Medical image recognition method and device, training method and device, equipment and storage medium

Abstract

The invention relates to the technical field of image recognition, and provides a medical image recognition method and device, a training method and device, equipment and a storage medium, and the method comprises the steps: obtaining a medical image; based on a feature extraction network, extracting image features of the medical image; determining a target sub-model from a preset medical image recognition model according to the image features of the medical image; the medical image recognition model comprises a plurality of sub-models, and each sub-model is used for recognizing image features of a corresponding disease category; and based on the target sub-model, identifying image features of the medical image to obtain a disease category corresponding to the medical image. The method can be applied to service systems such as medical treatment, health, old-age care and the like, and the target sub-model which is good at processing the medical image is determined from the multiple sub-models of the medical image recognition model according to the image features of the medical image, so that the calculation complexity in the medical image processing process can be reduced, and the recognition efficiency of the model is improved; and the expandability of the model can be enhanced.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and in particular to medical image recognition methods, training methods, devices, equipment and storage media. Background Art

[0002] In the business scenarios of healthcare and elderly care, the pathological features corresponding to medical images are relatively complex and diverse. The trained model can identify medical images and obtain the disease category corresponding to the medical images. For example, the trained model can identify pathological features such as nodules, tumors, and bleeding locations in CT images and obtain the disease category corresponding to the CT image.

[0003] In the related technology, it is impossible to effectively capture the complex features in medical images and the computational complexity of model training is large, resulting in low accuracy and efficiency of model recognition of medical images. Summary of the invention

[0004] The main purpose of this application is to provide a medical image recognition method, training method, device, equipment and storage medium, so that each medical image can be effectively processed by the corresponding sub-model, thereby improving the accuracy of identifying disease categories, which can not only reduce the computational complexity of the medical image processing process and improve the recognition efficiency of the model; but also enhance the scalability of the model.

[0005] In a first aspect, the present application provides a medical image recognition method, comprising:

[0006] Access to medical images;

[0007] Extracting image features of the medical image based on a feature extraction network;

[0008] According to the image features of the medical image, a target sub-model is determined from a preset medical image recognition model; the medical image recognition model includes a plurality of sub-models, each of which is used to recognize the image features of a corresponding disease category;

[0009] Based on the target sub-model, the image features of the medical image are identified to obtain the disease category corresponding to the medical image.

[0010] In a second aspect, the present application further provides a training method for a medical image recognition model, wherein the medical image recognition model includes a plurality of sub-models, and the training method includes:

[0011] Acquire multiple training samples, wherein the training samples include medical images and disease category labels corresponding to the medical images;

[0012] Extracting image features of the medical image based on a feature extraction network;

[0013] Determining a target sub-model from the multiple sub-models according to the image features of the medical image, wherein the image features of the medical images with the same disease category label correspond to the same target sub-model;

[0014] Based on the target sub-model, image features of the medical image with the disease category label are identified to obtain the disease category;

[0015] Based on a preset loss function, determining a loss value corresponding to the target sub-model according to the disease category and the disease category label;

[0016] According to the loss value, adjusting the model parameters of the target sub-model;

[0017] The medical image recognition model is updated according to the model parameters of the target sub-model.

[0018] In a third aspect, the present application further provides a medical image recognition device, comprising:

[0019] An acquisition module, used for acquiring medical images;

[0020] A feature extraction module, used for extracting image features of the medical image based on a feature extraction network;

[0021] A model determination module, used to determine a target sub-model from a preset medical image recognition model according to the image features of the medical image; the medical image recognition model includes a plurality of sub-models, each of which is used to identify image features corresponding to a disease category;

[0022] The recognition module is used to recognize the image features of the medical image based on the target sub-model to obtain the disease category corresponding to the medical image.

[0023] In a fourth aspect, the present application further provides a computer device, the computer device comprising a memory and a processor;

[0024] The memory is used to store computer programs;

[0025] The processor is used to execute the computer program and implement the medical image recognition method and the training method of the medical image recognition model as described above when executing the computer program.

[0026] In a fifth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the medical image recognition method and the steps of the medical image recognition model training method as described above are implemented.

[0027] The present application provides a medical image recognition method, training method, device, equipment and storage medium, wherein the method includes: obtaining a medical image; extracting the image features of the medical image based on a feature extraction network; determining a target sub-model from a preset medical image recognition model according to the image features of the medical image; the medical image recognition model includes multiple sub-models, each of which is used to identify the image features of the corresponding disease category; based on the target sub-model, the image features of the medical image are identified to obtain the disease category corresponding to the medical image. The medical image recognition model in the present application includes multiple sub-models that can identify medical images corresponding to different disease categories, and according to the image features of the medical image, a target sub-model that is better at processing the medical image can be determined from the multiple sub-models of the medical image recognition model, so that each medical image can be effectively processed by the corresponding sub-model, thereby improving the accuracy of identifying the disease category, which can reduce the computational complexity in the medical image processing process, improve the recognition efficiency of the model, and enhance the scalability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of 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 some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0029] Figure 1 A schematic diagram of a flow chart of a medical image recognition method provided in an embodiment of the present application;

[0030] Figure 2 A schematic diagram of the connection between the server and the terminal device provided in the embodiment of the present application;

[0031] Figure 3 A schematic diagram of a flow chart of a method for training a medical image recognition model provided in an embodiment of the present application;

[0032] Figure 4 A schematic block diagram of a medical image recognition device provided in an embodiment of the present application;

[0033] Figure 5 A schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0034] 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 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 ordinary technicians in this field without creative work are within the scope of protection of this application.

[0035] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0036] The embodiments of the present application provide a medical image recognition method, training method, apparatus, device and storage medium. Among them, the medical image recognition method can be applied to a terminal device, which can be a mobile phone, a tablet computer, a laptop computer, a desktop computer and other devices. It can also be applied to a server, which can be a separate server or a 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 (Content Delivery Network, CDN), and big data and artificial intelligence platforms.

[0037] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0038] See also Figure 1 , Figure 1 This is a flow chart of a medical image recognition method provided in an embodiment of the present application. It should be noted that the medical image recognition method provided in an embodiment of the present application can be used in a terminal device, and of course can also be used in a server.

[0039] like Figure 2 As shown, the medical image recognition method is applied to a server, and the server and the terminal device are connected in communication, and the disease category obtained by the medical image recognition method can be sent to the terminal device through the server. Of course, it is not limited to this and is not limited here.

[0040] In specific implementation, the terminal device includes but is not limited to: any one of a mobile phone, a tablet computer, a laptop computer, and a desktop computer; the server can be a single server or a server cluster, or a cloud server that provides cloud computing services.

[0041] like Figure 1As shown, the medical image recognition method includes steps S101 to S104.

[0042] Step S101: Acquire medical images.

[0043] For example, medical images may include X-ray images, computer tomography images (CT images), magnetic resonance imaging images (MRI images), ultrasound images, digital subtraction angiography images (DSA images), and endoscopic images.

[0044] Different medical images can be used to diagnose different types of diseases. Specifically, CT images can include images of the lungs, head, maxillofacial area, etc., for the diagnosis of different diseases. For example, lung CT images can be used to diagnose diseases such as lung tumors, nodules, hemorrhage, and pneumonia. Maxillofacial CT images can be used to diagnose the shape, position, and degree of damage of teeth.

[0045] Step S102: extracting image features of medical images based on a feature extraction network.

[0046] It is understandable that after acquiring the medical image, the embodiment of the present application needs to pre-process the medical image first, for example, adjusting the size, contrast and saturation of the medical image; normalizing the pixel values ​​in the medical image; performing data enhancement operations such as rotating, flipping, and cropping on the medical image; so as to increase data diversity and improve the clarity of the medical image.

[0047] For example, the embodiment of the present application can use a convolutional neural network to extract image features corresponding to medical images. It should be noted that a convolutional neural network is a deep learning model for processing image data, which may include a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer can be used to extract local features such as edges and textures of medical images; the pooling layer can be used to reduce the dimension of local features corresponding to medical images to reduce the amount of calculation; the fully connected layer can be used to extract the extracted image features as input to the medical image recognition model for the classification of disease categories.

[0048] Specifically, the preprocessed medical image (X-ray image, CT image, MRI image or endoscope image) is input into the convolution layer, and the local features of the medical image are extracted through multiple convolution layers. For example, the low-level convolution kernel can extract low-level features such as edges, textures, spots, etc. of the medical image, and the deep convolution kernel can extract more advanced features, such as lesion areas and organ structures. The extracted image features are input into the pooling layer to reduce the dimension of the image features using maximum pooling or average pooling. Finally, the image features extracted by the convolution layer are flattened and input into the fully connected layer.

[0049] Taking lung CT images as an example, the low-level convolution kernels of the convolutional neural network can extract low-level features such as the boundary between the lung and the chest wall, the texture difference between normal lung tissue and diseased tissue, and local bright or dark spots in the lung CT images. The middle-level convolution kernels can extract the local shape of blood vessels and bronchi; local lesions such as nodules and masses; and the local structure of lung lobes and segments. The deep convolution kernels can extract lesion areas such as tumors, infection areas, or fibrosis areas; the shape and distribution of the entire lung; and high-level features such as benign and malignant lesion types.

[0050] Step S103: determining a target sub-model from a preset medical image recognition model according to the image features of the medical image; the medical image recognition model includes a plurality of sub-models, each sub-model being used to recognize image features of a corresponding disease category.

[0051] For example, the medical image recognition model in the embodiment of the present application may include multiple sub-models, each of which is used to identify image features of a corresponding disease category. For example, the first sub-model is used to identify the location and size of lung nodules; the second sub-model is used to detect pneumonia; the third sub-model is used to detect lung tumors, etc. Different sub-models correspond to different recognition tasks, that is, to recognize image features of different disease categories.

[0052] The embodiment of the present application determines the activated target sub-model from multiple sub-models of the medical image recognition model according to the image features of the extracted medical image. Specifically, if the image feature of the extracted medical image is a tumor, the sub-model for detecting lung tumors can be activated; if the image feature of the extracted medical image is pneumonia, the sub-model for detecting pneumonia can be activated.

[0053] The embodiment of the present application can determine, based on the image features of the medical image, a target sub-model from multiple sub-models that is better at processing the image features, so as to effectively capture the complex features in the medical image and ensure the accuracy of model recognition.

[0054] Step S104: Based on the target sub-model, the image features of the medical image are identified to obtain the disease category corresponding to the medical image.

[0055] The multiple sub-models in the embodiments of the present application are all core components of the medical image recognition model, and each sub-model is an independent model that is specifically responsible for processing the image features of a specific disease category. Exemplarily, after determining the target sub-model based on the image features of the extracted medical image, the image features of the medical image can be input into the target sub-model, and after extracting high-level features based on the hidden layer of the target sub-model, the probability distribution of the disease category is output, thereby obtaining the disease category corresponding to the medical image.

[0056] For example, the sub-models in the medical image recognition model may include deep learning models such as convolutional neural networks, recurrent neural networks, and generative adversarial networks; they may also include models based on quantum computing; they may also include multimodal fusion models.

[0057] The embodiment of the present application combines multiple sub-models and assigns the extracted image features of the medical image to the corresponding sub-models for processing, which can flexibly process complex medical image data and improve the accuracy of disease category identification.

[0058] The medical image recognition method provided in the above embodiment includes: obtaining a medical image; extracting the image features of the medical image based on a feature extraction network; determining a target sub-model from a preset medical image recognition model according to the image features of the medical image; the medical image recognition model includes multiple sub-models, each sub-model is used to identify the image features of the corresponding disease category; based on the target sub-model, the image features of the medical image are identified to obtain the disease category corresponding to the medical image. The medical image recognition model in the embodiment of the present application includes multiple sub-models that can identify medical images corresponding to different disease categories, and according to the image features of the medical image, a target sub-model that is better at processing the medical image can be determined from the multiple sub-models of the medical image recognition model, so that each medical image can be effectively processed by the corresponding sub-model, thereby improving the accuracy of identifying the disease category, which can reduce the computational complexity in the medical image processing process, improve the recognition efficiency of the model, and enhance the scalability of the model.

[0059] In an exemplary implementation, step S103 may include step S1031 and step S1032.

[0060] Step S1031 : Based on a preset gating network and according to image features of the medical image, determine the gating scores corresponding to each of the multiple sub-models in the medical image recognition model.

[0061] Step S1032: determining a target sub-model from the multiple sub-models according to the gating scores corresponding to the multiple sub-models in the medical image recognition model.

[0062] For example, the gating network can assign a corresponding gating score, i.e., a weight, to each sub-model by learning the image features of the input medical image, and then determine the sub-model to be activated, i.e., the target sub-model, based on the weight. It should be noted that the gating network in the embodiment of the present application can be a fully connected layer. Through the gating network, the sub-model most relevant to the image features of the medical image can be dynamically selected as the target sub-model, thereby improving the recognition efficiency of the model. Since only a few sub-models need to be activated to identify the image features of the medical image, the amount of computation in the medical image processing process can be reduced.

[0063] Specifically, the extracted image features of the medical image are input into the gating network to obtain an unnormalized weight vector (i.e., gating score), the dimension of which is the number of sub-models; the gating score is then normalized to obtain a probability distribution, which includes the weight of each sub-model; then, based on Top-K selection (i.e., sparsity constraint), K sub-models with higher weights are selected from multiple weights of the probability distribution as target sub-models, where K is an integer greater than or equal to 1.

[0064] In an exemplary implementation, step S103 may further include step S1033.

[0065] Step S1033: adding Laplace noise to the gated scores corresponding to each of the multiple sub-models in the medical image recognition model.

[0066] Correspondingly, step S1032 may specifically include: determining a target sub-model from the multiple sub-models according to the gating scores to which Laplace noise is added corresponding to each of the multiple sub-models.

[0067] It is understandable that Laplace noise is one of the core mechanisms of differential privacy, which is random noise sampled from the Laplace distribution. By adding Laplace noise, sensitive information can be prevented from being leaked.

[0068] Specifically, after the gating score of each sub-model is output through the gating network, independent Laplace noise is added to each gating score; then based on Top-K selection (i.e., sparsity constraint), the sub-models corresponding to the higher K gating scores are selected as the target sub-models from the multiple gating scores with Laplace noise added, where K is an integer greater than or equal to 1.

[0069] For example, after the image features of the medical image are input into the gated network to obtain the gated scores of each sub-model, the embodiment of the present application can add Laplace noise to the gated scores corresponding to the sub-model to hide the specific information of the sub-model and protect data privacy, so that attackers cannot infer the input image features. Then, a sub-model with a higher score is selected as the target sub-model from the gated scores with Laplace noise added to reduce the possibility of attackers inferring the input image features through routing decisions.

[0070] In an exemplary embodiment, the number of target sub-models includes at least two; step S104 may include step S1041 and step S1042.

[0071] Step S1041: input the image features of the medical image into at least two target sub-models to obtain the category information output by each target sub-model.

[0072] Step S1042: Obtain the disease category corresponding to the medical image according to the category information output by each target sub-model.

[0073] It is understandable that the target sub-model may include at least two. After the image features of the medical image are input into at least two target sub-models, the category information (i.e., disease category probability) output by each target sub-model can be obtained; after the category information output by the target sub-model is weighted and summed, the final disease category probability can be obtained; then, the disease category with the highest probability is selected from the disease category probabilities as the disease category corresponding to the medical image. The embodiment of the present application can combine the prediction results of multiple sub-models to obtain a more accurate disease category probability distribution by weighted summing the disease category probabilities output by at least two target sub-models, thereby improving the flexibility of the model.

[0074] See also Figure 3 , Figure 3 A flowchart of a training method for a medical image recognition model provided in an embodiment of the present application. The medical image recognition model in the embodiment of the present application includes multiple sub-models, and the training method includes steps S201 to S207.

[0075] Step S201: Acquire multiple training samples, where the training samples include medical images and disease category labels corresponding to the medical images.

[0076] It is understandable that the embodiments of the present application can collect a large number of training samples under the premise of obtaining legal authorization and without involving user privacy, and each training sample may include a medical image (for example, an X-ray image, a CT image, an MRI image, or an endoscopic image), and a disease type label corresponding to the medical image.

[0077] Step S202: extracting image features of medical images based on a feature extraction network.

[0078] Step S203: determining a target sub-model from the multiple sub-models according to the image features of the medical images, wherein the image features of the medical images with the same disease category label correspond to the same target sub-model.

[0079] Step S204: Based on the target sub-model, the image features of the medical image with the disease category label are identified to obtain the disease category.

[0080] It should be noted that the relevant discussion of step S202 to step S204 can refer to the relevant embodiments of the aforementioned medical image recognition method, which will not be repeated here.

[0081] Step S205: Based on a preset loss function, determine the loss value corresponding to the target sub-model according to the disease category and the disease category label.

[0082] Step S206: adjust the model parameters of the target sub-model according to the loss value.

[0083] It is understandable that the preset loss function is mainly in the training stage of multiple sub-models of the medical image recognition model. After each batch of training samples is input into the corresponding sub-model, the predicted value (i.e., the identified disease category) can be output through forward propagation, and then the predicted value and the true value, i.e., the difference between the disease category and the disease category label, that is, the loss value, can be calculated according to the loss function. After obtaining the loss value, the corresponding sub-model can update the model parameters through back propagation to reduce the loss between the true value and the predicted value, so that the predicted value output by the sub-model is closer to the true value, thereby achieving the purpose of learning and improving the accuracy of model recognition. Specifically, the embodiment of the present application can be based on the preset loss function, after the image features of the medical image are identified through the target sub-model to obtain the disease category, the difference between the disease category and the disease category label corresponding to the medical image can be calculated to obtain the loss value corresponding to the target sub-model, and then the model parameters corresponding to the target sub-model are updated according to the loss value.

[0084] Step S207: Update the medical image recognition model according to the model parameters of the target sub-model.

[0085] For example, the medical image recognition model includes multiple sub-models. When training the medical image recognition model, only a few sub-models can be trained. For example, according to the routing selection algorithm, only the sub-model for detecting bronchi and the sub-model for detecting pneumonia may be activated during the training process, while the sub-model for identifying tumors will not be activated. Therefore, after the target sub-model is trained, it is necessary to update the medical image recognition model according to the model parameters of the target sub-model, that is, only the target sub-model that has participated in the training needs to be uploaded to the medical image recognition model, and other untrained (activated) sub-models do not need to be uploaded to the medical image recognition model, thereby reducing the data transmission load.

[0086] The training method of the medical image recognition model provided in the above embodiment includes: obtaining multiple training samples, the training samples include medical images and disease category labels corresponding to the medical images; extracting image features of the medical images based on a feature extraction network; determining a target sub-model from multiple sub-models based on the image features of the medical images, wherein the image features of the medical images with the same disease category labels correspond to the same target sub-model; based on the target sub-model, identifying the image features of the medical images with the disease category labels to obtain the disease category; based on a preset loss function, determining the loss value corresponding to the target sub-model based on the disease category and the disease category label; adjusting the model parameters of the target sub-model based on the loss value; and updating the medical image recognition model based on the model parameters of the target sub-model. The embodiment of the present application divides the global model of the medical image recognition model into multiple sub-models, each of which is used to process medical images with corresponding disease category labels, without the need to fully train the entire model, thereby reducing the amount of calculation and data transmission load.

[0087] Furthermore, during the training of the medical image recognition model, the gradient information of the trained target sub-model can be dynamically quantized and compressed to reduce communication overhead. Specifically, the gradient information generally includes specific parameters of the model such as the convolution kernel weights for detecting shadows and the weights of the fully connected layer. The importance of the gradient information of the target sub-model can be evaluated so that when compressing the gradient information, the unimportant auxiliary gradient information can be compressed as much as possible and the key gradient information can be retained, thereby ensuring the stable progress of the training.

[0088] For dynamic task allocation, the training tasks that need to be completed by different medical, health and elderly care institutions can be evaluated based on the differences in the sub-models uploaded by each terminal device each time and the characteristics of the data features.

[0089] In another embodiment, a sub-model can also be created in the terminal device as a local exclusive model. The local exclusive model does not need to be uploaded to the global model. It is designed and trained by the terminal device itself, does not participate in the global update, and is only used for local detail fine-tuning to meet personalized needs.

[0090] In an exemplary implementation, step S207 may include step S2071 and step S2072.

[0091] Step S2071: Aggregate the model parameters corresponding to the multiple target sub-models to obtain aggregated model parameters.

[0092] Step S2072: Update the parameters corresponding to the multiple target sub-models in the medical image recognition model according to the aggregated model parameters.

[0093] For example, the trained target sub-models can be marked to determine the sub-models that need to be uploaded to the medical image recognition model. For example, the target sub-model A after the image features of the medical image corresponding to the lung nodule are recognized and processed is marked as 1, so that the target sub-model A can be quickly identified and uploaded to the global model for global aggregation.

[0094] Specifically, the model parameters corresponding to multiple target sub-models can be weighted averaged through a weighted averaging algorithm to obtain aggregated model parameters; the aggregated model parameters are then updated to the medical image recognition model to replace the parameters of the original corresponding sub-models; then, the parameters of all sub-models are loaded according to the preset model architecture order; finally, the global model of the medical image recognition model can be compressed to reduce the overall communication volume and ensure stable training.

[0095] In an exemplary embodiment, the training method may further include step S100 and step S200.

[0096] Step S100: deploy multiple sub-models of the medical image recognition model to their corresponding terminal devices.

[0097] Step S200: assigning permission levels to terminal devices corresponding to each of the multiple sub-models in the deployed medical image recognition model.

[0098] The embodiment of the present application can deploy multiple sub-models of the medical image recognition model to different terminal devices. And assign permission levels to the terminal devices corresponding to each sub-model in the training process. For example, terminal device A with a higher permission level can access terminal devices with more important sub-models deployed, while terminal device B with a lower permission level can only access terminal devices with more basic sub-models deployed. In this way, the access of low-privilege terminal devices to terminal devices with sub-models of sensitive data deployed can be restricted to further improve the security of data.

[0099] See also Figure 4 , Figure 4 A schematic block diagram of a medical image recognition device provided in an embodiment of the present application. The medical image recognition device can be configured in a server or a terminal device to execute the aforementioned medical image recognition method.

[0100] like Figure 4 As shown, the medical image recognition device includes: an acquisition module 110 , a feature extraction module 120 , a model determination module 130 and a recognition module 140 .

[0101] The acquisition module 110 is used to acquire medical images.

[0102] The feature extraction module 120 is used to extract image features of medical images based on a feature extraction network.

[0103] The model determination module 130 is used to determine a target sub-model from a preset medical image recognition model according to the image features of the medical image; the medical image recognition model includes multiple sub-models, each sub-model is used to identify the image features of the corresponding disease category.

[0104] The recognition module 140 is used to recognize the image features of the medical image based on the target sub-model to obtain the disease category corresponding to the medical image.

[0105] In an exemplary embodiment, the model determination module 130 may include a score determination submodule and a model determination submodule.

[0106] The score determination submodule is used to determine the gating scores corresponding to each of the multiple sub-models in the medical image recognition model based on a preset gating network and according to the image features of the medical image.

[0107] The model determination submodule is used to determine a target submodel from multiple submodels in the medical image recognition model according to the gating scores corresponding to each of the multiple submodels.

[0108] In an exemplary embodiment, the apparatus may further include a noise adding submodule.

[0109] The noise adding submodule is used to add Laplace noise to the gated scores corresponding to each of the multiple sub-models in the medical image recognition model.

[0110] Correspondingly, the model determination submodule can be specifically used to determine the target submodel from multiple submodels according to the gating scores with Laplace noise added corresponding to each of the multiple submodels.

[0111] In an exemplary embodiment, the number of target sub-models includes at least two; the recognition module 140 may include a recognition sub-module and a category determination sub-module.

[0112] The recognition submodule is used to input the image features of the medical image into at least two target submodels to obtain the category information output by each target submodel.

[0113] The category determination submodule is used to obtain the disease category corresponding to the medical image based on the category information output by each target submodel.

[0114] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0115] The method of the present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0116] Exemplarily, the above method and apparatus may be implemented in the form of a computer program, which may be run on a computer device.

[0117] See also Figure 5 , Figure 5 A schematic block diagram of the structure of a computer device provided in an embodiment of the present application. The computer device may be a server or a terminal device.

[0118] like Figure 5 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a storage medium and an internal memory.

[0119] The storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute the steps of any one of the medical image recognition methods and the steps of the training method of the medical image recognition model.

[0120] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0121] The internal memory provides an environment for the operation of the computer program in the storage medium. When the computer program is executed by the processor, the processor can execute the steps of any one of the medical image recognition methods and the steps of the medical image recognition model training method.

[0122] The network interface is used for network communication, such as sending assigned tasks.

[0123] Those skilled in the art will understand that Figure 5The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0124] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0125] In one embodiment, the processor is used to execute a computer program and can implement the following steps when executing the computer program:

[0126] Access to medical images;

[0127] Extract image features of medical images based on feature extraction network;

[0128] According to the image features of the medical image, a target sub-model is determined from a preset medical image recognition model; the medical image recognition model includes a plurality of sub-models, each sub-model is used to recognize the image features of a corresponding disease category;

[0129] Based on the target sub-model, the image features of the medical image are identified to obtain the disease category corresponding to the medical image.

[0130] Correspondingly, the processor is used to execute the computer program and can also implement the following steps when executing the computer program:

[0131] Obtain multiple training samples, where the training samples include medical images and disease category labels corresponding to the medical images;

[0132] Extract image features of medical images based on feature extraction network;

[0133] Determine a target sub-model from multiple sub-models according to image features of the medical image, wherein image features of medical images with the same disease category label correspond to the same target sub-model;

[0134] Based on the target sub-model, the image features of the medical images with disease category labels are identified to obtain the disease category;

[0135] Based on the preset loss function, the loss value corresponding to the target sub-model is determined according to the disease category and the disease category label;

[0136] According to the loss value, adjust the model parameters of the target sub-model;

[0137] Update the medical image recognition model according to the model parameters of the target sub-model.

[0138] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process of medical image recognition described above can refer to the corresponding process in the embodiment of the aforementioned medical image recognition method, and the specific training process of the medical image recognition model described above can refer to the corresponding process in the embodiment of the aforementioned medical image recognition model training method, which will not be repeated here.

[0139] The present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps can be implemented:

[0140] Access to medical images;

[0141] Extract image features of medical images based on feature extraction network;

[0142] According to the image features of the medical image, a target sub-model is determined from a preset medical image recognition model; the medical image recognition model includes a plurality of sub-models, each sub-model is used to recognize the image features of a corresponding disease category;

[0143] Based on the target sub-model, the image features of the medical image are identified to obtain the disease category corresponding to the medical image.

[0144] Correspondingly, when the computer program is executed by the processor, the following steps may also be implemented:

[0145] Obtain multiple training samples, where the training samples include medical images and disease category labels corresponding to the medical images;

[0146] Extract image features of medical images based on feature extraction network;

[0147] Determine a target sub-model from multiple sub-models according to image features of the medical image, wherein image features of medical images with the same disease category label correspond to the same target sub-model;

[0148] Based on the target sub-model, the image features of the medical images with disease category labels are identified to obtain the disease category;

[0149] Based on the preset loss function, the loss value corresponding to the target sub-model is determined according to the disease category and the disease category label;

[0150] According to the loss value, adjust the model parameters of the target sub-model;

[0151] Update the medical image recognition model according to the model parameters of the target sub-model.

[0152] The computer-readable storage medium may be an internal storage unit of the computer device of the aforementioned embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device.

[0153] It should be noted that the functions or steps that can be implemented by the above-mentioned computer-readable storage medium can correspond to the embodiments of the aforementioned medical image recognition method and the embodiments of the training method of the medical image recognition model.

[0154] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0155] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0156] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A medical image recognition method, characterized in that: include: Access to medical images; Extracting image features of the medical image based on a feature extraction network; According to the image features of the medical image, a target sub-model is determined from a preset medical image recognition model; the medical image recognition model includes a plurality of sub-models, each of which is used to recognize the image features of a corresponding disease category; Based on the target sub-model, the image features of the medical image are identified to obtain the disease category corresponding to the medical image.

2. The medical image recognition method according to claim 1, characterized in that: Determining a target sub-model from a preset medical image recognition model according to the image features of the medical image includes: Based on a preset gating network, and according to image features of the medical image, determining gating scores corresponding to each of the plurality of sub-models in the medical image recognition model; According to the gating scores corresponding to the multiple sub-models in the medical image recognition model, a target sub-model is determined from the multiple sub-models.

3. The medical image recognition method according to claim 2, characterized in that: The step of determining a target sub-model from a preset medical image recognition model according to the image features of the medical image further includes: Adding Laplace noise to the gated scores corresponding to each of the multiple sub-models in the medical image recognition model; The step of determining a target sub-model from the multiple sub-models according to the gating scores corresponding to the multiple sub-models in the medical image recognition model comprises: A target sub-model is determined from the multiple sub-models according to the gating scores to which Laplace noise is added corresponding to each of the multiple sub-models.

4. The medical image recognition method according to any one of claims 1 to 3, characterized in that: The number of the target sub-models includes at least two; the identifying the image features of the medical image based on the target sub-model to obtain the disease category corresponding to the medical image includes: Inputting the image features of the medical image into at least two of the target sub-models to obtain category information output by each of the target sub-models; According to the category information output by each of the target sub-models, the disease category corresponding to the medical image is obtained.

5. A training method for a medical image recognition model, characterized in that: The medical image recognition model includes multiple sub-models, and the training method includes: Acquire multiple training samples, wherein the training samples include medical images and disease category labels corresponding to the medical images; Extracting image features of the medical image based on a feature extraction network; Determining a target sub-model from the multiple sub-models according to the image features of the medical image, wherein the image features of the medical images with the same disease category label correspond to the same target sub-model; Based on the target sub-model, image features of the medical image with the disease category label are identified to obtain the disease category; Based on a preset loss function, determining a loss value corresponding to the target sub-model according to the disease category and the disease category label; According to the loss value, adjusting the model parameters of the target sub-model; The medical image recognition model is updated according to the model parameters of the target sub-model.

6. The training method of the medical image recognition model according to claim 5, characterized in that: The updating of the medical image recognition model according to the model parameters of the target sub-model includes: Aggregating model parameters corresponding to each of the plurality of target sub-models to obtain aggregated model parameters; According to the aggregated model parameters, the parameters corresponding to the multiple target sub-models in the medical image recognition model are updated.

7. The training method of the medical image recognition model according to claim 5, characterized in that: Also includes: Deploy multiple sub-models of the medical image recognition model to their respective corresponding terminal devices; Assign permission levels to terminal devices corresponding to each of the multiple sub-models in the medical image recognition model.

8. A medical image recognition device, characterized in that: include: An acquisition module, used for acquiring medical images; A feature extraction module, used for extracting image features of the medical image based on a feature extraction network; A model determination module, used to determine a target sub-model from a preset medical image recognition model according to the image features of the medical image; the medical image recognition model includes a plurality of sub-models, each of which is used to identify image features corresponding to a disease category; The recognition module is used to recognize the image features of the medical image based on the target sub-model to obtain the disease category corresponding to the medical image.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the medical image recognition method as described in any one of claims 1 to 4 and the training method of the medical image recognition model as described in any one of claims 5 to 7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the medical image recognition method described in any one of claims 1 to 4 and the steps of the medical image recognition model training method described in any one of claims 5 to 7 are implemented.

Images