aspect of the present application is executed.

[0051] In a fourth aspect, the present application provides a computer-readable storage medium, which stores multiple instructions, and the instructions are suitable for a processor to load to execute the method provided by the first aspect of the present application or any possible implementation of the first aspect of the present application.

[0052] From the above content, it can be concluded that the present application has the following beneficial effects:

[0053] For the disease of syncope, after obtaining the sample ECG data, the present application configures corresponding annotation results for the sample ECG data, wherein the annotation results are used to indicate whether the user corresponding to the sample ECG data has experienced syncope, and the annotation results are also used to indicate the type of syncope when the user corresponding to the sample ECG data has experienced syncope, and the syncope types include cardiac inhibition type, vascular inhibition type and mixed type. Then, based on the annotated sample ECG data, a syncope prediction model is trained, wherein the syncope prediction model is used to predict whether the user corresponding to the ECG data input into the model will experience syncope within a preset time period in the future and the specific type of syncope when syncope occurs based on the ECG data input into the model. Under this processing mechanism, the present application has created a syncope prediction model, so that the syncope prediction model can use long-term ECG data within a normal time period as input to accurately predict whether the user being tested will experience syncope and the specific type of syncope, thereby providing good data support for syncope diagnosis and treatment in terms of decision-making, and thus helping to achieve better diagnosis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0055] Figure 1 A schematic diagram of a flow chart of a method for configuring a syncope prediction model of the present application;

[0056] Figure 2 A schematic diagram of a network architecture of a one-dimensional convolutional neural network of this application;

[0057] Figure 3 A schematic diagram of a network architecture of a two-dimensional convolutional neural network for this application;

[0058] Figure 4 A structural schematic diagram of a configuration device for a syncope prediction model of the present application;

[0059] Figure 5 A schematic diagram of the structure of the processing equipment of this application. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0061] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in a sequence other than that illustrated or described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products or devices. The naming or numbering of steps in this application does not mean that the steps in the method flow must be executed in the time / logical sequence indicated by the naming or numbering. The process steps that have been named or numbered can change the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0062] The division of modules in this application is a logical division. There may be other division methods when it is implemented in actual applications. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection between modules can be electrical or other similar forms, which are not limited in this application. In addition, the modules or submodules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed in multiple circuit modules, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present application.

[0063] Before introducing the configuration method of the syncope prediction model provided by the present application, the background content involved in the present application is first introduced.

[0064] The configuration method, device and computer-readable storage medium of the syncope prediction model provided in the present application can be applied to a processing device to create a syncope prediction model for the disease of syncope. In this way, the syncope prediction model can use long-term electrocardiogram data in a normal time period as input to accurately predict whether the user being tested will experience syncope and the specific type of syncope, thereby providing good data support for syncope diagnosis and treatment in decision-making, and thus helping to achieve better diagnosis and treatment.

[0065] The configuration method of the syncope prediction model mentioned in the present application can be executed by a configuration device of the syncope prediction model, or a server, a physical host, or a user equipment (UE) or other different types of processing devices integrated with the configuration device of the syncope prediction model. Among them, the configuration device of the syncope prediction model can be implemented in hardware or software, and the UE can be a terminal device such as a smart phone, a tablet computer, a laptop computer, a desktop computer, or a personal digital assistant (PDA), and the processing device can be set in a device cluster.

[0066] It can be understood that the solution of the present application mainly involves the training and processing of the syncope prediction model. Generally speaking, it only needs to have the corresponding data processing capabilities. Therefore, the processing equipment that executes the configuration method of the syncope prediction model of the present application, or the processing equipment equipped with the application service corresponding to the configuration method of the syncope prediction model of the present application, its device type and device deployment form are relatively flexible.

[0067] If it also involves the real-time collection and processing of electrocardiogram data, or the application of syncope prediction models in clinical work, it is easy to understand that the equipment type and equipment deployment form of the processing equipment can be adaptively adjusted as the application scenario changes.

[0068] As an example, the processing equipment can be configured in the form of a device cluster, which includes an electrocardiograph for collecting electrocardiograph data, a central device located in the background for training the model, and a terminal device located on the doctor's side responsible for model application.

[0069] Next, the configuration method of the syncope prediction model provided by the present application is introduced.

[0070] First, see Figure 1 , Figure 1 A flow chart of a method for configuring a syncope prediction model of the present application is shown. The method for configuring a syncope prediction model provided by the present application may specifically include the following steps S101 to S103:

[0071] Step S101, obtaining sample electrocardiogram data, wherein the sample electrocardiogram data is specifically long-term electrocardiogram data in a normal time period;

[0072] It can be understood that the configuration target of the syncope prediction model using electrocardiogram data as input may involve the configuration work of the training samples.

[0073] In this regard, the present application can obtain sample electrocardiogram data.

[0074] The sample ECG data may be ready-made data pulled from the hospital's relevant online system, ready-made data manually entered by staff, data obtained by transformation / secondary processing based on ready-made data, or directly generated data. All of these are possible in actual situations.

[0075] It is worth noting that the sample ECG data is the ECG data after stripping or excluding the content related to the patient's personal privacy, and only retains the core content of the ECG related to syncope, so there will be no privacy issues.

[0076] In addition, it is also important to note that the sample ECG data is usually not the ECG data obtained by the upright tilt test, but the long-term ECG data during the normal time period. It is the ECG data collected for a long time under the patient's usual / daily / normal state. More specifically, the state here refers to the state where the patient is not affected by the examination / treatment such as the upright tilt test. For example, in some cases, the data of the first 10 minutes of the upright tilt test may also meet the requirements.

[0077] Step S102, configuring corresponding annotation results for the sample electrocardiogram data, wherein the annotation results are used to indicate whether the user corresponding to the sample electrocardiogram data has syncope, and the annotation results are also used to indicate the syncope type when the user corresponding to the sample electrocardiogram data has syncope, and the syncope types include cardiac inhibition type, vascular inhibition type and mixed type;

[0078] It can be understood that for the training samples used to train the model, it is necessary to add / assign annotations / labels that the model can recognize as a true value to guide the training of the model.

[0079] In this regard, after obtaining the sample electrocardiogram data, the present application can perform annotation processing on it, so as to configure corresponding annotation results for the sample electrocardiogram data.

[0080] During the annotation process, it can be completed manually or using corresponding automated annotation tools. The automated annotation tools involve pre-configured automated annotation logic / strategies when performing automated annotation processing. Both types of annotation methods are possible. Considering that the annotation processing can directly use existing technologies, no specific explanation will be given this time.

[0081] It is worth noting that the present application specifically involves the prediction and processing of three types of syncope, namely cardiac inhibition type, vascular inhibition type and mixed type (the mixed type has the characteristics of both cardiac inhibition type and vascular inhibition type). It can be understood that these three types of syncope are clinically clear syncope types. It can be seen that the syncope prediction model that is subsequently trained can effectively make accurate predictions on these three types of syncope in terms of details, and provide more delicate data support and decision support.

[0082] Step S103, training a syncope prediction model based on the labeled sample ECG data, wherein the syncope prediction model is used to predict whether the user will experience syncope within a preset time period in the future and the specific syncope type if syncope occurs based on the ECG data input into the model.

[0083] It can be understood that the syncope prediction model configured in the present application is itself a machine learning model, usually a neural network model, which can use powerful artificial intelligence (AI) performance to complete the syncope prediction process designed in the present application.

[0084] In the training process based on the labeled training samples, i.e., sample ECG data, the following processing is usually included:

[0085] In each round of training, a group of training samples are input into the model, so that the model can perform corresponding syncope prediction processing to realize forward propagation. Then, based on the syncope prediction results output by the model, the loss function is calculated in combination with the annotation results of the training samples. The model parameters are optimized according to the calculation results of the loss function to realize reverse propagation. In this way, when the preset model training requirements such as prediction accuracy, number of trainings or training time are met, the model training can be completed and a syncope prediction model that can be put into practical use can be obtained.

[0086] Among them, for the specific model architecture adopted (such as VGG-16 model, ResNet, etc.), model training architecture (for example, the samples can be divided into training set and verification set in a ratio of 7:3, the training set is used to train the model, and the verification set is used to verify the model training effect) and loss function, it can be understood that the existing solution can be adopted, or further optimization and improvement can be made on the basis of the existing solution, or a novel solution can be developed independently. These are all possible in actual situations and can be adjusted according to actual needs.

[0087] In this way, a syncope prediction model can be obtained through the model training processing here, which is used to predict whether the user will experience syncope within a preset time period in the future and the specific type of syncope if syncope occurs, providing strong data support for clinical medical staff for the possibility of syncope and syncope symptoms of patients.

[0088] Corresponding to the application aspects of the model, the application method may include:

[0089] Step S104, obtaining the electrocardiogram data of the current user;

[0090] Step S105, inputting the electrocardiogram data of the current user into the syncope prediction model to perform corresponding syncope prediction processing;

[0091] Step S106, extracting the syncope prediction processing result output by the syncope prediction model;

[0092] Step S107: Based on the syncope prediction processing result, output a syncope intervention measure reminder to the current user.

[0093] It can be understood that the input of the model in the actual application process, that is, the electrocardiogram data of the current user, corresponds to / aligns with the sample electrocardiogram data in data format.

[0094] After the syncope prediction model has performed the currently required syncope prediction processing, it is obvious that the syncope prediction processing result output by the model can be extracted, and the syncope prediction processing result can be processed according to the preset data usage strategy.

[0095] For example, it can be stored, forwarded, and output a prompt for completed processing, output a pre-adapted syncope diagnosis / treatment plan (which can be further subdivided according to the type of syncope), or perform further analysis and processing. These are all possible, and the specific strategy content can be configured according to clinical work and actual conditions.

[0096] For example, the present application may specifically relate to a reminder mechanism for syncope intervention measures. When syncope is predicted, the user may be directly or indirectly prompted to perform the Valsava maneuver, take a deep breath and hold it, sit down, lie down, drink water to divert attention, register or go to the emergency room, etc., so as to improve symptoms or treat the condition for syncope that has occurred or is about to occur.

[0097] Among them, corresponding to the reminder mechanism here, the processing device of this application can output reminders on the current user side in different specific reminder modes such as text, image, voice, light or vibration through its own equipment (for example, it can include monitoring instruments such as wearable devices) or through external equipment (for example, it can be connected to monitoring instruments such as wearable devices). These settings can obviously be more flexible in actual situations and can be adjusted according to actual needs. Of course, the specific reminder output can also be completed manually. The processing device only needs to give the specific reminder content required for medical staff to remind the current user of syncope intervention measures, which is also possible in actual situations.

[0098] The specific reminder content outputted may be general reminder content, reminder content adapted to the specific syncope prediction type / situation, or personalized / customized reminder content, which may also be configured according to actual needs.

[0099] from Figure 1It can be seen from the illustrated embodiment that for the disease of syncope, after acquiring the sample ECG data, the present application configures corresponding annotation results for the sample ECG data, wherein the annotation results are used to indicate whether the user corresponding to the sample ECG data has experienced syncope, and the annotation results are also used to indicate the type of syncope when the user corresponding to the sample ECG data has experienced syncope, and the syncope types include cardiac inhibition type, vascular inhibition type and mixed type. Then, based on the annotated sample ECG data, a syncope prediction model is trained, wherein the syncope prediction model is used to predict whether the user corresponding to the ECG data input into the model will experience syncope within a preset time period in the future and the specific type of syncope when syncope occurs, based on the ECG data input into the model. Under this processing mechanism, the present application has created a syncope prediction model, so that the syncope prediction model can use the long-term ECG data within a normal time period as input to accurately predict whether the user under test will experience syncope and the specific type of syncope, thereby providing good data support for syncope diagnosis and treatment in terms of decision-making, and thus helping to achieve better diagnosis and treatment.

[0100] Continue to the above Figure 1 Each step of the illustrated embodiment and possible implementation methods thereof in practical applications are described in detail.

[0101] Regarding the processing architecture of the syncope prediction model of the present application, the present application also provides two types of configuration schemes from the specific operation level, specifically, it can involve a one-dimensional model configuration scheme and a two-dimensional model configuration scheme.

[0102] The so-called one-dimensional model configuration scheme means that the model output is specifically configured as a one-dimensional data format. Similarly, the two-dimensional model configuration scheme corresponds to the model output in a two-dimensional data format.

[0103] Next, we will further explain these two types of configuration solutions.

[0104] As an exemplary implementation, in a one-dimensional model configuration scheme, the labeling results use 0 to represent the normal type, 1 to represent the cardiac inhibition type, 2 to represent the vascular inhibition type, and 3 to represent the mixed type.

[0105] Correspondingly, the syncope prediction result of the syncope prediction model is a one-dimensional array of length 4, and the four output numbers represent the probabilities of normal type, cardiac inhibition type, vascular inhibition type and mixed type, respectively.

[0106] In this way, through a one-dimensional array of length 4, the prediction results of the syncope prediction model on whether syncope exists and the specific syncope type can be conveniently identified, and on this basis, a final syncope prediction result can be fused, and the maximum probability value and its corresponding specific syncope type are taken as output.

[0107] In addition, in order to ensure processing accuracy and efficiency, this application also makes more specific restrictions on the input format of the syncope prediction model.

[0108] (1) The sample electrocardiogram data is specifically the electrocardiogram data measured by lead II.

[0109] It can be understood that the so-called electrocardiogram data may involve or include electrocardiogram data measured by different leads. Taking a 12-lead electrocardiograph as an example, the electrocardiogram data measured by the 12-lead electrocardiograph is 12-lead electrocardiogram data.

[0110] However, the present application believes that some leads have outstanding contributions to the goal of syncope prediction, while some leads do not have outstanding contributions to the goal of syncope prediction and may even cause interference.

[0111] In this regard, this application specifically focuses on the II lead of the electrocardiogram data involved in the solution. The corresponding electrocardiograph only needs to be able to collect the II lead electrocardiogram data, and there is no need to limit the lead format of the specific electrocardiograph.

[0112] In this way, in actual operation, the amount of data involved can be highly simplified, taking into account both processing efficiency and processing accuracy, and it can also help reduce the training cost of the model or the configuration cost of the model training work. In addition, in the subsequent model application link, it can also help to achieve lower application costs. For example, the electrocardiograph can only collect the electrocardiograph data of the second lead, and other leads do not need to work, which greatly simplifies the workload of medical staff responsible for measuring electrocardiograms. At the same time, it can also greatly reduce the storage / transmission cost of the model input data, and the model can carry out more efficient and high-precision syncope prediction processing.

[0113] (2) Taking the training sample as an example, based on the labeled sample electrocardiogram data, before training the syncope prediction model, the method of the present application may also include:

[0114] The first two minutes of sample ECG data are discarded to obtain the remaining data;

[0115] Based on the sliding window method, the remaining data is cut into multiple segments according to the cutting unit of 7200 data length corresponding to ten minutes. In addition, for different types of sample ECG data, each normal user randomly retains 6 segments of data, each cardiac inhibition user retains all segments, each vascular inhibition user and each mixed user randomly retains 24 segments of data to complete the update of the sample ECG data.

[0116] It can be understood that the configuration of the embodiment here is a pre-processing means.

[0117] The present application believes that, in actual situations, the electrocardiogram data measured in the first two minutes may contain blanks or interference, and therefore they are discarded / eliminated.

[0118] In addition, in order to simplify the input data and take into account both training quality and training efficiency, the ECG data can be cut / split / segmented / truncated through a sliding window method based on a cutting unit of 7200 data length corresponding to ten minutes to obtain several segments of high-quality ECG data.

[0119] In addition, in order to further enhance the training effect, taking into account the large difference in sample size of each type of syncope in actual situations, and also considering the possibility of facing different types of syncope in the actual prediction process of the model under balance, the data amount of each sample user, that is, the number of segments, is limited based on it to balance the sample size differences between different types of syncope.

[0120] (3) Under the above settings, the syncope prediction model can be configured as a tensor with an input shape of [n, 1, 72000] and a label corresponding to each sequence segment (the label is consistent with the label before segmentation), where n is the number of sequence segments.

[0121] In addition, under the one-dimensional model configuration scheme, this application also provides a practical configuration scheme from the perspective of specific model architecture.

[0122] For details, see Figure 2 A schematic diagram of a network architecture of a one-dimensional convolutional neural network of the present application is shown. As an exemplary embodiment, the syncope prediction model can be a one-dimensional variant network based on VGG-16. Correspondingly, the model architecture of the syncope prediction model has the following configuration content:

[0123] The number of convolution kernels in the first convolution layer is set to 16, and the number of convolution kernels doubles after each pooling;

[0124] The two convolutional layers connected before and after ( Figure 2 Convolution) followed by a pooling layer ( Figure 2 Then continue to connect two front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then continue to connect two front-to-back fully connected layers configured with ReLU functions ( Figure 2 , followed by a fully connected layer with a Softmax function ( Figure 2The network is shown as Fully connected+Softmax), with a total of 13 convolutional layers and 3 fully connected layers.

[0125] Among them, it is understandable that for the convolutional layers, pooling layers, ReLU functions and fully connected layers in different positions, the parameters will vary. In the above content, for the convenience of explanation, they are simply called convolutional layers, pooling layers, ReLU functions and fully connected layers, that is, the types of specific model structures are pointed out without making further distinctions in the name details.

[0126] It can be understood that the embodiment here is a specific model architecture involved in the one-dimensional model configuration solution, which provides a set of specific implementation supporting solutions.

[0127] In addition, for the model, in specific applications, of course, other configuration work may also be involved.

[0128] For example, you can use the Adam optimizer to optimize, and set the initial learning rate to 0.0001 to adaptively adjust the learning rate. When verifying the performance of the model on the validation set, you can use accuracy, precision, recall, f1, and AUC as indicators for evaluating the model. These evaluation indicators are common indicators, so they are not described in detail here.

[0129] In addition, during the model training process, a five-fold cross-validation method can also be used to assist in evaluating the model training effect.

[0130] Next, we will focus on the two-dimensional model configuration scheme. As an exemplary implementation method, in the two-dimensional model configuration scheme, in the annotation results, the label assigned to the normal type is [0,0], the label assigned to the cardiac inhibition type is [1,0], the label assigned to the vascular inhibition type is [0,1], and the label assigned to the mixed type is [1,1].

[0131] It can be seen that in the embodiment here, a multi-label method is adopted, and the labels are specifically in the form of a two-dimensional array.

[0132] In this case, the syncope prediction result of the syncope prediction model may include two-digit prediction probability values ​​(also a two-dimensional array), and the output is 1 if the first probability value is greater than 0.5, and the output is 1 if the second prediction value is greater than 0.5.

[0133] In this way, the two predicted values ​​are combined to obtain the final prediction result.

[0134] At the same time, for the two-dimensional model configuration scheme, this application also provides a practical configuration scheme from the perspective of specific model architecture.

[0135] For details, see Figure 3 A schematic diagram of a network architecture of a two-dimensional convolutional neural network of the present application is shown. As an exemplary embodiment, the syncope prediction model can be a two-dimensional variant network based on VGG-16. Correspondingly, the model architecture of the syncope prediction model has the following configuration content:

[0136] The number of convolution kernels connected to each other is 64. Figure 3 Conv) is followed by a pooling layer ( Figure 3 Then continue to connect two convolutional layers with 128 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 256 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 512 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 512 convolution kernels connected front and back, then continue to connect three fully connected layers with ReLU function connected front and back ( Figure 3 FC), followed by a Sigmoid function ( Figure 3 Sigmoid), and the convolutional layers are configured with ReLU functions ( Figure 3 denoted as ReLU).

[0137] Among them, it is understandable that for the convolutional layers, pooling layers, ReLU functions and fully connected layers in different positions, the parameters will vary. In the above content, for the convenience of explanation, they are simply called convolutional layers, pooling layers, ReLU functions and fully connected layers, that is, the types of specific model structures are pointed out without making further distinctions in the name details.

[0138] It can be understood that the embodiment here is a specific model architecture involved in the two-dimensional model configuration solution, and provides a set of specific implementation supporting solutions.

[0139] The above is an introduction to the configuration method of the syncope prediction model provided in the present application. In order to facilitate better implementation of the configuration method of the syncope prediction model provided in the present application, the present application also provides a configuration device of the syncope prediction model from the perspective of functional modules.

[0140] See also Figure 4 , Figure 4 4 is a schematic diagram of a configuration device for a syncope prediction model of the present application. In the present application, the configuration device 400 for a syncope prediction model may specifically include the following structure:

[0141] An acquisition unit 401 is used to acquire sample electrocardiogram data, wherein the sample electrocardiogram data is specifically long-term electrocardiogram data in a normal time period;

[0142] A configuration unit 402 is used to configure corresponding annotation results for the sample electrocardiogram data, wherein the annotation results are used to indicate whether the user corresponding to the sample electrocardiogram data has a syncope situation, and the annotation results are also used to indicate the syncope type when the user corresponding to the sample electrocardiogram data has a syncope situation, and the syncope types include cardiac inhibition type, vascular inhibition type and mixed type;

[0143] The training unit 403 is used to train a syncope prediction model based on labeled sample ECG data, wherein the syncope prediction model is used to predict whether the user will experience syncope within a preset time period in the future according to the ECG data input into the model, as well as the specific syncope type if syncope occurs.

[0144] In an exemplary embodiment, in a one-dimensional model configuration scheme, the labeling result uses 0 to represent a normal type, 1 to represent a cardiac inhibition type, 2 to represent a vascular inhibition type, and 3 to represent a mixed type;

[0145] The syncope prediction result of the syncope prediction model is a one-dimensional array with a length of 4, and the 4 output numbers represent the probabilities of normal type, cardiac inhibition type, vascular inhibition type and mixed type respectively.

[0146] In yet another exemplary embodiment, the apparatus further includes a preprocessing unit 404, configured to:

[0147] The first two minutes of sample ECG data are discarded to obtain the remaining data;

[0148] Based on the sliding window method, the remaining data is cut into multiple segments according to the cutting unit of 7200 data length corresponding to ten minutes. Moreover, for different types of sample ECG data, each normal user randomly retains 6 segments of data, each cardiac inhibition user retains all data, each vascular inhibition user and each mixed user randomly retains 24 segments of data, thus completing the update of the sample ECG data;

[0149] The syncope prediction model is specifically configured with input data of a tensor with a shape of [n, 1, 72000] and a label corresponding to each sequence segment, where n is the number of sequence segments.

[0150] In another exemplary embodiment, the syncope prediction model is specifically a one-dimensional variant network based on VGG-16, and the model architecture of the syncope prediction model has the following configuration content:

[0151] The number of convolution kernels in the first convolution layer is set to 16, and the number of convolution kernels doubles after each pooling;

[0152] After the two front-to-back convolutional layers, connect a pooling layer, then continue to connect two front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then connect a pooling layer, then continue to connect three front-to-back convolutional layers, then continue to connect two front-to-back fully-connected layers configured with ReLU functions, and then continue with a fully-connected layer configured with Softmax function.

[0153] In another exemplary embodiment, in the two-dimensional model configuration scheme, in the annotation result, the label assigned to the normal type is [0,0], the label assigned to the cardiac inhibition type is [1,0], the label assigned to the vascular inhibition type is [0,1], and the label assigned to the mixed type is [1,1];

[0154] The syncope prediction result of the syncope prediction model includes two predicted probability values. If the first probability value is greater than 0.5, the output is 1, and if the second predicted value is greater than 0.5, the output is 1.

[0155] In another exemplary embodiment, the syncope prediction model is specifically a two-dimensional variant network based on VGG-16, and the model architecture of the syncope prediction model has the following configuration content:

[0156] After the two convolutional layers with 64 convolution kernels connected front and back, connect a pooling layer, then continue to connect two convolutional layers with 128 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 256 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 512 convolution kernels connected front and back, then connect a pooling layer, then continue to connect three convolutional layers with 512 convolution kernels connected front and back, then continue to connect three fully connected layers configured with ReLU functions, then continue with a Sigmoid function, and all convolutional layers are configured with ReLU functions.

[0157] In yet another exemplary embodiment, the apparatus further includes an application unit 405, configured to:

[0158] Get the current user's electrocardiogram data;

[0159] Inputting the electrocardiogram data of the current user into the syncope prediction model to perform corresponding syncope prediction processing;

[0160] extracting the syncope prediction processing result output by the syncope prediction model;

[0161] Based on the syncope prediction processing result, a syncope intervention measure reminder is output to the current user.

[0162] This application also provides a processing device from the perspective of hardware structure, see Figure 5 , Figure 5 501, a memory 502, and an input / output device 503. The processor 501 is used to execute the computer program stored in the memory 502 to implement the following Figure 1 The steps of the configuration method of the syncope prediction model in the corresponding embodiment; or, the processor 501 is used to execute the computer program stored in the memory 502 to implement the following Figure 4 Corresponding to the functions of each unit in the embodiment, the memory 502 is used to store the processor 501 executing the above Figure 1 A computer program required for the method of configuring the syncope prediction model in the corresponding embodiment.

[0163] Exemplarily, the computer program may be divided into one or more modules / units, one or more modules / units are stored in the memory 502, and executed by the processor 501 to complete the present application. One or more modules / units may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the computer device.

[0164] The processing device may include, but is not limited to, a processor 501, a memory 502, and an input / output device 503. Those skilled in the art will appreciate that the illustration is merely an example of a processing device and does not constitute a limitation on the processing device, and may include more or fewer components than shown in the illustration, or a combination of certain components, or different components. For example, the processing device may also include a network access device, a bus, etc., and the processor 501, the memory 502, the input / output device 503, etc. are connected via a bus.

[0165] The processor 501 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the processing device, and uses various interfaces and lines to connect various parts of the entire device.

[0166] The memory 502 can be used to store computer programs and / or modules. The processor 501 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 502 and calling the data stored in the memory 502. The memory 502 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the processing device, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0167] When the processor 501 is used to execute the computer program stored in the memory 502, the following functions can be implemented:

[0168] Acquire sample electrocardiogram data, wherein the sample electrocardiogram data is specifically long-term electrocardiogram data within a normal time period;

[0169] Configuring corresponding annotation results for the sample electrocardiogram data, wherein the annotation results are used to indicate whether the user corresponding to the sample electrocardiogram data has experienced syncope, and the annotation results are also used to indicate the syncope type when the user corresponding to the sample electrocardiogram data has experienced syncope, and the syncope types include cardiac inhibition type, vascular inhibition type, and mixed type;

[0170] A syncope prediction model is trained based on the labeled sample ECG data, wherein the syncope prediction model is used to predict whether the user will experience syncope within a preset time period in the future and the specific syncope type if syncope occurs based on the ECG data input into the model.

[0171] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the configuration device, processing device and corresponding units of the syncope prediction model described above can refer to the following. Figure 1 The description of the configuration method of the syncope prediction model in the corresponding embodiment will not be repeated here.

[0172] A person of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0173] To this end, the present application provides a computer-readable storage medium, in which a plurality of instructions are stored, and the instructions can be loaded by a processor to execute the present application as follows: Figure 1 The steps of the configuration method of the syncope prediction model in the corresponding embodiment, the specific operation can refer to the following Figure 1 The description of the configuration method of the syncope prediction model in the corresponding embodiment will not be repeated here.

[0174] The computer-readable storage medium may include: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0175] Due to the instructions stored in the computer-readable storage medium, the present application can be executed. Figure 1 The steps of the configuration method of the syncope prediction model in the corresponding embodiment, therefore, the present application can be implemented as follows Figure 1 The beneficial effects that can be achieved by the configuration method of the syncope prediction model in the corresponding embodiment are detailed in the previous description and will not be repeated here.

[0176] The configuration method, device, processing equipment and computer-readable storage medium of the syncope prediction model provided by the present application are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for configuring a syncope prediction model, characterized in that: The method comprises: Acquiring sample electrocardiogram data, wherein the sample electrocardiogram data is specifically long-term electrocardiogram data within a normal time period; Configuring corresponding annotation results for the sample electrocardiogram data, wherein the annotation results are used to indicate whether the user corresponding to the sample electrocardiogram data has a syncope, and the annotation results are also used to indicate the syncope type when the user corresponding to the sample electrocardiogram data has a syncope, and the syncope types include cardiac inhibition type, vascular inhibition type and mixed type; Based on the labeled sample ECG data, a syncope prediction model is trained, wherein the syncope prediction model is used to predict whether the user corresponding to the ECG data input by the model will experience syncope within a preset time period in the future, as well as the specific syncope type if syncope occurs, based on the ECG data input by the model.

2. The method according to claim 1, characterized in that In the one-dimensional model configuration scheme, the labeling result uses 0 to represent the normal type, 1 to represent the cardiac inhibition type, 2 to represent the vascular inhibition type, and 3 to represent the mixed type; The syncope prediction result of the syncope prediction model is a one-dimensional array with a length of 4, and the 4 output numbers respectively represent the probabilities of the normal type, the cardiac inhibition type, the vascular inhibition and the mixed type.

3. The method according to claim 2, characterized in that The sample electrocardiogram data is specifically the electrocardiogram data measured by lead II. Before training the syncope prediction model based on the labeled sample electrocardiogram data, the method further includes: Discarding the first two minutes of data of the sample electrocardiogram data to obtain the remaining data; Based on the sliding window method, the remaining data is cut into multiple segments of data according to the cutting unit of 7200 data length corresponding to ten minutes, and for different types of sample electrocardiogram data, each normal user randomly retains 6 segments of data, each cardiac inhibition user retains all data, each vascular inhibition user and each mixed user randomly retains 24 segments of data, thereby completing the update of the sample electrocardiogram data; The syncope prediction model is specifically configured to input a tensor of shape [n, 1, 72000] and a label corresponding to each sequence segment for the input data of the model, where n is the number of sequence segments.

4. The method according to claim 3, characterized in that The syncope prediction model is specifically a one-dimensional variant network based on VGG-16, and the model architecture of the syncope prediction model has the following configuration content: The number of convolution kernels in the first convolution layer is set to 16, and the number of convolution kernels is doubled after each pooling; After the two convolutional layers connected front and back, a pooling layer is connected, followed by two convolutional layers connected front and back, followed by another pooling layer, followed by three convolutional layers connected front and back, followed by another pooling layer, followed by three convolutional layers connected front and back, followed by another pooling layer, followed by three convolutional layers connected front and back, followed by two fully connected layers connected front and back configured with a ReLU function, followed by a fully connected layer configured with a Softmax function.

5. The method according to claim 1, characterized in that In the two-dimensional model configuration scheme, in the annotation result, the label assigned to the normal type is [0,0], the label assigned to the cardiac inhibition type is [1,0], the label assigned to the vascular inhibition type is [0,1], and the label assigned to the mixed type is [1,1]; The syncope prediction result of the syncope prediction model includes two predicted probability values. If the first probability value is greater than 0.5, the output is 1, and if the second predicted value is greater than 0.5, the output is 1.

6. The method according to claim 5, characterized in that The syncope prediction model is specifically a two-dimensional variant network based on VGG-16. The model architecture of the syncope prediction model has the following configuration contents: After the two convolutional layers with 64 convolution kernels connected front and back, a pooling layer is connected, and then the two convolutional layers with 128 convolution kernels connected front and back are connected, and then one pooling layer is connected, and then the three convolutional layers with 256 convolution kernels connected front and back are connected, and then one pooling layer is connected, and then the three convolutional layers with 512 convolution kernels connected front and back are connected, and then one pooling layer is connected, and then the three convolutional layers with 512 convolution kernels connected front and back are connected, and then the three fully connected layers configured with ReLU functions are connected front and back, and then a Sigmoid function is connected, and the convolutional layers are all configured with the ReLU function.

7. The method according to claim 1, characterized in that The method further comprises: Get the current user's electrocardiogram data; Inputting the electrocardiogram data of the current user into the syncope prediction model to perform corresponding syncope prediction processing; extracting the syncope prediction processing result output by the syncope prediction model; Based on the syncope prediction processing result, a syncope intervention measure reminder is output to the current user.

8. A device for configuring a syncope prediction model, characterized in that: The device comprises: An acquisition unit, used for acquiring sample electrocardiogram data, wherein the sample electrocardiogram data is specifically long-term electrocardiogram data in a normal time period; a configuration unit, configured to configure corresponding annotation results for the sample electrocardiogram data, wherein the annotation results are used to indicate whether the user corresponding to the sample electrocardiogram data has a syncope, and the annotation results are also used to indicate the syncope type when the user corresponding to the sample electrocardiogram data has a syncope, and the syncope types include cardiac inhibition type, vascular inhibition type and mixed type; A training unit is used to train a syncope prediction model based on the labeled sample electrocardiogram data, wherein the syncope prediction model is used to predict whether the user corresponding to the electrocardiogram data input by the model will experience syncope within a preset time period in the future and the specific syncope type when syncope occurs, based on the electrocardiogram data input by the model.

9. A processing device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and the processor executes the method according to any one of claims 1 to 7 when calling the computer program in the memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the method according to any one of claims 1 to 7.

Images