Model training method and device, computer equipment, storage medium and program product

Abstract

The invention relates to a model training method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the steps of obtaining demand description information and a training sample set based on an interaction page, determining an initial medical image processing model and a hyper-parameter of the initial medical image processing model according to the demand description information, and obtaining a medical image processing model according to the training sample set and the hyper-parameter. And training the initial medical image processing model to obtain a target medical image processing model. By adopting the method, the training efficiency of the medical image processing model can be improved.

Description

Technical Field

[0001] This application relates to the field of model training technology, and in particular to a model training method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] With the continuous development of computer technology, various industries have begun to use computer capabilities to assist in work processing. In the medical industry, pre-trained medical image processing models can be used to classify medical images, saving medical workers a lot of work. However, training medical image processing models requires medical workers to master computer technology, which presents certain technical obstacles.

[0003] In existing technologies, medical image processing models are mainly trained through collaboration between medical professionals and computer technicians, with medical professionals repeatedly consulting with computer technicians. This results in low training efficiency for medical image processing models. Summary of the Invention

[0004] Therefore, it is necessary to provide a model training method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the training efficiency of medical image processing models in response to the above-mentioned technical problems.

[0005] In a first aspect, this application provides a model training method, comprising: obtaining requirement description information and training sample set based on an interactive page, wherein the requirement description information represents the training requirements of a medical image processing model; determining an initial medical image processing model and its hyperparameters based on the requirement description information; and training the initial medical image processing model based on the training sample set and the hyperparameters to obtain a target medical image processing model.

[0006] In one embodiment, determining an initial medical image processing model and its hyperparameters based on the requirement description information includes: identifying image type information, hyperparameter requirement information, and region information included in the requirement description information; selecting an initial medical image processing model from multiple candidate medical image processing models based on the image type information and region information; and determining hyperparameters based on the hyperparameter requirement information, wherein the hyperparameters include at least one of learning rate, training epochs, and target loss value.

[0007] In one embodiment, an initial medical image processing model is selected from multiple candidate medical image processing models based on image type information and region information. This includes: obtaining the identification information of each candidate medical image processing model; for each identification information, determining a first similarity between the identification information and image type information, and a second similarity between the identification information and region information; determining a third similarity based on the first and second similarities; determining the target identification information corresponding to the target similarity greater than a similarity threshold in each third similarity; and selecting the initial medical image processing model corresponding to the target identification information from among the candidate medical image processing models.

[0008] In one embodiment, the process of obtaining the training sample set includes: obtaining multiple initial sample medical images and anomaly type labels corresponding to each initial sample medical image based on the interactive page; performing privacy-de-identifying processing on the initial sample medical images to obtain privacy-de-identifying initial sample medical images, and performing image enhancement on the privacy-de-identifying initial sample medical images to obtain each sample medical image; and constructing a training sample set based on each sample medical image and anomaly type labels.

[0009] In one embodiment, there are multiple initial medical image processing models. The initial medical image processing models are trained based on a training sample set and hyperparameters to obtain a target medical image processing model. This includes: performing a preset number of initial training rounds on each initial medical image processing model based on the training sample set and hyperparameters to obtain each first medical image processing model corresponding to each initial medical image processing model; selecting a second medical image processing model from among the first medical image processing models based on their training metrics, wherein the matching degree between the training metrics and hyperparameters of the second medical image processing model is greater than a preset matching degree threshold; and training the second medical image processing model based on the training sample set and hyperparameters to obtain the target medical image processing model.

[0010] In one embodiment, the method further includes: obtaining a trained image processing model, freezing the convolutional layers of the image processing model and adding a classification layer to the image processing model to obtain a first image processing model; training the first image processing model according to preset general medical images to obtain an initial medical image processing model, wherein the general medical images include medical images that match the requirement description information.

[0011] Secondly, this application also provides a model training apparatus, comprising: a training sample acquisition module, used to acquire requirement description information and a training sample set based on an interactive page, wherein the requirement description information represents the training requirements of a medical image processing model; a training data determination module, used to determine an initial medical image processing model and its hyperparameters based on the requirement description information; and a model training module, used to train the initial medical image processing model based on the training sample set and hyperparameters to obtain a target medical image processing model.

[0012] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in the first aspect.

[0013] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0014] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect above.

[0015] The aforementioned model training method, apparatus, computer equipment, computer-readable storage medium, and computer program product acquire user-submitted requirement description information and training sample set based on an interactive page. The requirement description data characterizes the user's training requirements for the medical image processing model. Based on the requirement description information, the initial medical image processing model and its hyperparameters are determined. The initial medical image processing model and hyperparameters to be trained are determined through computer recognition, improving model training efficiency. Based on the training sample set and hyperparameters, the initial medical image processing model is trained to obtain the target medical image processing model. The automatic model training via computer equipment based on the input requirement description information and training sample set further enhances the training efficiency of the medical image processing model. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a diagram illustrating the application environment of a model training method in one embodiment;

[0018] Figure 2 This is a flowchart illustrating a model training method in one embodiment;

[0019] Figure 3 This is a flowchart illustrating step 202 in one embodiment;

[0020] Figure 4 This is a flowchart illustrating step 302 in one embodiment;

[0021] Figure 5 This is a schematic diagram illustrating the process of obtaining the training sample set in one embodiment;

[0022] Figure 6 This is a flowchart illustrating step 203 in one embodiment;

[0023] Figure 7 This is a schematic diagram of the training process of the initial medical image processing model in one embodiment;

[0024] Figure 8 This is a flowchart illustrating the model training method in another embodiment;

[0025] Figure 9 This is a structural block diagram of a model training device in one embodiment;

[0026] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0028] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0029] The model training method provided in this application embodiment can be applied to, for example... Figure 1 The application environment shown includes at least terminal 101 and server 102.

[0030] Terminal 101 is used for user interaction. Users can upload training sample sets and requirement descriptions to Terminal 101. Terminal 101 can also be used to display model training results to users. Terminal 101 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. Users can be institutional users of medical institutions.

[0031] Server 102 receives the training sample set and requirement description information sent by terminal 101, determines the initial medical image processing model and hyperparameters based on the requirement description information, and trains the initial medical image processing model based on the training sample set and hyperparameters to obtain the target medical image processing model. The initial medical image processing model can be stored in server 102 or in an external database of server 102. Server 102 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. Terminal 101 communicates with server 102 via a network. It should be noted that server 102 can also be configured with human-computer interaction functions and / or human-computer interaction pages.

[0032] In one exemplary embodiment, such as Figure 2 As shown, a model training method is provided, which can be applied to... Figure 1 The following steps, 201 to 203, are used as an example to illustrate the process of using a server in the example.

[0033] In real-world scenarios, technologies such as artificial intelligence and computer processing have been applied to various industries, improving the work efficiency of staff in these industries. Among them, image processing technology is used by various industries due to its intuitive and convenient advantages. For example, in the medical industry, medical institutions can use image processing technology to classify medical images, thereby improving the efficiency of medical image classification.

[0034] However, institutional users in medical institutions typically lack computer skills and are unable to independently develop and configure image processing algorithms and models. Meanwhile, computer technicians often lack the ability to screen sample sets and configure training labels, resulting in low efficiency in the development of medical image processing models. To address this, this application automatically identifies user requirements and obtains labeled training sample sets. Based on the requirements and training sample sets, the system automatically trains the model to obtain a target medical image processing model that matches the user's requirements. This achieves automatic training according to the needs of institutional users, improving the training efficiency of medical image processing models.

[0035] Step 201: Obtain the requirement description information and training sample set based on the interactive page.

[0036] In this application, the interactive page refers to the front-end page provided by the server for interaction with institutional users of medical institutions. Institutional users of medical institutions can enter requirement description information in the interactive area, and can also upload training sample sets on the interactive page.

[0037] In real-world scenarios, medical images typically have regional or year-specific type information, and the recognition and / or classification results for different input images may also differ. Therefore, institutional users in medical institutions can pre-store sample images in the same folder and name the folder, using the name as a label. Institutional users can also edit their training requirements for the model as a requirement description and input it into the interactive page to indicate the training strategy for the model.

[0038] The requirement description information can be textual data. Institutional users of medical institutions can describe the training requirements of the medical image processing model through the requirement description information. The requirement description information represents the training requirements of the medical image processing model and may include at least one of the following: region information, image type of input image, model application scenario, time limit for model training, resource limit for model training, model sensitivity, and model response speed.

[0039] The training sample set may include multiple sample images required for training and corresponding labels for the sample images. The labels may be the classification results and / or recognition results of the sample images, and the sample images may be authorized medical images obtained by institutional users of medical institutions. Optionally, medical images include ocular medical images.

[0040] During implementation, institutional users can upload their requirements description information and training sample set in advance on the interactive page and trigger the confirmation control on the interactive page to start model training. The server then obtains the requirements description information and training sample set.

[0041] In real-world scenarios, the training sample set can be pre-input. To improve the efficiency of model training, image preprocessing can be performed on the sample images in the training sample set. Optionally, the training sample set includes preprocessed sample images. During execution, the server can obtain the initial training sample set uploaded by the user, preprocess the initial sample images in the initial training sample set to obtain sample images, and construct the training sample set based on the sample images.

[0042] Step 202: Based on the requirements description information, determine the initial medical image processing model and its hyperparameters.

[0043] In real-world scenarios, a server can store multiple initial models that have undergone basic training. These initial models can be image processing models that have undergone basic training, or initial models that have been trained with basic image processing functions such as convolutional layers, feature extraction layers, and feature recognition layers. Furthermore, the initial models can also be trained on general medical images to obtain initial medical image processing models. Multiple initial medical image processing models can be stored in the server's model library to perform targeted training on the initial medical image processing models that have undergone basic training, ensuring the accuracy of model recognition while improving model training efficiency.

[0044] To address this, this application determines the initial medical image processing model to be trained based on the requirement description information. During implementation, the server performs semantic recognition on the requirement description information to obtain multiple model requirement information and multiple hyperparameter requirement information. The initial medical image processing model is determined based on the model requirement information, and the hyperparameters are determined using the multiple hyperparameter requirement information. Optionally, the model requirement information includes image type information of the input image, model function information, and / or region information. The hyperparameter requirement information includes learning rate, training epochs, target loss value, and / or recall rate.

[0045] During execution, the server can perform matching processing in the model library based on the semantic recognition results of the requirement description information to obtain an initial medical image processing model that matches the requirement description information. There can be one or more initial medical image processing models. Optionally, there can be multiple initial medical image processing models.

[0046] In this application, hyperparameters refer to parameters that need to be set before model training begins to control the model training process and structure. Hyperparameters may include at least one of the following: number of network layers, number of iterations, learning rate, number of training epochs, and target loss value.

[0047] In the process of determining hyperparameters, the hyperparameter requirements can be determined based on the hyperparameter requirements in the requirement description information. For example, if the hyperparameter requirement information is "high sensitivity", it can indicate that the institution user has a high requirement for the response speed in the model recognition process. Based on this, the number of network layers can be determined. Alternatively, it can indicate that the user has a short requirement for the model training process time. Based on this, the learning rate can be determined.

[0048] Step 203: Based on the training sample set and hyperparameters, train the initial medical image processing model to obtain the target medical image processing model.

[0049] During implementation, the server inputs sample images from the training sample set into the initial medical image processing model, obtains the output of the initial medical image processing model, calculates the training loss based on the results and labels, adjusts the parameters of the initial medical image processing model based on the training loss, and repeats the above process until the training indicators of the model meet the hyperparameters, thus obtaining the target medical image processing model.

[0050] During execution, the parameter step size during model training can be determined based on the learning rate, the number of parameter tuning attempts during model training can be determined based on the training rounds, and the training termination condition during model training can be determined based on the target loss value.

[0051] In the above model training method, the user-submitted requirement description information and training sample set are obtained based on the interactive page. The requirement description data represents the user's training requirements for the medical image processing model. Based on the requirement description information, the initial medical image processing model and its hyperparameters are determined. The initial medical image processing model and hyperparameters to be trained are determined by computer recognition, which improves the efficiency of model training. Based on the training sample set and hyperparameters, the initial medical image processing model is trained to obtain the target medical image processing model. The model training is automatically performed by computer equipment based on the input requirement description information and training sample set, which improves the training efficiency of the medical image processing model.

[0052] Based on the above exemplary embodiment, the following provides a model training method in one or more exemplary embodiments, which is applied to... Figure 1 Taking the server in the example, the following content will be used for explanation.

[0053] During the identification of requirement description information, model requirement information and hyperparameter requirement information included in the requirement description information can be identified. An initial medical image processing model is determined based on the model requirement information, and hyperparameters are determined based on the hyperparameter requirement information. In one optional implementation provided by this application, such as... Figure 3 As shown, step 202 includes steps 301 to 303:

[0054] Step 301: Identify the image type information, hyperparameter requirement information, and region information included in the requirement description information.

[0055] During implementation, the server performs semantic segmentation and keyword extraction on the requirement description information to obtain keywords corresponding to the image type information of the input image, keywords corresponding to the hyperparameter requirement information, and keywords corresponding to the region information.

[0056] During execution, the requirement description information can be segmented based on the punctuation marks in the requirement description information, and it can also be identified through semantic recognition and morpheme recognition.

[0057] For example, if the requirement description includes "use fundus color photography data", then the image type of the data image can be determined to be fundus color photography data.

[0058] Step 302: Select an initial medical image processing model from multiple candidate medical image processing models based on image type information and region information.

[0059] During implementation, the server can match the image type and region information with the identification information of multiple candidate medical image processing models in the model library, and select the initial medical image processing model based on the matching results.

[0060] During implementation, the identifiers corresponding to image type information and region information can be queried. The identifiers can be matched with the identifiers of multiple candidate medical image processing models in the model library, and the initial medical image processing model with a matching degree greater than a threshold can be selected. Alternatively, the similarity between image type information and region information and the identifiers of multiple candidate medical image processing models in the model library can be calculated, and the initial medical image processing model with a similarity greater than a threshold can be selected.

[0061] Step 303: Determine the hyperparameters based on the hyperparameter requirement information.

[0062] During implementation, hyperparameters corresponding to hyperparameter requirements can be preset in the data table. The server queries the data table for hyperparameter requirements to obtain hyperparameters that match the requirement description. Hyperparameters include at least one of the following: learning rate, number of training epochs, and target loss value.

[0063] For example, if the requirement description includes "rapid deployment is required", the learning rate will be increased and the number of training rounds will be reduced.

[0064] One optional implementation provided in this application achieves automatic model training by identifying demand description information and determining the initial medical image processing model and hyperparameters based on the demand description information. This eliminates the need for medical institution members to configure model parameters, thereby improving the efficiency of model training.

[0065] In the initial medical image processing model selection process, to improve the accuracy and reliability of model selection, the identification information of each candidate medical image processing model can be obtained. For each identification information, the similarity between the identification information and the model requirement information is calculated. The initial medical image processing corresponding to the target identification information with a similarity greater than a similarity threshold is determined. In one optional implementation provided by this application, such as… Figure 4 As shown, step 302 includes steps 401 to 402:

[0066] Step 401: Obtain the identification information of each candidate medical image processing model. For each identification information, determine the first similarity between the identification information and the image type information, and the second similarity between the identification information and the region information. Determine the third similarity based on the first and second similarities.

[0067] In this application, the identification information of the medical image processing model is information used to characterize the functional classification of the medical image processing model. There can be multiple identification information, which can be used to characterize the type of input data, the type of output data, the target area, etc. of the medical image processing model.

[0068] During implementation, the server acquires the identification information of each candidate medical image processing model. For each identification information of each candidate medical image processing model, it determines the first similarity between the representation information and the image type information, and calculates the second similarity between the identification information and the region information. Based on the first and second similarities, a third similarity is determined. The third similarity characterizes the degree of matching between the requirement description information and the candidate medical image processing model.

[0069] During execution, multiple first similarities of candidate medical image processing models can be integrated into one first similarity, and multiple second similarities can be integrated into one second similarity. The third similarity can be determined based on the first and second similarities. Alternatively, multiple third similarities can be calculated and integrated into one third similarity.

[0070] Step 402: Determine the target identification information corresponding to the target similarity greater than the similarity threshold in each third similarity, and select the initial medical image processing model corresponding to the target identification information from each candidate medical image processing model.

[0071] During implementation, the server detects target similarities greater than the similarity threshold in each third similarity, determines the target identification information corresponding to the target similarity, and selects the initial medical image processing model corresponding to the target identification information from among the candidate medical image processing models in the model library.

[0072] In one optional implementation provided by this application, model selection is performed by calculating the similarity between the requirement description information and each candidate medical image processing model in the model library, thereby improving the reliability and accuracy of the initial medical image processing model.

[0073] In real-world scenarios, medical images from healthcare institutions may contain some private information. To address this, the sample images in the training sample set can first undergo privacy-de-sensitizing processing, and then a training sample set can be constructed based on the privacy-de-sensitizing sample images. One optional implementation provided in this application is as follows: Figure 5 As shown, the process of obtaining the training sample set includes steps 501 to 503:

[0074] Step 501: Obtain multiple initial sample medical images and the corresponding abnormality type labels for each initial sample medical image based on the interactive page.

[0075] During implementation, medical institutions can submit multiple initial sample medical images and corresponding abnormality type labels in advance through an interactive page. The server then obtains the multiple initial sample medical images and abnormality type labels.

[0076] During the process, institutional users can package the initial sample medical images and upload them to the interactive page, then rename the packaged images to the text corresponding to the abnormality type label; alternatively, institutional users can upload the initial sample medical images sequentially, and each initial sample medical image can be labeled with an abnormality type label.

[0077] Step 502: Perform privacy-de-identification processing on the initial sample medical images to obtain privacy-de-identified initial sample medical images, and perform image enhancement on the privacy-de-identified initial sample medical images to obtain individual sample medical images.

[0078] In real-world scenarios, to improve model training performance and efficiency, image enhancement can be performed on sample images. In this application, after de-privacy processing of the initial sample medical image, image enhancement is performed on the de-privacy processed image to obtain the sample medical image.

[0079] During implementation, the server removes the identification information, time information and / or institutional information contained in the initial sample medical image to obtain a privacy-de-identified initial sample medical image, and then performs image enhancement on the privacy-de-identified initial sample medical image to obtain the sample medical image.

[0080] During execution, image sharpness enhancement, resolution enhancement, and artifact removal can be performed on the initial privacy-de-scarce medical images. Furthermore, feature enhancement processing can be applied to the initial privacy-de-scarce medical images based on the requirements description to improve the effectiveness of model training.

[0081] Step 503: Construct a training sample set based on the medical images and abnormality type labels of each sample.

[0082] During implementation, the server constructs a training sample set based on the medical images and abnormality type labels of each sample, so that the model can be trained based on the training sample set after the institution members instruct the model to train.

[0083] One optional implementation provided in this application involves de-privacy processing of initial sample medical images uploaded by institutional users, which improves the purity and security of the sample images. At the same time, image enhancement is performed on the de-privacy images to improve the effectiveness of the sample images, thereby improving the effectiveness of model training.

[0084] In real-world scenarios, there can be multiple initial medical image processing models that meet the required description information. The optimal model can be selected using the Adaptive Model Racing Protocol (AMRP). Initial training can be performed on each initial medical image processing model. A second medical image processing model is selected based on the training metrics of the initial training. The second medical image processing model is then trained using the training sample set and hyperparameters to obtain the target medical image processing model. In one optional implementation provided in this application, such as… Figure 6 As shown, step 203 includes steps 601 to 603:

[0085] Step 601: Based on the training sample set and hyperparameters, perform initial training on each initial medical image processing model for a preset number of rounds to obtain each first medical image processing model corresponding to each initial medical image processing model.

[0086] This application selects multiple initial medical image processing models and then filters them after initial training. This method uses fewer training rounds to select the model that best matches the requirement description information, saving training resources while improving the usability of the target medical image processing model.

[0087] During implementation, the server performs a preset number of initial training rounds on each initial medical image processing model based on the training sample set and hyperparameters, and obtains the corresponding first medical image processing model after the initial training, as well as the training indicators during the initial training process.

[0088] In this application, the training metrics can be the indicator data of each initial medical image processing model during the initial training process. The training metrics can characterize the training status of each initial medical image processing model. The training metrics can include at least one of recall, accuracy, response speed and training cost.

[0089] Step 602: Based on the training metrics of each first medical image processing model, select the second medical image processing model from among the first medical image processing models.

[0090] During implementation, the server matches the training metrics of each first medical image processing model with the requirement description information to determine the training metrics that best match the requirement description information. Based on the training metrics, a second medical image processing model is selected from each first medical image processing model. The matching degree between the training metrics and hyperparameters of the second medical image processing model is greater than the preset matching degree threshold.

[0091] For example, during the warm-up phase (2-3 rounds of training), the stability of the model is evaluated, and candidate models with stagnant loss are eliminated; finally, the model that performs best on the target metric (such as accuracy) is selected for deep fine-tuning.

[0092] Step 603: Based on the training sample set and hyperparameters, train the second medical image processing model to obtain the target medical image processing model.

[0093] During implementation, the server continues to train the selected second medical image processing model based on the training sample set and hyperparameters to obtain the target medical image processing model. The specific training process can be referred to the above embodiment, and will not be repeated here.

[0094] One optional implementation provided in this application improves the effectiveness and reliability of the target medical image processing model by screening among multiple initial medical image processing models. At the same time, the screening method after initial training also saves training resources.

[0095] In real-world scenarios, image processing models often contain common components, such as general convolutional layers, feature extraction layers, and feature computation layers. Transfer learning can be used to freeze the parameters of these common model layers and add a classification layer. Training the model with the added classification layer yields an initial medical image processing model, which can then quickly generate multiple candidate medical image processing models from a model library. One optional implementation provided in this application is as follows: Figure 7 As shown, the training process of the initial medical image processing model includes steps 701 to 702:

[0096] Step 701: Obtain the trained image processing model, freeze the convolutional layers of the image processing model, and add a classification layer to the image processing model to obtain the first image processing model.

[0097] During implementation, the server can select a pre-trained image processing model, freeze the convolutional layers and / or feature extraction layers of the image processing model, and directly use the pre-trained convolutional layers and / or feature extraction layers to add a classification layer to the image processing module to obtain the first image processing model.

[0098] Step 702: Train the first image processing model based on the preset general medical images to obtain the initial medical image processing model.

[0099] During implementation, the server trains the first image processing model based on preset general medical images to train the classification layer of the first image processing model into the medical image domain, thereby obtaining an initial medical image processing model. The general medical images include medical images that match the requirement description information.

[0100] For example, by using transfer learning, based on the general features of pre-trained models, only a small amount of labeled data is needed to achieve the performance of traditional platforms with full data, thus adapting to scenarios where data is scarce.

[0101] One optional implementation method provided in this application is to quickly train an initial medical image processing model in a model library through transfer learning. Using the parameters of the trained image processing model can improve the training efficiency of the model while ensuring its accuracy.

[0102] In one embodiment, see Figure 8 The document illustrates a flowchart of a model training method provided in an embodiment of this application, which can be applied to... Figure 1 In the server shown. For example Figure 8 As shown, the model training method may include the following steps:

[0103] Step 801: Obtain the requirement description information and initial training sample set based on the interactive page.

[0104] Step 802: Identify the image type information, hyperparameter requirement information, and region information included in the requirement description information.

[0105] Step 803: Based on image type information and region information, select an initial medical image processing model from multiple candidate medical image processing models.

[0106] Step 804: Determine the hyperparameters based on the hyperparameter requirement information.

[0107] Step 805: Perform privacy-de-identifying processing on the initial sample medical images in the initial training sample set to obtain privacy-de-identifying initial sample medical images.

[0108] Step 806: Perform image enhancement on the initial sample medical images after privacy removal to obtain the medical images of each sample.

[0109] Step 807: Construct a training sample set based on the medical images and abnormality type labels of each sample.

[0110] Step 808: Based on the training sample set and hyperparameters, perform initial training on each initial medical image processing model for a preset number of rounds to obtain each first medical image processing model corresponding to each initial medical image processing model.

[0111] Step 809: Based on the training metrics of each first medical image processing model, select the second medical image processing model from among the first medical image processing models.

[0112] Optionally, the matching degree between the training metrics and hyperparameters of the second medical image processing model is greater than a preset matching degree threshold.

[0113] Step 810: Based on the training sample set and hyperparameters, train the second medical image processing model to obtain the target medical image processing model.

[0114] It should be noted that any one or more of steps 801 to 810 can be combined to form a new implementation method according to the needs of implementation and deployment. Furthermore, any one or more technical features in the technical solution composed of steps 801 to 810 can also be combined to form a new implementation method according to the actual deployment needs, or technical features in one or more optional implementation methods provided by one or more of the above embodiments can be combined to form a new implementation method. These will not be elaborated on here.

[0115] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0116] Based on the same inventive concept, this application also provides a model training apparatus for implementing the model training method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more model training apparatus embodiments provided below can be found in the limitations of the model training method described above, and will not be repeated here.

[0117] In one exemplary embodiment, such as Figure 9 As shown, a model training device is provided, including: a training sample acquisition module 901, a training data determination module 902, and a model training module 903, wherein: the training sample acquisition module 901 is used to acquire requirement description information and a training sample set based on an interactive page, the requirement description information representing the training requirements of the medical image processing model; the training data determination module 902 is used to determine the initial medical image processing model and its hyperparameters based on the requirement description information; the model training module 903 is used to train the initial medical image processing model based on the training sample set and hyperparameters to obtain the target medical image processing model.

[0118] In one embodiment, the training data determination module 902 includes a requirement identification unit, a model selection unit, and a hyperparameter determination unit, wherein: the requirement identification unit is used to identify the image type information, hyperparameter requirement information, and region information included in the requirement description information; the model selection unit is used to select an initial medical image processing model from multiple candidate medical image processing models based on the image type information and region information; and the hyperparameter determination unit is used to determine hyperparameters based on the hyperparameter requirement information, wherein the hyperparameters include at least one of learning rate, training epochs, and target loss value.

[0119] In one embodiment, the model screening unit includes a similarity calculation module and a model selection module, wherein: the similarity calculation module is used to obtain the identification information of each candidate medical image processing model, and for each identification information, determine the first similarity between the identification information and the image type information, and the second similarity between the identification information and the region information, and determine the third similarity based on the first similarity and the second similarity; the model selection module is used to determine the target identification information corresponding to the target similarity greater than the similarity threshold in each third similarity, and select the initial medical image processing model corresponding to the target identification information from each candidate medical image processing model.

[0120] In one embodiment, the device further includes an initial sample acquisition module, a privacy removal module, and a training sample set construction module, wherein: the initial sample acquisition module is used to acquire multiple initial sample medical images and anomaly type labels corresponding to each initial sample medical image based on an interactive page; the privacy removal module is used to perform privacy removal processing on the initial sample medical images to obtain privacy-removed initial sample medical images, and to perform image enhancement on the privacy-removed initial sample medical images to obtain each sample medical image; the training sample set construction module is used to construct a training sample set based on each sample medical image and anomaly type labels.

[0121] In one embodiment, the model training module 903 includes an initial training unit, a second model selection unit, and a second model training unit, wherein: the initial training unit is used to perform a preset number of initial training rounds on each initial medical image processing model according to the training sample set and hyperparameters to obtain each first medical image processing model corresponding to each initial medical image processing model; the second model selection unit is used to select a second medical image processing model from each first medical image processing model according to the training indicators of each first medical image processing model, wherein the matching degree between the training indicators and hyperparameters of the second medical image processing model is greater than a preset matching degree threshold; the second model training unit is used to train the second medical image processing model according to the training sample set and hyperparameters to obtain a target medical image processing model.

[0122] In one embodiment, the apparatus further includes a first transfer learning module and a second transfer learning module, wherein: the first transfer learning module is used to acquire the trained image processing model, freeze the convolutional layers of the image processing model and add a classification layer to the image processing model to obtain a first image processing model; the second transfer learning module is used to train the first image processing model according to preset general medical images to obtain an initial medical image processing model, wherein the general medical images include medical images that match the requirement description information.

[0123] Each module in the aforementioned model training device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0124] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a model training method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0125] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0126] In one exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: obtaining requirement description information and a training sample set based on an interactive page, wherein the requirement description information characterizes the training requirements of a medical image processing model; determining an initial medical image processing model and its hyperparameters based on the requirement description information; and training the initial medical image processing model based on the training sample set and the hyperparameters to obtain a target medical image processing model.

[0127] In one embodiment, when the processor executes the computer program, it specifically implements the following steps: identifying the image type information, hyperparameter requirement information, and region information included in the requirement description information; selecting an initial medical image processing model from multiple candidate medical image processing models based on the image type information and region information; and determining hyperparameters based on the hyperparameter requirement information, wherein the hyperparameters include at least one of learning rate, training epochs, and target loss value.

[0128] In one embodiment, when the processor executes the computer program, it specifically implements the following steps: obtaining the identification information of each candidate medical image processing model; for each identification information, determining the first similarity between the identification information and the image type information, and the second similarity between the identification information and the region information; determining the third similarity based on the first and second similarities; determining the target identification information corresponding to the target similarity greater than the similarity threshold in each third similarity; and selecting the initial medical image processing model corresponding to the target identification information from among the candidate medical image processing models.

[0129] In one embodiment, when the processor executes the computer program, it further implements the following steps: acquiring multiple initial sample medical images and anomaly type labels corresponding to each initial sample medical image based on an interactive page; performing privacy-de-sensitizing processing on the initial sample medical images to obtain privacy-de-sensitizing initial sample medical images, and performing image enhancement on the privacy-de-sensitizing initial sample medical images to obtain each sample medical image; and constructing a training sample set based on each sample medical image and anomaly type labels.

[0130] In one embodiment, when the processor executes the computer program, it specifically implements the following steps: performing initial training on each initial medical image processing model for a preset number of rounds based on the training sample set and hyperparameters to obtain each first medical image processing model corresponding to each initial medical image processing model; selecting a second medical image processing model from among the first medical image processing models based on the training metrics of each first medical image processing model, wherein the matching degree between the training metrics and hyperparameters of the second medical image processing model is greater than a preset matching degree threshold; and training the second medical image processing model based on the training sample set and hyperparameters to obtain the target medical image processing model.

[0131] In one embodiment, when the processor executes the computer program, it further performs the following steps: obtaining a trained image processing model, freezing the convolutional layers of the image processing model and adding a classification layer to the image processing model to obtain a first image processing model; training the first image processing model according to a preset general medical image to obtain an initial medical image processing model, wherein the general medical image includes medical images that match the requirement description information.

[0132] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it performs the following steps: obtaining requirement description information and a training sample set based on an interactive page, wherein the requirement description information characterizes the training requirements of a medical image processing model; determining an initial medical image processing model and its hyperparameters based on the requirement description information; and training the initial medical image processing model based on the training sample set and the hyperparameters to obtain a target medical image processing model.

[0133] In one embodiment, when the computer program is executed by the processor, it specifically implements the following steps: identifying the image type information, hyperparameter requirement information, and region information included in the requirement description information; selecting an initial medical image processing model from multiple candidate medical image processing models based on the image type information and region information; and determining hyperparameters based on the hyperparameter requirement information, wherein the hyperparameters include at least one of learning rate, training epochs, and target loss value.

[0134] In one embodiment, when the computer program is executed by the processor, it specifically implements the following steps: obtaining the identification information of each candidate medical image processing model; for each identification information, determining the first similarity between the identification information and the image type information, and the second similarity between the identification information and the region information; determining the third similarity based on the first and second similarities; determining the target identification information corresponding to the target similarity greater than the similarity threshold in each third similarity; and selecting the initial medical image processing model corresponding to the target identification information from among the candidate medical image processing models.

[0135] In one embodiment, when the computer program is executed by the processor, it specifically implements the following steps: acquiring multiple initial sample medical images and anomaly type labels corresponding to each initial sample medical image based on the interactive page; performing privacy-de-sensitivity processing on the initial sample medical images to obtain privacy-de-sensitivity initial sample medical images, and performing image enhancement on the privacy-de-sensitivity initial sample medical images to obtain each sample medical image; and constructing a training sample set based on each sample medical image and anomaly type labels.

[0136] In one embodiment, when the computer program is executed by the processor, it specifically implements the following steps: performing initial training on each initial medical image processing model for a preset number of rounds based on the training sample set and hyperparameters to obtain each first medical image processing model corresponding to each initial medical image processing model; selecting a second medical image processing model from among the first medical image processing models based on the training metrics of each first medical image processing model, wherein the matching degree between the training metrics and hyperparameters of the second medical image processing model is greater than a preset matching degree threshold; and training the second medical image processing model based on the training sample set and hyperparameters to obtain the target medical image processing model.

[0137] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining a trained image processing model, freezing the convolutional layers of the image processing model and adding a classification layer to the image processing model to obtain a first image processing model; training the first image processing model according to preset general medical images to obtain an initial medical image processing model, wherein the general medical images include medical images that match the requirement description information.

[0138] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps: obtaining requirement description information and a training sample set based on an interactive page, wherein the requirement description information characterizes the training requirements of a medical image processing model; determining an initial medical image processing model and its hyperparameters based on the requirement description information; and training the initial medical image processing model based on the training sample set and the hyperparameters to obtain a target medical image processing model.

[0139] In one embodiment, when the computer program is executed by the processor, it specifically implements the following steps: identifying the image type information, hyperparameter requirement information, and region information included in the requirement description information; selecting an initial medical image processing model from multiple candidate medical image processing models based on the image type information and region information; and determining hyperparameters based on the hyperparameter requirement information, wherein the hyperparameters include at least one of learning rate, training epochs, and target loss value.

[0140] In one embodiment, when the computer program is executed by the processor, it specifically implements the following steps: obtaining the identification information of each candidate medical image processing model; for each identification information, determining the first similarity between the identification information and the image type information, and the second similarity between the identification information and the region information; determining the third similarity based on the first and second similarities; determining the target identification information corresponding to the target similarity greater than the similarity threshold in each third similarity; and selecting the initial medical image processing model corresponding to the target identification information from among the candidate medical image processing models.

[0141] In one embodiment, when the computer program is executed by the processor, it specifically implements the following steps: acquiring multiple initial sample medical images and anomaly type labels corresponding to each initial sample medical image based on the interactive page; performing privacy-de-sensitivity processing on the initial sample medical images to obtain privacy-de-sensitivity initial sample medical images, and performing image enhancement on the privacy-de-sensitivity initial sample medical images to obtain each sample medical image; and constructing a training sample set based on each sample medical image and anomaly type labels.

[0142] In one embodiment, when the computer program is executed by the processor, it specifically implements the following steps: performing initial training on each initial medical image processing model for a preset number of rounds based on the training sample set and hyperparameters to obtain each first medical image processing model corresponding to each initial medical image processing model; selecting a second medical image processing model from among the first medical image processing models based on the training metrics of each first medical image processing model, wherein the matching degree between the training metrics and hyperparameters of the second medical image processing model is greater than a preset matching degree threshold; and training the second medical image processing model based on the training sample set and hyperparameters to obtain the target medical image processing model.

[0143] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining a trained image processing model, freezing the convolutional layers of the image processing model and adding a classification layer to the image processing model to obtain a first image processing model; training the first image processing model according to preset general medical images to obtain an initial medical image processing model, wherein the general medical images include medical images that match the requirement description information.

[0144] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0145] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0146] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0147] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A model training method, characterized in that, The method includes: The requirement description information and training sample set are obtained based on the interactive page. The requirement description information represents the training requirements of the medical image processing model. Based on the requirements description information, determine the initial medical image processing model and the hyperparameters of the initial medical image processing model; The initial medical image processing model is trained based on the training sample set and the hyperparameters to obtain the target medical image processing model.

2. The method according to claim 1, characterized in that, The step of determining the initial medical image processing model and its hyperparameters based on the requirement description information includes: The requirement description information includes image type information, hyperparameter requirement information, and region information; Based on the image type information and the region information, the initial medical image processing model is selected from multiple candidate medical image processing models; The hyperparameters are determined based on the hyperparameter requirement information, and the hyperparameters include at least one of learning rate, training epochs, and target loss value.

3. The method according to claim 2, characterized in that, The step of selecting the initial medical image processing model from multiple candidate medical image processing models based on the image type information and the region information includes: Obtain the identification information of each candidate medical image processing model; for each identification information, determine the first similarity between the identification information and the image type information, and the second similarity between the identification information and the region information; and determine the third similarity based on the first similarity and the second similarity. The target identification information corresponding to the target similarity greater than the similarity threshold in each of the third similarity is determined, and the initial medical image processing model corresponding to the target identification information is selected from each of the candidate medical image processing models.

4. The method according to claim 1, characterized in that, The process of obtaining the training sample set includes: Based on the interactive page, multiple initial sample medical images and the corresponding abnormality type labels for each initial sample medical image are obtained; The initial sample medical images are subjected to privacy-de-identification processing to obtain privacy-de-identification initial sample medical images, and the privacy-de-identification initial sample medical images are then enhanced to obtain individual sample medical images. The training sample set is constructed based on the medical images of each sample and the abnormality type label.

5. The method according to claim 1, characterized in that, The number of initial medical image processing models is multiple. The step of training the initial medical image processing models based on the training sample set and the hyperparameters to obtain the target medical image processing model includes: Based on the training sample set and the hyperparameters, each initial medical image processing model is initially trained for a preset number of rounds to obtain each first medical image processing model corresponding to each initial medical image processing model. Based on the training metrics of each of the first medical image processing models, a second medical image processing model is selected from each of the first medical image processing models, wherein the matching degree between the training metrics and the hyperparameters of the second medical image processing model is greater than a preset matching degree threshold. The second medical image processing model is trained based on the training sample set and the hyperparameters to obtain the target medical image processing model.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: Obtain the trained image processing model, freeze the convolutional layers of the image processing model, and add a classification layer to the image processing model to obtain the first image processing model; The first image processing model is trained based on preset general medical images to obtain the initial medical image processing model, wherein the general medical images include medical images that match the requirement description information.

7. A model training device, characterized in that, The device includes: The training sample acquisition module is used to acquire requirement description information and training sample set based on the interactive page. The requirement description information represents the training requirements of the medical image processing model. The training data determination module is used to determine the initial medical image processing model and the hyperparameters of the initial medical image processing model based on the requirement description information. The model training module is used to train the initial medical image processing model based on the training sample set and the hyperparameters to obtain the target medical image processing model.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images