Training method and prediction method of pulmonary hypertension occurrence probability prediction model

Abstract

The present application discloses a training method and prediction method for a pulmonary hypertension occurrence probability prediction model. The training method includes acquiring multiple sets of historical image data; fusing a YOLOv6 network model and a feature-adaptive attention model to obtain an initial prediction model, wherein the feature-adaptive attention model is used to adaptively select different convolution kernels to extract and shuffle channel features of different scales; and inputting multiple sets of historical image data into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model. This method implements the construction of a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model based on the YOLOv6 network model and the feature-adaptive attention model, which is beneficial to improving the prediction accuracy and efficiency of predicting the occurrence probability of chronic thromboembolic pulmonary hypertension based on the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model.

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and in particular relates to a training method and a prediction method for a pulmonary hypertension occurrence probability prediction model. Background Art

[0002] Chronic thromboembolic pulmonary hypertension (CTEPH) is a clinical and pathophysiological syndrome characterized by structural or functional alterations in the pulmonary vasculature, resulting from heterogeneous diseases or etiologies and diverse pathogenic mechanisms, leading to elevated pulmonary vascular resistance and pulmonary artery pressure. Early screening for CTEPH is challenging, treatment is challenging, and the prognosis is poor. Due to its insidious early symptoms, CTEPH patients are often misdiagnosed or missed. The average duration of diagnosis for CTEPH patients from symptom onset to diagnosis is 2-4 years. Persistent CTEPH leads to persistent right ventricular overload, ultimately progressing to right heart failure and death. In my country, the average five-year survival rate for untreated CTEPH patients is only 20.8%. Studies have shown that patients with CTEPH who are screened early and receive effective treatment have significantly higher long-term survival rates than those in the middle and late stages of the disease. Therefore, early screening for CTEPH is crucial to reduce mortality and adverse effects of CTEPH, increase life expectancy, and improve the quality of life of patients with CTEPH.

[0003] Currently, the common method for early screening of CTEPH risk is mainly to perform computed tomography (CT) pulmonary angiography on patients, calculate blood flow characteristics based on the pulmonary vascular images of CT pulmonary angiography, and predict the patient's CTEPH probability based on the blood flow characteristics.

[0004] However, the method of predicting the probability of CTEPH in patients based on blood flow characteristics mainly relies on mathematical models and fluid mechanics equations, which are greatly affected by the boundary conditions of the modeling settings, resulting in low prediction accuracy of the probability of CTEPH. Summary of the Invention

[0005] In view of this, an embodiment of the present application provides a training method and a prediction method for a pulmonary hypertension occurrence probability prediction model to overcome the above problems of the prior art.

[0006] In a first aspect, an embodiment of the present application provides a training method for a pulmonary hypertension occurrence probability prediction model, comprising: obtaining multiple sets of historical image data, each set of historical image data comprising a user's historical pulmonary angiography dynamic images, historical pulmonary vascular images, and historical electrocardiograms, and each set of historical image data being annotated with a historical probability of occurrence of chronic thromboembolic pulmonary hypertension; fusing a YOLOv6 network model and a feature-adaptive attention model to obtain an initial prediction model, the feature-adaptive attention model being used to adaptively select different convolution kernels to extract and shuffle channel features of different scales; inputting multiple sets of historical image data into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model.

[0007] In a second aspect, an embodiment of the present application provides a method for predicting the probability of pulmonary hypertension, comprising: obtaining current imaging data of the patient, the current imaging data including a current pulmonary angiography dynamic image, a current pulmonary vascular image, and a current electrocardiogram; inputting the current imaging data into a target chronic thromboembolic pulmonary hypertension probability prediction model to obtain the corresponding current chronic thromboembolic pulmonary hypertension probability; wherein, the target chronic thromboembolic pulmonary hypertension probability prediction model is trained by the training method of the pulmonary hypertension probability prediction model provided in the first aspect above.

[0008] In a third aspect, an embodiment of the present application provides a training device for a pulmonary hypertension occurrence probability prediction model, comprising a historical data acquisition module, a model fusion module, and a training module. The historical data acquisition module is used to acquire multiple sets of historical image data, each set of historical image data includes a user's historical pulmonary angiography dynamic image, historical pulmonary vascular image, and historical electrocardiogram, and each set of historical image data is annotated with a historical probability of occurrence of chronic thromboembolic pulmonary hypertension; the model fusion module is used to fuse the YOLOv6 network model and the feature adaptive attention model to obtain an initial prediction model, and the feature adaptive attention model is used to adaptively select different convolution kernels to extract and shuffle channel features of different scales; the training module is used to input multiple sets of historical image data into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model.

[0009] In a fourth aspect, an embodiment of the present application provides an electronic device comprising a memory; one or more processors coupled to the memory; and one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more applications are configured to execute the training method of the pulmonary hypertension occurrence probability prediction model provided in the first aspect above, and / or the prediction method of the pulmonary hypertension occurrence probability provided in the second aspect above.

[0010] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, in which program code is stored. The program code can be called by a processor to execute the training method of the pulmonary hypertension occurrence probability prediction model provided in the first aspect above, and / or the prediction method of the pulmonary hypertension occurrence probability provided in the second aspect above.

[0011] In the seventh aspect, an embodiment of the present application provides a computer program product. When the computer program product is run on a computer device, the computer device executes the training method of the pulmonary hypertension occurrence probability prediction model provided in the first aspect above, and / or the prediction method of the pulmonary hypertension occurrence probability provided in the second aspect above.

[0012] The solution provided in this application obtains multiple sets of historical image data, each set of historical image data includes a user's historical pulmonary angiography dynamic image, historical pulmonary vascular image and historical electrocardiogram, each set of historical image data is annotated with a historical probability of chronic thromboembolic pulmonary hypertension, and integrates the YOLOv6 network model and the feature adaptive attention model to obtain an initial prediction model, the feature adaptive attention model is used to adaptively select different convolution kernels to extract and shuffle channel features of different scales, and input multiple sets of historical image data into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension probability prediction model, and realizes the training of the initial prediction model constructed based on the YOLOv6 network model and the feature adaptive attention model according to the historical image data to obtain a target chronic thromboembolic pulmonary hypertension probability prediction model, the prediction accuracy and prediction efficiency of the chronic thromboembolic pulmonary hypertension probability prediction model based on the target chronic thromboembolic pulmonary hypertension probability prediction model are both high, which is conducive to improving the prediction accuracy and prediction efficiency of the chronic thromboembolic pulmonary hypertension probability prediction.

[0013] Furthermore, the feature-adaptive attention model can adaptively select different convolution kernels to extract and shuffle channel features of different scales, which is beneficial to improving the robustness of the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby helping to further improve the prediction accuracy of the chronic thromboembolic pulmonary hypertension probability prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0015] Figure 1 A schematic diagram of a scenario of a system for predicting the probability of occurrence of pulmonary hypertension provided in an embodiment of the present application is shown.

[0016] Figure 2 A flow chart of a method for training a pulmonary hypertension occurrence probability prediction model provided in an embodiment of the present application is shown.

[0017] Figure 3 Shown Figure 2 A structural diagram of a feature adaptive attention model in a training method for a pulmonary hypertension probability prediction model is shown.

[0018] Figure 4 Shown Figure 2 A structural schematic diagram of an initial prediction model in a training method for a pulmonary hypertension occurrence probability prediction model is shown.

[0019] Figure 5 Another flowchart of the method for training a pulmonary hypertension occurrence probability prediction model provided in an embodiment of the present application is shown.

[0020] Figure 6 A flow chart of a method for predicting the probability of occurrence of pulmonary hypertension provided in an embodiment of the present application is shown.

[0021] Figure 7 A structural block diagram of a training device for a pulmonary hypertension occurrence probability prediction model provided in an embodiment of the present application is shown.

[0022] Figure 8 A structural block diagram of a device for predicting the probability of occurrence of pulmonary hypertension provided in an embodiment of the present application is shown.

[0023] Figure 9 A functional block diagram of an electronic device provided in an embodiment of the present application is shown.

[0024] Figure 10 A computer-readable storage medium provided in an embodiment of the present application is shown for storing or carrying program code for implementing a training method for a pulmonary hypertension occurrence probability prediction model provided in an embodiment of the present application, and / or a method for predicting the occurrence probability of pulmonary hypertension.

[0025] Figure 11 A computer program product provided in an embodiment of the present application is shown for storing or carrying program code for implementing a training method for a pulmonary hypertension occurrence probability prediction model provided in an embodiment of the present application, and / or a method for predicting the occurrence probability of pulmonary hypertension. DETAILED DESCRIPTION

[0026] In order to make the purpose, features, and advantages of the invention of this application more obvious and easy to understand, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the embodiments described below are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0027] It will be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0028] It should also be understood that the terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0029] It should be further understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0030] In addition, in the description of the present application, the terms "first", "second", "third", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0031] Chronic thromboembolic pulmonary hypertension (CTEPH) is a clinical and pathophysiological syndrome characterized by structural or functional alterations in the pulmonary vasculature, resulting from heterogeneous diseases or etiologies and diverse pathogenic mechanisms, leading to elevated pulmonary vascular resistance and pulmonary artery pressure. Early screening for CTEPH is challenging, treatment is challenging, and the prognosis is poor. Due to its insidious early symptoms, CTEPH patients are often misdiagnosed or missed. The average duration of diagnosis for CTEPH patients from symptom onset to diagnosis is 2-4 years. Persistent CTEPH leads to persistent right ventricular overload, ultimately progressing to right heart failure and death. In my country, the average five-year survival rate for untreated CTEPH patients is only 20.8%. Studies have shown that patients with CTEPH who are screened early and receive effective treatment have significantly higher long-term survival rates than those in the middle and late stages of the disease. Therefore, early screening for CTEPH is crucial to reduce mortality and adverse effects of CTEPH, increase life expectancy, and improve the quality of life of patients with CTEPH.

[0032] Currently, the common method for early screening of CTEPH risk is mainly to perform computed tomography (CT) pulmonary angiography on patients, calculate blood flow characteristics based on the pulmonary vascular images of CT pulmonary angiography, and predict the patient's CTEPH probability based on the blood flow characteristics.

[0033] However, the method of predicting the probability of CTEPH in patients based on blood flow characteristics mainly relies on mathematical models and fluid mechanics equations, which are greatly affected by the boundary conditions of the modeling settings, resulting in low prediction accuracy of the probability of CTEPH.

[0034] In response to the above problems, the training method and prediction method of the pulmonary hypertension occurrence probability prediction model provided in the embodiment of the present application are obtained by obtaining multiple sets of historical image data, each set of historical image data includes a user's historical pulmonary artery angiography dynamic image, historical pulmonary vascular image and historical electrocardiogram, each set of historical image data is annotated with the historical probability of chronic thromboembolic pulmonary hypertension, and the YOLOv6 network model and the feature adaptive attention model are integrated to obtain the initial prediction model, the feature adaptive attention model is used to adaptively select different convolution kernels to extract and shuffle channel features of different scales, and input multiple sets of historical image data. The initial prediction model was trained based on the YOLOv6 network model and the feature adaptive attention model according to the historical image data to obtain the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model. The prediction accuracy and efficiency of the probability of chronic thromboembolic pulmonary hypertension predicted based on the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model are high, which is conducive to improving the prediction accuracy and efficiency of the probability of chronic thromboembolic pulmonary hypertension.

[0035] Furthermore, the feature-adaptive attention model can adaptively select different convolution kernels to extract and shuffle channel features of different scales, which is beneficial to improving the robustness of the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby helping to further improve the prediction accuracy of the chronic thromboembolic pulmonary hypertension probability prediction.

[0036] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application.

[0037] See also Figure 1 , which shows a schematic diagram of an application scenario of the prediction system for the probability of pulmonary hypertension provided in an embodiment of the present application, which may include a computed tomography (CT) device 100, an angiogram 200, an electrocardiograph 300 and a processing device 400. The processing device 400 is communicatively connected to the CT device 100, the angiogram 200 and the electrocardiograph 300, and exchanges data with the CT device 100, the angiogram 200 and the electrocardiograph 300.

[0038] The CT device 100 may be used to capture images of the patient's pulmonary arteries, obtain pulmonary vascular images, and send the pulmonary vascular images to the processing device 400 .

[0039] The angiography device 200 can be used to perform angiography on the patient's pulmonary artery, obtain a dynamic image of the pulmonary artery angiography, and send the dynamic image of the pulmonary artery angiography to the processing device 400 .

[0040] The electrocardiograph 300 can be used to collect an electrocardiogram from a patient, obtain the electrocardiogram, and send the electrocardiogram to the processing device 400 .

[0041] The processing device 400 can be used to receive the pulmonary vascular image sent by the CT device 100, receive the pulmonary artery angiography dynamic image sent by the angiogram 200, and receive the electrocardiogram sent by the electrocardiograph 300, and predict the probability of occurrence of chronic thromboembolic pulmonary hypertension based on the pulmonary vascular image, the pulmonary artery angiography dynamic image and the electrocardiogram.

[0042] The processing device 400 may be a terminal device or a server, etc. The type of the processing device 400 is not limited here and can be specifically configured according to actual needs.

[0043] The terminal device can be a mobile terminal device (for example, any one of a mobile phone, a personal digital assistant (PDA), a tablet personal computer (Tablet PC), a laptop computer, a smart watch, a smart bracelet or a wearable device, etc.), or a fixed terminal device (desktop computer, smart panel, etc.).

[0044] The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or any cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), big data or artificial intelligence platforms, etc.

[0045] See also Figure 2 , which shows a flow chart of a method for training a pulmonary hypertension occurrence probability prediction model provided by an embodiment of the present application. In a specific embodiment, the method for training a pulmonary hypertension occurrence probability prediction model can be applied to Figure 1 The processing device 400 in the prediction system for the occurrence probability of pulmonary hypertension is shown below. Taking the processing device 400 as an example, Figure 2 The process shown in FIG. 1 is described in detail. The training method of the pulmonary hypertension occurrence probability prediction model may include the following steps S110 to S130.

[0046] Step S110: Acquire multiple sets of historical image data.

[0047] In an embodiment of the present application, the processing device can acquire multiple sets of historical image data. Each set of historical image data can include a user's historical pulmonary angiography dynamic images, historical pulmonary vascular images, and historical electrocardiograms, and each set of historical image data can be annotated with a historical probability of chronic thromboembolic pulmonary hypertension.

[0048] The historical pulmonary angiography dynamic images, historical pulmonary vascular images, and historical electrocardiograms in each set of historical image data are all collected during the same period. For example, the historical pulmonary angiography dynamic images, historical pulmonary vascular images, and historical electrocardiograms in each set of historical image data can all be collected on the same day. The historical pulmonary angiography dynamic images, historical pulmonary vascular images, and historical electrocardiograms in each set of historical image data can all be collected in the same week, etc.

[0049] In some implementations, the processing device pre-stores multiple sets of historical image data, and the processing device can read the pre-stored multiple sets of historical image data.

[0050] In some implementations, the processing device may generate upload prompt information and receive multiple sets of historical image data uploaded by the user according to the upload prompt information.

[0051] Among them, the upload prompt information can be used to prompt the user to upload multiple sets of historical image data to the processing device according to the upload prompt information. The upload prompt information can be at least any one of text prompt information, sound prompt information or light prompt information, etc., which is not limited here.

[0052] Step S120: Fuse the YOLOv6 network model and the feature adaptive attention model to obtain an initial prediction model.

[0053] In an embodiment of the present application, the processing device can fuse the YOLOv6 network model and the feature adaptive attention model to obtain an initial prediction model.

[0054] The YOLOv6 network model consists of three parts: the backbone network, the neck network, and the head network. The backbone network extracts features from the input image to generate feature maps, which are then passed to the neck network. The neck network fuses multi-scale features from the feature maps to generate fused features, which are then passed to the head network. The head network performs the final regression prediction.

[0055] The feature adaptive attention model can be used to adaptively select different convolution kernels to extract and shuffle channel features of different scales. The feature adaptive attention model can include a first feature extraction module, a second feature extraction module, a feature shuffling module, a third feature extraction module, and a fusion module. The first feature extraction module can be connected to the second feature extraction module and the feature shuffling module, the second feature extraction module can be connected to the feature shuffling module, and the feature shuffling module can be connected to the third feature extraction module and the fusion module. Figure 3 shown.

[0056] The first feature extraction module may include a first 1x1 convolutional layer, which can be used to extract features from multiple sets of historical image data to obtain a first feature map. The second feature extraction module may include a 5x5 depthwise separable convolutional layer, a Selective Kernel Networks (SKNets) layer, and a 9x9 depthwise separable convolutional layer. The 5x5 depthwise separable convolutional layer can be used to perform deep feature extraction on the first feature map to obtain a second feature map. The SKNets layer can be used to assign weights to the first feature map to obtain a third feature map. The 9x9 depthwise separable convolutional layer can be used to perform deep feature extraction on the first feature map to obtain a fourth feature map.

[0057] The feature shuffling module may include a chunk() function, which may be used to perform a shuffling operation on the first feature map, the second feature map, the third feature map, and the fourth feature map to obtain a fifth feature map.

[0058] The third feature extraction module may include a second 1x1 convolutional layer and a third 1x1 convolutional layer connected in sequence. The second 1x1 convolutional layer may be used to extract features from the fifth feature map to obtain a sixth feature map. The third 1x1 convolutional layer may be used to extract features from the sixth feature map to obtain a seventh feature map. The fusion module may include a Concat layer. The Concat layer may be used to fuse the fifth feature map and the seventh feature map to obtain an eighth feature map.

[0059] The feature adaptive attention model can be connected between the Spatial Pyramid Pooling-Fast (SPPF) module and the neck network in the backbone network of the YOLOv6 network model, such as Figure 4 shown.

[0060] In some embodiments, the processing device can fuse the SPPF module and the feature adaptive attention model in the backbone network of the YOLOv6 network model to obtain an initial fusion model, and delete the cross-stage partial (CSP) module in the initial fusion model to obtain an initial prediction model. Due to the addition of the feature adaptive attention model, channel features of different scales can be combined, enriching the feature information of the initial fusion model. When the CSP module in the initial fusion model is deleted, the model expression performance of the obtained initial prediction model remains unchanged, and the network structure of the initial prediction model is simplified, which is beneficial to reducing the amount of computation of the initial prediction model, thereby improving the prediction efficiency of predicting the probability of occurrence of chronic thromboembolic pulmonary hypertension.

[0061] In some embodiments, after acquiring multiple sets of historical image data, the processing device can fuse the first feature extraction module, the second feature extraction module, the feature shuffling module, the third feature extraction module and the fusion module, so that the first feature extraction module is connected to the second feature extraction module and the feature shuffling module, the second feature extraction module is connected to the feature shuffling module, and the feature shuffling module is connected to the third feature extraction module and the fusion module to obtain a feature adaptive attention model, and fuse the SPPF module and the feature adaptive attention model in the backbone network of the YOLOv6 network model to obtain an initial fusion model, and delete the CSP module in the initial fusion model to obtain an initial prediction model.

[0062] It should be noted that in the embodiment of the present application, there is no order between step S110 and step S120. The processing device can fuse the YOLOv6 network model and the feature adaptive attention model after obtaining multiple sets of historical image data to obtain an initial prediction model. The processing device can also obtain multiple sets of historical image data after fusing the YOLOv6 network model and the feature adaptive attention model to obtain the initial prediction model. This is not limited here.

[0063] Step S130: Input multiple sets of historical image data into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model.

[0064] In an embodiment of the present application, the processing device can input multiple sets of historical image data into the initial prediction model, the initial prediction model receives and responds to the multiple sets of historical image data, and is trained based on the multiple sets of historical image data to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model, thereby realizing the training of the initial prediction model constructed based on the YOLOv6 network model and the feature adaptive attention model based on the historical image data to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model, and the prediction accuracy and prediction efficiency of predicting the probability of chronic thromboembolic pulmonary hypertension based on the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model are both high, which is conducive to improving the prediction accuracy and prediction efficiency of predicting the probability of chronic thromboembolic pulmonary hypertension occurrence.

[0065] Furthermore, the feature-adaptive attention model can adaptively select different convolution kernels to extract and shuffle channel features of different scales, which is beneficial to improving the robustness of the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby helping to further improve the prediction accuracy of the chronic thromboembolic pulmonary hypertension probability prediction.

[0066] The solution provided in this application obtains multiple sets of historical image data, each set of historical image data includes a user's historical pulmonary angiography dynamic image, historical pulmonary vascular image and historical electrocardiogram, each set of historical image data is annotated with a historical probability of chronic thromboembolic pulmonary hypertension, and integrates the YOLOv6 network model and the feature adaptive attention model to obtain an initial prediction model, the feature adaptive attention model is used to adaptively select different convolution kernels to extract and shuffle channel features of different scales, and input multiple sets of historical image data into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension probability prediction model, and realizes the training of the initial prediction model constructed based on the YOLOv6 network model and the feature adaptive attention model according to the historical image data to obtain a target chronic thromboembolic pulmonary hypertension probability prediction model, the prediction accuracy and prediction efficiency of the chronic thromboembolic pulmonary hypertension probability prediction model based on the target chronic thromboembolic pulmonary hypertension probability prediction model are both high, which is conducive to improving the prediction accuracy and prediction efficiency of the chronic thromboembolic pulmonary hypertension probability prediction.

[0067] Furthermore, the feature-adaptive attention model can adaptively select different convolution kernels to extract and shuffle channel features of different scales, which is beneficial to improving the robustness of the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby helping to further improve the prediction accuracy of the chronic thromboembolic pulmonary hypertension probability prediction.

[0068] See also Figure 5 , which shows a flow chart of a method for training a pulmonary hypertension occurrence probability prediction model provided by another embodiment of the present application. In a specific embodiment, the method for training a pulmonary hypertension occurrence probability prediction model can be applied to Figure 1 The processing device 400 in the prediction system for the occurrence probability of pulmonary hypertension is shown below. Taking the processing device 400 as an example, Figure 5 The process shown in FIG. 1 is described in detail. The training method of the pulmonary hypertension occurrence probability prediction model may include the following steps S210 to S250.

[0069] Step S210: Acquire multiple sets of historical image data.

[0070] Step S220: Fuse the YOLOv6 network model and the feature adaptive attention model to obtain an initial prediction model.

[0071] In this embodiment, step S210 and step S220 may refer to the contents of the corresponding steps in the aforementioned embodiment, and will not be repeated here.

[0072] Step S230: Divide the multiple groups of historical image data into training sets and test sets.

[0073] In this embodiment, the processing device can annotate multiple groups of historical image data to obtain corresponding annotated images, and divide the annotated images according to preset division rules to obtain a training set and a test set.

[0074] Among them, when the processing equipment annotates multiple groups of historical image data, it mainly annotates the blood flow characteristics in the historical pulmonary angiography dynamic images and historical pulmonary vascular images in each group of historical image data, and annotates the abnormal heart rate values ​​in the historical electrocardiogram, and annotates the corresponding historical probability of chronic thromboembolic pulmonary hypertension for each group of historical image data.

[0075] The preset partitioning rule is an artificial partitioning rule. For example, the preset partitioning rule may be a training set:test set = 9:1 artificial partitioning rule. When the number of historical image data sets is 20,000, the training set is 18,000 and the test set is 2,000. The preset partitioning rule may be a training set:test set = 8:1 artificial partitioning rule. When the number of historical image data sets is 27,000, the training set is 24,000 and the test set is 3,000. The partitioning method of the preset partitioning rule is not limited here and can be set according to actual needs.

[0076] Step S240: performing data enhancement processing on the training set to obtain an enhanced training set.

[0077] In this embodiment, the processing device may perform data enhancement processing on the training set to obtain an enhanced training set.

[0078] Among them, data enhancement processing can include at least any one of brightness enhancement processing, grayscale enhancement processing, contrast enhancement processing and transparency enhancement processing. The type of data enhancement processing is not limited here and can be set according to actual needs.

[0079] Step S250: inputting the enhanced training set into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model.

[0080] In this embodiment, the processing device can input the enhanced training set into the initial prediction model, the initial prediction model receives and responds to the enhanced training set, and is trained according to the enhanced training set to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model, thereby realizing the training of the initial prediction model according to the enhanced training set, avoiding the low robustness of the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model obtained by training the initial prediction model with a smaller training set due to less historical imaging data, and increasing the robustness of the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model.

[0081] In some embodiments, after the processing device inputs the enhanced training set into the initial prediction model for training and obtains the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model, it can determine whether the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model has converged based on the test set, so as to determine whether the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model is stable.

[0082] Among them, the processing device can input the test set into the target chronic thromboembolic pulmonary hypertension probability prediction model, the target chronic thromboembolic pulmonary hypertension probability prediction model receives and responds to the test set, performs testing according to the test set, obtains corresponding test results, and determines the test accuracy corresponding to the test results based on the test results and the test set, and determines whether the target chronic thromboembolic pulmonary hypertension probability prediction model converges based on the test accuracy.

[0083] When the test accuracy is greater than or equal to the preset accuracy threshold, it is determined that the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model has converged; when the test accuracy is less than the preset accuracy threshold, it is determined that the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model has not converged.

[0084] The preset accuracy threshold can be used to characterize the minimum accuracy corresponding to the convergence of the target chronic thromboembolic pulmonary hypertension probability prediction model. The preset accuracy threshold can be an accuracy pre-set by the user, or it can be an accuracy automatically generated by the processing equipment based on the prediction process of multiple predictions of the probability of chronic thromboembolic pulmonary hypertension. The setting method of the preset accuracy threshold is not limited here, and it can be set according to actual needs.

[0085] The solution provided in this embodiment obtains multiple groups of historical image data and integrates the YOLOv6 network model and the feature adaptive attention model to obtain an initial prediction model, divides the multiple groups of historical image data to obtain a training set and a test set, performs data enhancement processing on the training set to obtain an enhanced training set, and inputs the enhanced training set into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model. It realizes the training of the initial prediction model constructed based on the YOLOv6 network model and the feature adaptive attention model according to the historical image data to obtain the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model. The prediction accuracy and prediction efficiency of predicting the probability of chronic thromboembolic pulmonary hypertension based on the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model are both high, which is conducive to improving the prediction accuracy and prediction efficiency of predicting the probability of chronic thromboembolic pulmonary hypertension.

[0086] Furthermore, the feature-adaptive attention model can adaptively select different convolution kernels to extract and shuffle channel features of different scales, which is beneficial to improving the robustness of the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby helping to further improve the prediction accuracy of the chronic thromboembolic pulmonary hypertension probability prediction.

[0087] Furthermore, the initial prediction model is trained based on the enhanced training set to avoid the low robustness of the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model obtained by training the initial prediction model with a smaller training set due to less historical imaging data, thereby increasing the robustness of the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model.

[0088] See also Figure 6 , which shows a flow chart of a method for predicting the probability of occurrence of pulmonary hypertension provided by an embodiment of the present application. In a specific embodiment, the method for predicting the probability of occurrence of pulmonary hypertension can be applied to Figure 1 The processing device 400 in the prediction system for the occurrence probability of pulmonary hypertension is shown below. Taking the processing device 400 as an example, Figure 6 The process shown in FIG3 is described in detail. The method for predicting the probability of occurrence of pulmonary hypertension may include the following steps S310 to S320.

[0089] Step S310: Acquire the patient's current image data.

[0090] In this embodiment, the patient's current imaging data may include the current pulmonary angiography dynamic image, the current pulmonary vascular image, and the current electrocardiogram, etc. The current pulmonary angiography dynamic image, the current pulmonary vascular image, and the current electrocardiogram in the current imaging data are all collected during the same period.

[0091] For example, the current pulmonary angiography dynamic image, the current pulmonary vascular image and the current electrocardiogram in the current image data can all be collected on the same day, the current pulmonary angiography dynamic image, the current pulmonary vascular image and the current electrocardiogram in the current image data can all be collected in the same week, etc.

[0092] When the user needs to predict the probability of chronic thromboembolic pulmonary hypertension in a patient, a prediction instruction can be sent to the processing device. The processing device receives and responds to the prediction instruction, and sends image acquisition instructions to the angiography device, CT device and electrocardiograph respectively. The angiography device receives and responds to the image acquisition instruction, angiograms the patient's pulmonary artery, obtains the current pulmonary artery angiography dynamic image, and sends the current pulmonary artery angiography dynamic image to the processing device. The CT device receives and responds to the image acquisition instruction, acquires an image of the patient's pulmonary artery, obtains the current pulmonary vascular image, and sends the current pulmonary vascular image to the processing device. The electrocardiograph receives and responds to the image acquisition instruction, acquires an electrocardiogram of the patient, obtains the current electrocardiogram, and sends the current electrocardiogram to the processing device. The processing device receives the current pulmonary artery angiography dynamic image returned by the angiography device, receives the current pulmonary vascular image returned by the CT device, and receives the current electrocardiogram returned by the electrocardiograph.

[0093] In some embodiments, the processing device can detect the user's operation. When it is determined based on the detected user operation that the user has entered a prediction instruction for predicting the probability of occurrence of chronic thromboembolic pulmonary hypertension in the patient, the prediction instruction for predicting the probability of occurrence of chronic thromboembolic pulmonary hypertension in the patient is received.

[0094] For example, when a user needs to predict the probability of a patient suffering from chronic thromboembolic pulmonary hypertension, the user can perform a touch operation on the operation panel of the processing device. The processing device responds to the user's touch operation, generates a corresponding touch signal, and analyzes the touch signal. When it is determined that the touch signal is a preset signal for representing the prediction of the probability of a patient suffering from chronic thromboembolic pulmonary hypertension, it is determined that a prediction instruction for predicting the probability of a patient suffering from chronic thromboembolic pulmonary hypertension has been received.

[0095] In some embodiments, the processing device may be provided with a voice recognition module. When a user needs to predict the probability of occurrence of chronic thromboembolic pulmonary hypertension in a patient, the user may send a voice message within the voice collection range of the voice recognition module. The voice recognition module collects the voice message sent by the user and performs voice recognition on the collected voice message. Based on the recognition result of the voice recognition, it is determined that the recognition result contains keywords for indicating the prediction of the probability of occurrence of chronic thromboembolic pulmonary hypertension in the patient, such as "prediction of the probability of occurrence of chronic thromboembolic pulmonary hypertension", and for example, "probability of occurrence of chronic thromboembolic pulmonary hypertension" and "prediction", etc., and it is determined that a prediction instruction for predicting the probability of occurrence of chronic thromboembolic pulmonary hypertension in the patient has been received.

[0096] As an example, the voice information sent by the user is: predict the probability of occurrence of chronic thromboembolic pulmonary hypertension in the patient, and the recognition result of the voice recognition contains the keywords "probability of occurrence of chronic thromboembolic pulmonary hypertension" and "prediction", then it is determined that a prediction instruction for predicting the probability of occurrence of chronic thromboembolic pulmonary hypertension in the patient has been received.

[0097] In some embodiments, the system for predicting the probability of occurrence of pulmonary hypertension may further include a client, which is connected to the processing device via a network and exchanges data with the processing device via the network.

[0098] When a user needs to predict the probability of chronic thromboembolic pulmonary hypertension in a patient, the user can send a prediction instruction to the client. The client receives and responds to the prediction instruction, forwards the prediction instruction to the processing device through the network, and the processing device receives the prediction instruction forwarded by the client.

[0099] Among them, the client can be any one of a mobile client (for example, a mobile phone client, a PDA client, a Tablet PC client, a laptop client, a smart watch client, a smart bracelet client or a wearable client, etc.) or a fixed client (for example, a desktop computer client, a smart panel client, etc.). The type of client is not limited here and can be set according to actual needs.

[0100] The network may be any one of a ZigBee network, a Bluetooth (BT) network, a Wireless Fidelity (Wi-Fi) network, a Thread network, a Long Range Radio (LoRa) network, a Low-Power Wide-Area Network (LPWAN), an infrared network, a Narrow Band Internet of Things (NB-IoT), a Controller Area Network (CAN), a Digital Living Network Alliance (DLNA) network, a Wide Area Network (WAN), a Local Area Network (LAN), a Metropolitan Area Network (MAN), or a Wireless Personal Area Network (WPAN), etc., without limitation here.

[0101] Step S320: inputting the current image data into the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model to obtain the corresponding current chronic thromboembolic pulmonary hypertension occurrence probability.

[0102] In this embodiment, after obtaining the patient's current image data, the processing device can input the current image data into the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model. The target chronic thromboembolic pulmonary hypertension occurrence probability prediction model receives and responds to the current image data, predicts the occurrence probability of chronic thromboembolic pulmonary hypertension based on the current image data, obtains the corresponding current chronic thromboembolic pulmonary hypertension occurrence probability, and outputs the current chronic thromboembolic pulmonary hypertension occurrence probability to the processing device. The processing device receives the current chronic thromboembolic pulmonary hypertension occurrence probability output by the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model, realizes the prediction of the occurrence probability of chronic thromboembolic pulmonary hypertension based on the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model, and improves the prediction accuracy and prediction efficiency of the prediction of the occurrence probability of chronic thromboembolic pulmonary hypertension.

[0103] Furthermore, the feature-adaptive attention model can adaptively select different convolution kernels to extract and shuffle channel features of different scales, thereby improving the robustness of the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby further improving the prediction accuracy of the chronic thromboembolic pulmonary hypertension probability prediction.

[0104] The target chronic thromboembolic pulmonary hypertension occurrence probability prediction model can be obtained by training the pulmonary hypertension occurrence probability prediction model training method in the aforementioned embodiment.

[0105] The solution provided in this embodiment obtains the patient's current image data and inputs the current image data into the target chronic thromboembolic pulmonary hypertension probability prediction model to obtain the corresponding current chronic thromboembolic pulmonary hypertension probability. This achieves the prediction of the chronic thromboembolic pulmonary hypertension probability based on the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby improving the prediction accuracy and efficiency of the chronic thromboembolic pulmonary hypertension probability prediction.

[0106] Furthermore, the feature-adaptive attention model can adaptively select different convolution kernels to extract and shuffle channel features of different scales, thereby improving the robustness of the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby further improving the prediction accuracy of the chronic thromboembolic pulmonary hypertension probability prediction.

[0107] See also Figure 7 , which shows a training device 500 for a pulmonary hypertension occurrence probability prediction model provided by an embodiment of the present application. The training device 500 for a pulmonary hypertension occurrence probability prediction model can be applied to Figure 1 The processing device 400 in the prediction system for the occurrence probability of pulmonary hypertension is shown below. Taking the processing device 400 as an example, Figure 7 The training device 500 for the pulmonary hypertension occurrence probability prediction model shown in FIG. 5 is described in detail. The training device 500 for the pulmonary hypertension occurrence probability prediction model may include a historical data acquisition module 510 , a model fusion module 520 and a training module 530 .

[0108] The historical data acquisition module 510 can be used to acquire multiple sets of historical image data, each set of historical image data can include a user's historical pulmonary angiography dynamic images, historical pulmonary vascular images and historical electrocardiograms, and each set of historical image data can be annotated with the historical probability of occurrence of chronic thromboembolic pulmonary hypertension; the model fusion module 520 can be used to fuse the YOLOv6 network model and the feature adaptive attention model to obtain an initial prediction model, and the feature adaptive attention model can be used to adaptively select different convolution kernels to extract and shuffle channel features of different scales; the training module 530 can be used to input multiple sets of historical image data into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension probability prediction model.

[0109] In some implementations, the model fusion module 520 may include a fusion unit and a deletion unit.

[0110] The fusion unit can be used to fuse the fast pyramid pooling module and the feature adaptive attention model in the backbone network of the YOLOv6 network model to obtain an initial fusion model; the deletion unit can be used to delete the cross-stage local modules in the initial fusion model to obtain an initial prediction model.

[0111] In some embodiments, the training device 500 for the pulmonary hypertension occurrence probability prediction model may further include a module fusion module.

[0112] The fusion module can be used for the model fusion module 520 to fuse the YOLOv6 network model and the feature adaptive attention model. Before obtaining the initial prediction model, the first feature extraction module, the second feature extraction module, the feature shuffling module, the third feature extraction module and the fusion module are fused to obtain the feature adaptive attention model.

[0113] The first feature extraction module can be connected to the second feature extraction module and the feature shuffling module, the second feature extraction module can be connected to the feature shuffling module, and the feature shuffling module can be connected to the third feature extraction module and the fusion module.

[0114] In some embodiments, the first feature extraction module may include a first 1x1 convolutional layer, which may be used to perform feature extraction on multiple sets of historical image data to obtain a first feature map. The second feature extraction module may include a 5x5 depthwise separable convolutional layer, a selective kernel network layer, and a 9x9 depthwise separable convolutional layer. The 5x5 depthwise separable convolutional layer may be used to perform deep feature extraction on the first feature map to obtain a second feature map. The selective kernel network layer may be used to assign weights to the first feature map to obtain a third feature map. The 9x9 depthwise separable convolutional layer may be used to perform deep feature extraction on the first feature map to obtain a fourth feature map.

[0115] The feature shuffling module may include a chunk() function, which may be used to perform a shuffling operation on the first feature map, the second feature map, the third feature map, and the fourth feature map to obtain a fifth feature map.

[0116] The third feature extraction module may include a second 1x1 convolutional layer and a third 1x1 convolutional layer connected in sequence, the second 1x1 convolutional layer can be used to perform feature extraction on the fifth feature map to obtain a sixth feature map, and the third 1x1 convolutional layer can be used to perform feature extraction on the sixth feature map to obtain a seventh feature map; the fusion module may include a Concat layer, and the Concat layer can be used to fuse the fifth feature map and the seventh feature map to obtain an eighth feature map.

[0117] In some embodiments, the training device 500 for the pulmonary hypertension occurrence probability prediction model may further include a division module and a processing module.

[0118] The division module can be used for the training module 530 to input multiple groups of historical image data into the initial prediction model for training, and before obtaining the target chronic thromboembolic pulmonary hypertension probability prediction model, the multiple groups of historical image data are divided to obtain a training set and a test set; the processing module can be used to perform data enhancement processing on the training set to obtain an enhanced training set.

[0119] In some implementations, the training module 530 may include a training unit.

[0120] The training unit can be used to input the enhanced training set into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model.

[0121] In some embodiments, the training device 500 for the pulmonary hypertension occurrence probability prediction model may further include a determination module.

[0122] The determination module can be used to determine whether the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model converges based on the test set.

[0123] The solution provided in this embodiment obtains multiple sets of historical image data, each set of historical image data includes a user's historical pulmonary angiography dynamic image, historical pulmonary vascular image and historical electrocardiogram, each set of historical image data is annotated with a historical probability of chronic thromboembolic pulmonary hypertension, and integrates the YOLOv6 network model and the feature adaptive attention model to obtain an initial prediction model, the feature adaptive attention model is used to adaptively select different convolution kernels to extract and shuffle channel features of different scales, and multiple sets of historical image data are input into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension probability prediction model. The initial prediction model constructed based on the YOLOv6 network model and the feature adaptive attention model is trained according to the historical image data to obtain a target chronic thromboembolic pulmonary hypertension probability prediction model. The prediction accuracy and efficiency of the chronic thromboembolic pulmonary hypertension probability prediction model based on the target chronic thromboembolic pulmonary hypertension probability prediction model are both high, which is conducive to improving the prediction accuracy and efficiency of the chronic thromboembolic pulmonary hypertension probability prediction.

[0124] Furthermore, the feature-adaptive attention model can adaptively select different convolution kernels to extract and shuffle channel features of different scales, which is beneficial to improving the robustness of the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby helping to further improve the prediction accuracy of the chronic thromboembolic pulmonary hypertension probability prediction.

[0125] See also Figure 8 , which shows a device 600 for predicting the probability of occurrence of pulmonary hypertension provided by an embodiment of the present application. The device 600 for predicting the probability of occurrence of pulmonary hypertension can be applied to Figure 1 The processing device 400 in the prediction system for the occurrence probability of pulmonary hypertension is shown below. Taking the processing device 400 as an example, Figure 8 The device 600 for predicting the probability of occurrence of pulmonary hypertension shown in FIG. 6 is described in detail. The device 600 for predicting the probability of occurrence of pulmonary hypertension may include a current data acquisition module 610 and an input module 620 .

[0126] The current data acquisition module 610 can be used to acquire the patient's current image data, which may include the current pulmonary angiography dynamic image, the current pulmonary vascular image and the current electrocardiogram; the input module 620 can be used to input the current image data into the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model to obtain the corresponding current chronic thromboembolic pulmonary hypertension occurrence probability; wherein, the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model is trained by the training method of the pulmonary hypertension occurrence probability prediction model in the aforementioned embodiment.

[0127] The solution provided in this embodiment obtains the patient's current image data and inputs the current image data into the target chronic thromboembolic pulmonary hypertension probability prediction model to obtain the corresponding current chronic thromboembolic pulmonary hypertension probability. This achieves the prediction of the chronic thromboembolic pulmonary hypertension probability based on the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby improving the prediction accuracy and efficiency of the chronic thromboembolic pulmonary hypertension probability prediction.

[0128] Furthermore, the feature-adaptive attention model can adaptively select different convolution kernels to extract and shuffle channel features of different scales, thereby improving the robustness of the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby further improving the prediction accuracy of the chronic thromboembolic pulmonary hypertension probability prediction.

[0129] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to in detail. For the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple. For the relevant parts, refer to the partial description of the method embodiments. Any processing method described in the method embodiment can be implemented by the corresponding processing module in the device embodiment, and will not be repeated in detail in the device embodiment.

[0130] In addition, the functional modules in the various embodiments of the present application may be integrated into a processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The above-mentioned integrated modules may be implemented in the form of hardware or software functional modules.

[0131] See also Figure 9 , which shows a functional block diagram of an electronic device 700 provided by an embodiment of the present application. The electronic device 700 may include one or more of the following components: a memory 710, a processor 720, and one or more applications, wherein the one or more applications may be stored in the memory 710 and configured to be executed by the one or more processors 720, and the one or more applications are configured to execute the method described in the aforementioned method embodiment.

[0132] The memory 710 may include a random access memory (RAM) or a read-only memory (Read-Only Memory). The memory 710 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 710 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as obtaining multiple sets of historical image data, marking historical chronic thromboembolic pulmonary hypertension occurrence probabilities, fusing the YOLOv6 network model and the feature adaptive attention model, obtaining an initial prediction model, extracting features, shuffling features, inputting multiple sets of historical image data, training the initial prediction model, obtaining a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model, fusing the fast pyramid pooling module and the feature adaptive attention model, obtaining an initial fusion module, deleting cross-stage local modules, fusing the first feature extraction module, the second Feature extraction module, feature shuffling module, third feature extraction module and fusion module, obtain feature adaptive attention model, obtain first feature map, extract deep features, obtain second feature map, weight assignment, obtain third feature map, obtain fourth feature map, shuffling operation, obtain fifth feature map, obtain sixth feature map, obtain seventh feature map, obtain eighth feature map, divide multiple groups of historical image data, obtain training set, obtain test set, data enhancement processing, obtain enhanced training set, input enhanced training set, determine whether the model converges, obtain current image data, input current image data and obtain current probability of occurrence of chronic thromboembolic pulmonary hypertension, etc.), instructions for implementing the following method embodiments, etc. The data storage area can also store data created by the electronic device 700 during use (such as multiple sets of historical image data, historical pulmonary angiography dynamic images, historical pulmonary vascular images, historical electrocardiograms, historical chronic thromboembolic pulmonary hypertension probability, YOLOv6 network model, feature adaptive attention model, initial prediction model, different convolution kernels, channel features of different scales, target chronic thromboembolic pulmonary hypertension probability prediction model, backbone network, fast pyramid pooling module, initial fusion module, cross-stage local module, first feature extraction module, second feature extraction module, feature extraction module, etc. Feature shuffling module, third feature extraction module, fusion module, first 1x1 convolution layer, first feature map, 5x5 depth-wise separable convolution layer, selective kernel network layer, 9x9 depth-wise separable convolution layer, second feature map, third feature map, fourth feature map, chunk() function, fifth feature map, second 1x1 convolution layer, third 1x1 convolution layer, sixth feature map, seventh feature map, Concat layer, eighth feature map, training set, test set, enhanced training set, current image data, current pulmonary vascular image, current electrocardiogram, and current probability of occurrence of chronic thromboembolic pulmonary hypertension), etc.

[0133] The processor 720 may include one or more processing cores. The processor 720 utilizes various interfaces and circuits to connect various components within the electronic device 700. It executes instructions, programs, code sets, or instruction sets stored in the memory 710, and accesses data stored in the memory 710 to perform various functions and process data within the electronic device 700. Optionally, the processor 720 may be implemented using at least one hardware form factor selected from the group consisting of a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 720 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. The CPU primarily processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing display content; and the modem handles wireless communications. It is understood that the modem may not be integrated into the processor 720 and may instead be implemented via a separate communications chip.

[0134] Please refer to Figure 10 , which shows a block diagram of a computer-readable storage medium provided in an embodiment of the present application. The computer-readable storage medium 800 stores program code 810, which can be called by a processor to execute the method described in the above method embodiment.

[0135] The computer-readable storage medium 800 can be an electronic memory such as a flash memory, an EEPROM (Electrically Erasable Programmable Read-Only Memory), an EPROM, a hard disk, or a ROM. Alternatively, the computer-readable storage medium 800 includes a non-transitory computer-readable storage medium. The computer-readable storage medium 800 has storage space for program code 810 for executing any of the method steps described above. These program codes can be read from or written to one or more computer program products. The program code 810 can be compressed, for example, in a suitable form.

[0136] Please refer to Figure 11, which shows a block diagram of the structure of a computer program product 900 provided in an embodiment of the present application. Computer program product 900 includes a computer program / instructions 910, which is stored in a computer-readable storage medium of a computer device. When computer program product 900 is executed on a computer device, the computer device's processor reads computer program / instructions 910 from the computer-readable storage medium and executes computer program / instructions 910, causing the computer device to perform the method described in the above method embodiment.

[0137] The solution provided in this embodiment obtains multiple sets of historical image data, each set of historical image data includes a user's historical pulmonary angiography dynamic image, historical pulmonary vascular image and historical electrocardiogram, each set of historical image data is annotated with a historical probability of chronic thromboembolic pulmonary hypertension, and integrates the YOLOv6 network model and the feature adaptive attention model to obtain an initial prediction model, the feature adaptive attention model is used to adaptively select different convolution kernels to extract and shuffle channel features of different scales, and multiple sets of historical image data are input into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension probability prediction model. The initial prediction model constructed based on the YOLOv6 network model and the feature adaptive attention model is trained according to the historical image data to obtain a target chronic thromboembolic pulmonary hypertension probability prediction model. The prediction accuracy and efficiency of the chronic thromboembolic pulmonary hypertension probability prediction model based on the target chronic thromboembolic pulmonary hypertension probability prediction model are both high, which is conducive to improving the prediction accuracy and efficiency of the chronic thromboembolic pulmonary hypertension probability prediction.

[0138] Furthermore, the feature-adaptive attention model can adaptively select different convolution kernels to extract and shuffle channel features of different scales, which is beneficial to improving the robustness of the target chronic thromboembolic pulmonary hypertension probability prediction model, thereby helping to further improve the prediction accuracy of the chronic thromboembolic pulmonary hypertension probability prediction.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A training method for a pulmonary hypertension probability prediction model, characterized in that: include: Acquire multiple sets of historical image data, each set of historical image data including a user's historical pulmonary angiography dynamic image, historical pulmonary vascular image, and historical electrocardiogram, wherein each set of historical image data is annotated with a historical probability of occurrence of chronic thromboembolic pulmonary hypertension; The first feature extraction module, the second feature extraction module, the feature shuffling module, the third feature extraction module and the fusion module are integrated to obtain a feature adaptive attention model, wherein the feature adaptive attention model is used to adaptively select different convolution kernels to extract and shuffle channel features of different scales. The first feature extraction module is connected to the second feature extraction module and the feature shuffling module, the second feature extraction module is connected to the feature shuffling module, the feature shuffling module is connected to the third feature extraction module and the fusion module. The first feature extraction module includes a first 1x1 convolution layer, and the first 1x1 convolution layer is used to perform feature extraction on the multiple sets of historical image data to obtain a first feature map; the second feature extraction module includes a 5x5 depth-separable convolution layer, a selective kernel network layer and a 9x9 depth-separable convolution layer, and the 5x5 depth-separable convolution layer is used to perform deep feature extraction on the first feature map to obtain a first feature map. The selective kernel network layer is used to assign weights to the first feature map to obtain a third feature map, and the 9x9 depth-separable convolution layer is used to perform deep feature extraction on the first feature map to obtain a fourth feature map; the feature shuffling module includes a chunk() function, and the chunk() function is used to shuffle the first feature map, the second feature map, the third feature map, and the fourth feature map to obtain a fifth feature map; the third feature extraction module includes a second 1x1 convolution layer and a third 1x1 convolution layer connected in sequence, the second 1x1 convolution layer is used to perform feature extraction on the fifth feature map to obtain a sixth feature map, and the third 1x1 convolution layer is used to perform feature extraction on the sixth feature map to obtain a seventh feature map; the fusion module includes a Concat layer, and the Concat layer is used to fuse the fifth feature map and the seventh feature map to obtain an eighth feature map; Fusing the fast pyramid pooling module in the backbone network of the YOLOv6 network model and the feature adaptive attention model to obtain an initial fusion model; Deleting the cross-stage local modules in the initial fusion model to obtain the initial prediction model; The multiple groups of historical image data are input into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model.

2. The training method according to claim 1, characterized in that Before inputting the multiple sets of historical image data into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model, the training method further includes: Dividing the multiple groups of historical image data to obtain a training set and a test set; Performing data enhancement processing on the training set to obtain an enhanced training set; The inputting the multiple sets of historical image data into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model includes: The enhanced training set is input into the initial prediction model for training to obtain the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model.

3. The training method according to claim 2, characterized in that Also includes: Determine whether the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model converges based on the test set.

4. A method for predicting the probability of occurrence of pulmonary hypertension, characterized in that: include: Acquiring current imaging data of the patient, wherein the current imaging data includes a current pulmonary artery angiography dynamic image, a current pulmonary vascular image, and a current electrocardiogram; Inputting the current image data into a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model to obtain a corresponding current chronic thromboembolic pulmonary hypertension occurrence probability; Wherein, the target chronic thromboembolic pulmonary hypertension occurrence probability prediction model is obtained by training using the training method according to any one of claims 1 to 3.

5. A training device for a pulmonary hypertension probability prediction model, characterized in that: include: a historical data acquisition module, configured to acquire multiple sets of historical image data, each set of which includes a user's historical pulmonary angiography dynamic images, historical pulmonary vascular images, and historical electrocardiograms, and each set of which is annotated with a historical probability of occurrence of chronic thromboembolic pulmonary hypertension; A fusion module is used to fuse the first feature extraction module, the second feature extraction module, the feature shuffling module, the third feature extraction module and the fusion module to obtain a feature adaptive attention model, wherein the feature adaptive attention model is used to adaptively select different convolution kernels to extract and shuffle channel features of different scales. The first feature extraction module is connected to the second feature extraction module and the feature shuffling module, the second feature extraction module is connected to the feature shuffling module, and the feature shuffling module is connected to the third feature extraction module and the fusion module. The first feature extraction module includes a first 1x1 convolution layer, which is used to extract features from the multiple sets of historical image data to obtain a first feature map; the second feature extraction module includes a 5x5 depth-separable convolution layer, a selective kernel network layer and a 9x9 depth-separable convolution layer, and the 5x5 depth-separable convolution layer is used to perform deep feature extraction on the first feature map. A second feature map is obtained, the selective kernel network layer is used to assign weights to the first feature map to obtain a third feature map, the 9x9 depth-separable convolution layer is used to perform deep feature extraction on the first feature map to obtain a fourth feature map; the feature shuffling module includes a chunk() function, and the chunk() function is used to shuffle the first feature map, the second feature map, the third feature map, and the fourth feature map to obtain a fifth feature map; the third feature extraction module includes a second 1x1 convolution layer and a third 1x1 convolution layer connected in sequence, the second 1x1 convolution layer is used to perform feature extraction on the fifth feature map to obtain a sixth feature map, and the third 1x1 convolution layer is used to perform feature extraction on the sixth feature map to obtain a seventh feature map; the fusion module includes a Concat layer, and the Concat layer is used to fuse the fifth feature map and the seventh feature map to obtain an eighth feature map; A fusion unit, configured to fuse the fast pyramid pooling module in the backbone network of the YOLOv6 network model and the feature adaptive attention model to obtain an initial fusion model; a deletion unit, configured to delete the cross-stage local modules in the initial fusion model to obtain the initial prediction model; The training module is used to input the multiple groups of historical image data into the initial prediction model for training to obtain a target chronic thromboembolic pulmonary hypertension occurrence probability prediction model.

6. An electronic device, characterized in that: include: Memory; one or more processors coupled to the memory; One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by one or more processors, and the one or more applications are configured to execute the training method according to any one of claims 1 to 3, and / or the prediction method according to claim 4.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, which can be called by a processor to execute the training method according to any one of claims 1 to 3 and / or the prediction method according to claim 4.

Images