Model training method and device, computer equipment and readable storage medium

Abstract

The invention relates to a model training method and device, computer equipment and a readable storage medium, and the method comprises the steps: determining an initial model corresponding to a target task in response to a selection operation for the target task; in response to an acquisition operation of a training sample corresponding to the target task, determining a target training sample set; and in response to a training operation on the initial model, training the initial model according to the target training sample set to obtain a target model. By using the model training method provided by the invention, the user can adaptively adjust the deployed model, so that the model can meet the requirements of the user.

Description

Technical Field

[0001] The present application relates to the field of medical technology, and in particular to a model training method, apparatus, computer equipment and readable storage medium. Background Art

[0002] With the development of deep learning, artificial neural networks have been widely used in medical image processing, including medical image segmentation and classification. One of the most important applications is intelligent quality control, that is, medical image quality control based on artificial neural networks. Intelligent quality control can automatically judge the quality problems in medical images through pre-trained classification networks, so as to reduce image quality problems and improve inspection efficiency.

[0003] In traditional technology, after the server trains the intelligent quality control model, it deploys the intelligent quality control model in the medical equipment, and the user directly uses the intelligent quality control model.

[0004] However, in traditional technologies, the intelligent quality control model deployed in medical equipment is a pre-trained unified model that cannot meet the different needs of different users. Summary of the invention

[0005] Based on this, it is necessary to provide a model training method, device, computer equipment and readable storage medium that can meet the needs of different users in response to the above technical problems.

[0006] In a first aspect, the present application provides a model training method, the method comprising:

[0007] In response to a selection operation of a target task, determining an initial model corresponding to the target task;

[0008] In response to an operation of acquiring training samples corresponding to a target task, a target training sample set is determined; wherein the target training sample set includes an original training sample set and a labeled training sample set;

[0009] In response to the training operation on the initial model, the initial model is trained according to the target training sample set to obtain a target model.

[0010] In one embodiment, if the initial model is a trained model, in response to an operation of acquiring training samples corresponding to the target task, determining a target training sample set includes:

[0011] In response to an operation of acquiring training samples corresponding to the target task, determining an original training sample set;

[0012] In response to the use operation of the initial model, the original training sample set is input into the initial model to obtain an output result; and the labeled training sample set is determined according to the output result.

[0013] In one embodiment, determining a labeled training sample set according to the output result includes:

[0014] In response to a confirmation operation on the output result, if it is determined that the output result meets the preset requirement, the output result is determined as a label of the labeled training sample set, and the labeled training sample set is determined according to the label of the labeled training sample set and the original training sample set;

[0015] If it is determined that the output result does not meet the preset requirement, a labeled training sample set is obtained in response to a labeling operation on the original training sample set.

[0016] In one embodiment, if the initial model is an untrained model, in response to an operation of acquiring training samples corresponding to a target task, determining a target training sample set includes:

[0017] In response to an operation of acquiring training samples corresponding to the target task, determining an original training sample set;

[0018] In response to the labeling operation on the original training sample set, a labeled training sample set is obtained.

[0019] In one embodiment, in response to an operation of acquiring training samples corresponding to a target task, determining the original training sample set includes:

[0020] In response to a selection operation on stored training samples in a database, the selected stored training sample set is determined as an original training sample set; wherein the stored training samples include scan data acquired in different time periods.

[0021] In one embodiment, the method further comprises:

[0022] In response to an upload operation on the target model and / or the target training sample set, the target model and / or the target training sample set is uploaded to the cloud.

[0023] In one embodiment, in response to a selection operation of a target task, determining an initial model corresponding to the target task includes:

[0024] In response to a selection operation of a plurality of different types of models stored in the cloud, the selected model is determined as an initial model corresponding to the target task;

[0025] In response to an operation of acquiring a training sample corresponding to a target task, determining a target training sample set includes:

[0026] In response to a selection operation on a training sample stored in the cloud, the selected training sample is determined as a target training sample set corresponding to the target task.

[0027] In a second aspect, the present application also provides a model training device, the device comprising:

[0028] A selection module, configured to determine an initial model corresponding to the target task in response to a selection operation on the target task;

[0029] An acquisition module, configured to determine a target training sample set in response to an acquisition operation of training samples corresponding to a target task; wherein the target training sample set includes an original training sample set and a labeled training sample set;

[0030] The training operation is used to respond to the training operation on the initial model and train the initial model according to the target training sample set to obtain the target model.

[0031] In a third aspect, the present application further provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method provided in the first aspect are implemented.

[0032] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method provided in the first aspect above.

[0033] The above-mentioned model training method, device, computer equipment and readable storage medium, the method determines the initial model corresponding to the target task in response to the selection operation of the target task; determines the target training sample set in response to the acquisition operation of the training sample corresponding to the target task; in response to the training operation of the initial model, the initial model is trained according to the target training sample set to obtain the target model. In this embodiment, the user can flexibly select the initial model and the target training sample set according to their own needs, so as to train the initial model according to the target training sample set to obtain the target model required by the user, and the target model obtained in this way can meet the needs of the user, so that the model training method provided by this embodiment has high practicality and flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a schematic diagram of the structure of a computer device in one embodiment;

[0035] Figure 2 A schematic diagram of a step flow of a model training method in an embodiment;

[0036] Figure 3 A schematic diagram of the steps of a model training method in another embodiment;

[0037] Figure 4 A schematic diagram of the steps of a model training method in another embodiment;

[0038] Figure 5 A schematic diagram of the steps of a model training method in another embodiment;

[0039] Figure 6 A schematic diagram of the steps of a model training method in another embodiment;

[0040] Figure 7 A schematic diagram of the steps of a model training method in another embodiment;

[0041] Figure 8 is a schematic diagram of a display interface of a computer device in one embodiment;

[0042] Fig. 9 Another is a schematic diagram of a display interface of a computer device in an embodiment;

[0043] Fig.10 Another is a schematic diagram of a display interface of a computer device in an embodiment;

[0044] Fig.11 Another is a schematic diagram of a display interface of a computer device in an embodiment;

[0045] Fig.12 A schematic diagram of an application framework of a model training method in an embodiment;

[0046] Fig.13 A schematic diagram of an application environment of a model training method in an embodiment;

[0047] Fig.14 Schematic diagram of the structure of a model training device in one embodiment. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0049] First of all, before specifically introducing the technical solutions of the embodiments of the present application, the technical background or technical evolution context on which the embodiments of the present application are based is introduced. With the development of deep learning, artificial neural networks are widely used in the medical field to process medical images, including the segmentation and classification of medical images. One of the important applications is intelligent quality control, that is, medical image quality control based on artificial neural networks. Intelligent quality control can automatically judge the quality problems in medical images through pre-trained network models, and promptly issue prompts and suggestions to medical staff, which can effectively reduce the quality problems of images and improve the efficiency of medical image quality inspection.

[0050] In traditional technology, after the server has trained the intelligent quality control model, it deploys the intelligent quality control model in the medical equipment, and the medical staff directly use the intelligent charge model. However, due to the different quality requirements of different hospitals or medical staff for images, it is difficult to unify the standards of the intelligent quality control model. The functions of the intelligent quality control model deployed in the medical equipment are relatively fixed and single, and cannot meet the needs of different medical staff. To address this problem, the present application provides a model training method.

[0051] The following is an introduction to the technical solutions involved in the embodiments of the present application in combination with the scenarios in which the embodiments of the present application are applied.

[0052] The model training method provided in the embodiment of the present application can be applied to a computer device, and its internal structure diagram is as follows: Figure 1 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a model training method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball or a touch pad set on the computer device housing, or an external keyboard, touch pad or mouse, etc.

[0053] In one embodiment, Figure 2 As shown, a model training method is provided. This embodiment uses the method applied to a computer device as an example. In this embodiment, the method includes the following steps:

[0054] Step 200: In response to a selection operation on a target task, an initial model corresponding to the target task is determined.

[0055] The target task refers to the function that the model that the user needs to train can achieve. The target task can include any of the various tasks such as motion artifact classification, distinguishing whether the image fat compression is uniform, target recognition, distinguishing whether the medical image contains a complete region of interest, image segmentation, regression analysis, etc.

[0056] The computer device includes a task display interface, and a variety of different tasks are displayed on the task display interface. The multiple different tasks displayed on the task display interface can be pre-set by the user and stored in the computer device. The user performs a selection operation on the target task from the multiple different tasks on the task display interface, and the computer device determines the initial model corresponding to the target task in response to the selection operation. Among them, the selection operation performed by the user in the task display interface can be a click operation or a sliding operation on the target task among the multiple different tasks. The initial model corresponding to the target task refers to the corresponding neural network when processing the target task. The initial model can be an untrained model, that is, a completely untrained neural network, or a trained model, that is, a trained neural network. This embodiment does not limit the type of target task and the type of initial model corresponding to the target task, as long as its function can be achieved.

[0057] Step 210: In response to the operation of acquiring the training samples corresponding to the target task, determine a target training sample set; wherein the target training sample set includes an original training sample set and a labeled training sample set.

[0058] The computer device includes a training sample display interface, and different training samples are displayed on the training sample display interface. The different training samples displayed on the training sample display interface can be pre-set by the user and stored in the computer device. The user performs an acquisition operation on the training sample corresponding to the target task among the multiple different training samples on the training sample display interface, and the computer device determines the target training sample set in response to the acquisition operation. Among them, the acquisition operation performed by the user in the training sample display interface can be a click operation or a drag operation on the training sample corresponding to the target task. The target training sample set includes an original training sample set and a labeled training sample set. The original training sample set refers to a sample set that is directly acquired and not processed; the labeled training sample set refers to a sample set that is labeled after the sample is acquired, and the annotations in the labeled training sample set are the accurate output results of the model trained using the training sample set. The labeled training sample set is the gold standard sample set used when training the model.

[0059] Step 220: In response to the training operation on the initial model, the initial model is trained according to the target training sample set to obtain a target model.

[0060] After obtaining the initial model and the target training sample set corresponding to the target task, the computer device trains the initial model according to the target training sample set in response to the user's training operation on the initial model, and obtains the target model after the target training end condition is met. The user's training operation on the initial model may be a click operation or a drag operation, etc. The target training end condition may be a condition corresponding to the initial model, or may be a condition obtained in response to the user's selection operation of multiple training end conditions, which is not limited in this embodiment.

[0061] When the computer device trains the initial model according to the target training sample set, the labeled training samples in the target training sample set can be divided into a training set and a verification set; the training set is used to train the initial model, and the verification set is used to verify the initial model.

[0062] In an optional embodiment, after determining the target training sample set, the computer device returns to the task display interface, and a training control corresponding to the initial model is set in the area where the initial model is located on the task display interface. The user's training operation on the initial model can be a click operation on the training control.

[0063] In an optional embodiment, the training sample display interface of the computer device may include a training operation control. After the computer device obtains the initial model and target training sample set corresponding to the target task, the user clicks on the training operation control on the training sample display interface. In response to the click operation, the computer device trains the initial model according to the target training sample set to obtain the target model.

[0064] The training sample display interface may also include multiple training end conditions. After the computer device determines the target training sample, or before determining the target training sample, the user selects multiple training end conditions on the training sample display interface to determine the target training end condition.

[0065] In another optional embodiment, the computer device responds to the acquisition operation of the training sample corresponding to the target task, and after determining the target training sample set, a training interface pops up. The training interface includes a training operation control, and the user clicks the training operation control on the training interface, and the computer device responds to the click operation to determine the target model.

[0066] The training interface also includes a training end condition control. Before the user clicks the training operation control on the training interface, the user can click the training end condition control on the training interface. In response to the click operation, the computer device displays the training end condition interface. The user selects multiple training end conditions on the training end condition interface to determine the target training end condition.

[0067] The model training method provided in this embodiment determines the initial model corresponding to the target task in response to the selection operation of the target task; determines the target training sample set in response to the acquisition operation of the training sample corresponding to the target task; and trains the initial model according to the target training sample set to obtain the target model in response to the training operation of the initial model. In this embodiment, the user can flexibly select the initial model and the target training sample set according to their own needs, so as to train the initial model according to the target training sample set to obtain the target model required by the user, and the target model obtained in this way can meet the needs of the user, so that the model training method provided in this embodiment has high practicality and flexibility.

[0068] In one embodiment, as described in the above embodiment, the initial model corresponding to the target task can be a trained model or an untrained model. Figure 3 As shown, it is related to an implementation method for determining a target training sample set in response to an acquisition operation of a training sample corresponding to a target task, and the steps of the method include:

[0069] Step 300: In response to an operation of acquiring training samples corresponding to a target task, an original training sample set is determined.

[0070] The user performs an acquisition operation on a training sample corresponding to a target task from a plurality of different training samples on the training sample display interface, and the computer device determines the original sample set corresponding to the target task in response to the acquisition operation. The acquisition operation performed by the user on the training sample display interface may be a click operation or a drag operation on the training sample corresponding to the target task. The original training sample set may be a sample set used when training the initial model, or may be another sample set stored in the computer device.

[0071] Step 310: In response to the use operation of the initial model, the original training sample set is input into the initial model to obtain an output result; and the labeled training sample set is determined according to the output result.

[0072] After the computer device obtains the original training sample set, the user performs a use operation on the initial model. In response to the use operation, the computer device inputs the obtained original training sample set into the initial model to obtain the output result of the initial model.

[0073] In an optional embodiment, a use control corresponding to the initial model is also provided on the interface where the initial model is located, and the user's use operation on the initial model may be a click operation on the user control.

[0074] After obtaining the output result of the initial model, the computer device determines the labeled training sample set according to the output result.

[0075] In this embodiment, when the initial model is a trained model, in response to the acquisition operation of the training samples corresponding to the target task, the original training sample set in the target training sample set can be determined; in response to the use operation of the initial model, the output result of the initial model is obtained, and the labeled training sample set in the target sample set can be determined according to the output result. In this way, the user can flexibly obtain the target training samples according to the selected initial model type, so that the model training method provided by this embodiment has high flexibility and practicality.

[0076] In one embodiment, Figure 4 As shown, an implementation method for determining a labeled training sample set according to an output result includes the following steps:

[0077] Step 400: In response to the confirmation operation on the output result, if it is determined that the output result meets the preset requirements, the output result is determined as the label of the labeled training sample set, and the labeled training sample set is determined according to the label of the labeled training sample set and the original training sample set.

[0078] After the computer device obtains the output result of the initial model based on the original training sample set, the output result will be displayed in the computer device, and the user performs a confirmation operation on the output result. In response to the confirmation operation, the computer device determines whether the output result meets the preset requirements. The preset requirements can be preset by the user, and different users can have the same or different preset requirements for the output results of the output model, which is not limited in this embodiment.

[0079] If the computer device determines that the output result meets the preset requirements, it means that the initial model can accurately perform the target task, that is, the output result of the initial model is accurate and can meet the needs of the user, then the output result is directly determined as the label in the labeled training sample set, and the label in the labeled training sample set is added to the corresponding original training sample set to obtain the labeled training sample set. Assuming that the initial model is a classification model, the output result meets the preset requirements, indicating that the initial model can classify the original training sample set and obtain a classification result that meets the user's requirements. Assuming that the initial model is a model for distinguishing different levels of motion artifacts, the output result meets the preset requirements, indicating that the initial model can classify the levels of motion artifacts in the original training sample set in accordance with the user's requirements.

[0080] Step 410: If it is determined that the output result does not meet the preset requirement, a labeled training sample is obtained in response to a labeling operation on the original training sample set.

[0081] If the computer device determines that the output result does not meet the preset requirements, it means that the initial model performs the target task and the output result obtained is inaccurate, then the user will annotate the original training sample set, that is, the user annotates the original training sample set on the display interface. The computer device responds to the user's annotation operation and obtains annotated training samples. Assuming that the initial model is a classification model, the output result does not meet the preset requirements, indicating that the classification result after the initial model classifies the original training sample does not meet the user's classification requirements. Assuming that the initial model is a model for distinguishing different levels of motion artifacts, the output result does not meet the preset requirements, indicating that the initial model does not meet the user's requirements for the classification of motion artifacts in the original training sample set.

[0082] In an optional embodiment, the output result of the initial model is displayed on the display interface of the computer, and there is a selection control corresponding to the output result on the display result, and the selection control includes a first control and a second control. The first control indicates that the output result meets the preset requirements, and the second control indicates that the output result does not meet the preset requirements. The user clicks on the first control on the display interface, and the computer device responds to the click operation, determines that the output result meets the preset requirements, and directly determines the output result as a labeled training sample; the user clicks on the second control on the display interface, and the computer device responds to the click operation and determines that the output result does not meet the preset requirements. At this time, the user will annotate the original training sample set on the display interface, and the computer device responds to the user's annotation operation and determines the original training sample set annotated by the user as a labeled training sample.

[0083] In this embodiment, considering that the initial model is a trained model, in response to the confirmation operation of the output result of the initial model, it is determined whether the output result meets the preset requirements to determine the labeled training sample set. Different judgment results lead to different methods for determining the labeled training sample set. In this way, when the initial model is a trained model, the labeled training sample set can be determined flexibly and quickly, thereby improving the practicality of the model training method.

[0084] In one embodiment, when the initial model corresponding to the target task is an untrained model, such as Figure 5 As shown, another implementation method of determining a target training sample set in response to an acquisition operation of training samples corresponding to a target task is involved, and the steps of the implementation method include:

[0085] Step 500: In response to an operation of acquiring training samples corresponding to a target task, an original training sample set is determined.

[0086] The user performs an acquisition operation on a training sample corresponding to a target task from a plurality of different training samples on the training sample display interface, and the computer device determines an original sample set corresponding to the target task in response to the acquisition operation. The acquisition operation performed by the user on the training sample display interface may be a click operation or a drag operation on the training sample corresponding to the target task.

[0087] Step 510: In response to the labeling operation on the original training sample set, a labeled training sample set is obtained.

[0088] After determining the original training sample set, the computer device will display the original training sample set on a display interface. The user performs a labeling operation on the original training sample set on the display interface, and the computer device obtains the labeled training samples in response to the labeling operation.

[0089] In this embodiment, considering that the initial model is an untrained model, in response to the acquisition operation of the training samples corresponding to the target task, the original training sample set in the target training sample set is determined; in response to the labeling operation of the original training sample set, the labeled training sample set in the target training sample set is obtained. In this way, when the initial model is an untrained model, the target training sample set can be accurately obtained, thereby improving the practicality of the model training method.

[0090] In one embodiment, in response to an operation of acquiring training samples corresponding to a target task, an implementation method of determining an original training sample set includes:

[0091] In response to a selection operation on stored training samples in a database, the selected stored training sample set is determined as an original training sample set; wherein the stored training samples include scan data acquired in different time periods.

[0092] The stored training samples in the database may be scanned data acquired in different time periods. In other words, the data scanned by the scanning device may be stored in the storage database in real time.

[0093] The computer device includes a database of training samples, in which a variety of different training samples are stored. The user selects the training samples in the database, and the computer device determines the stored training sample set selected by the user as the original training sample set in response to the selection operation.

[0094] In this embodiment, the determined original training sample set is directly selected from the stored training samples in the database, so the method for determining the original training sample set is fast and easy to implement.

[0095] In one embodiment, the method further comprises:

[0096] In response to an upload operation on the target model and / or the target training sample set, the target model and / or the target training sample set is uploaded to the cloud.

[0097] The cloud refers to a data sharing platform associated with computer equipment. If the model training method is applied to the medical field, the cloud can be a data sharing platform for multiple hospitals.

[0098] After the computer device obtains the target training sample set or the target model, an upload control is set for the target training sample set or the target model on the display interface of the computer device. The user clicks the upload control on the display interface, and the computer device responds to the click operation and uploads the target training sample set or the target model to the cloud. The user can click the upload controls corresponding to the target training sample set and the target model, and the computer device responds to the click operation and uploads the target model and the target training sample set to the cloud.

[0099] The display interface of the computer device includes a total upload control. After determining the target model, the user clicks the total upload control on the display interface. In response to the click, the computer device uploads the target training sample set and the target model to the cloud at the same time.

[0100] In this embodiment, uploading the target model and / or the target training sample set to the cloud can enable other computer devices to obtain the target model and / or the target training sample set, thereby improving the practicality of the model training method.

[0101] In one embodiment, the cloud stores training sample sets and models uploaded by other computer devices when the computer device has a cloud with which it is connected for communication, such as Figure 6 As shown, it is related to an implementation method of determining an initial model corresponding to a target task in response to a selection operation of the target task, and the implementation method includes:

[0102] Step 600: In response to a selection operation of a plurality of different types of models stored in the cloud, the selected model is determined as an initial model corresponding to the target task;

[0103] The user performs a selection operation on a plurality of different types of models stored in the cloud displayed on the display interface of the computer device, and the computer device determines the model selected by the user as the initial model corresponding to the target task in response to the selection operation.

[0104] When a training sample set and a model uploaded by other computer devices are stored in the cloud and the computer device has a cloud with which it is communicated, an implementation method of determining a target training sample set in response to an acquisition operation of a training sample corresponding to a target task includes:

[0105] Step 610: In response to the selection operation of the training samples stored in the cloud, the selected training samples are determined as the target training sample set corresponding to the target task.

[0106] The user performs a selection operation on the training samples stored in the cloud displayed on the display interface of the computer device, and the computer device determines the selected training samples as the target training sample set that the target considers to be corresponding in response to the selection operation.

[0107] In an optional embodiment, the computer device includes a cloud login interface, the user enters login information on the cloud login interface, and the computer device enters the cloud in response to the login information. The initial interface of the cloud is displayed on the display interface of the computer device, and the initial interface of the cloud includes a model control and a sample control. The user clicks on the model control, and the computer device displays multiple different types of models stored in the cloud in response to the click operation. The user clicks on the sample control, and the computer device displays the training sample stored in the cloud in response to the click operation.

[0108] In this embodiment, more types of models and training samples are stored in the cloud. The computer device selects the model corresponding to the target task from the different types of models stored in the cloud, and selects the target training sample set from the training samples stored in the cloud, which can improve the flexibility and practicality of the model training method.

[0109] See also Figure 7 An embodiment of the present application provides a model training method, the steps of which include:

[0110] Step 700: In response to a selection operation on a target task, determining an initial model corresponding to the target task;

[0111] Step 710: In response to the selection operation of the stored training samples in the database, the selected stored training sample set is determined as the original training sample set; wherein the stored training samples include scanning data acquired in different time periods;

[0112] Step 720: If the initial model is a trained model, in response to the use operation of the initial model, the original training sample set is input into the initial model to obtain an output result;

[0113] Step 730: In response to the confirmation operation on the output result, if it is determined that the output result meets the preset requirement, the output result is determined as the label of the labeled training sample set, and the labeled training sample set is determined according to the label of the labeled sequence sample set and the original training sample set; if it is determined that the output result does not meet the preset requirement, in response to the labeling operation on the original training sample set, a labeled training sample set is obtained;

[0114] Step 740: If the initial model is an untrained model, in response to the labeling operation on the original training sample set, a labeled training sample set is obtained;

[0115] Step 750: In response to the training operation on the initial model, the initial model is trained according to the labeled training sample set to obtain a target model.

[0116] In one embodiment, the display interface of the computer device is as follows: Figure 8 As shown, in Figure 8 The function control display area includes a function control display area and a display area. In the function control display area, multiple function controls are displayed, such as: model structure control, training sample control, model training control and model deployment control. Different function controls correspond to different display interfaces. In other words, when a user clicks on different function controls in the function control display area, the interface corresponding to the function control can be displayed in the display area. Figure 8 The display area shown is the interface corresponding to the model control, namely the task display interface. The task display interface includes a variety of different tasks: classification, segmentation, target location, regression analysis and customization. The computer device receives the user's selection operation of the target task among the various tasks and determines the initial model.

[0117] In an alternative embodiment, if Figure 8 As shown, if the target task selected by the user on the task display interface is classification, the computer device will display the parameters corresponding to the initial model corresponding to the classification task in response to the selection operation of the classification task. The parameters may include multiple output formats, output category names, network structures, input formats, input dimensions, and confirmation controls of the initial model. If the computer device receives the user's selection operation of the three classification formats in the output format, the output category name includes the names of the three classification results. Assuming that the classification task is to distinguish motion artifacts, the output category name may include 0 for no motion artifacts, 1 for slight motion artifacts, and 2 for severe motion artifacts. The network structure of the initial model may include ResNet (residual neural network) and VGG (deep convolutional network structure), etc. The input format of the initial model may include a single image, multiple images, images and scanning parameters, etc. The input dimensions of the initial model may include the original size of the image, the image after reduction or cropping, and customization. The confirmation control of the initial model may include a completion control, a save control, and a reset control. The computer device obtains the initial model in response to the user clicking the complete control; the computer device saves the initial model to the computer device in response to the user clicking the save control; the computer device prompts the user to rewrite and set various parameters of the initial model in response to the user clicking the reset control.

[0118] like Fig. 9The interface corresponding to the sample control in the display area shown is the training sample display interface. The training sample display interface displays a newly created database created by the user, which is used to store the acquired target training sample set. The display interface includes the name and location of the newly created database. The user's selection operation is received, and the original training sample set is determined from the stored training samples of different time periods in the local database, or the original training sample set is obtained in the cloud. When the initial model is a trained model, the training sample display interface includes interactive annotation of existing data, that is, the original training sample is input into the initial model to obtain the output result. When the output result does not meet the preset requirements, a selection control is displayed on the training sample display interface, and the user clicks on the selection control, and the computer device selects whether to annotate online in response to the click operation; if the selection result is online annotation, the annotated training sample set is obtained in response to the user's annotation operation on the original training sample set. The training sample display interface also includes a target training sample set confirmation control, which may include a completion control, a save control, and a reset control. The computer device obtains the target training sample set in response to the user clicking the complete control; the computer device saves the target training sample set to a newly created database in response to the user clicking the save control; the computer device prompts the user to rewrite various parameters of the target training sample in response to the user clicking the reset control.

[0119] like Fig.10 The display area shown is an interface corresponding to the model training control, which includes multiple training parameters and multiple training end conditions. The computer device determines the target training parameters and the target training end conditions in response to the user's selection operation on the training parameters and the training end conditions, and trains the initial model based on the target training sample set, the target training parameters and the target training end conditions. The interface also includes a target model confirmation control, which may include a completion control, a save control, and a reset control. The computer device obtains the target model in response to the user's click operation on the completion control; the computer device saves the target model to the computer device in response to the user's click operation on the save control; the computer device prompts the user to rewrite the settings of the various parameters of the target model obtained by training in response to the user's click operation on the reset control.

[0120] like Fig.11The display area shown is the interface corresponding to the model deployment control, which includes multiple operation modes and multiple result display methods corresponding to the target model. When the target model is a classification model, the multiple operation modes corresponding to the target model include automatically classifying all new scanned data, checking whether to run the model in the protocol interface, and selecting to run the target model in the interface. The multiple result display methods corresponding to the target model include automatically popping up the classification results after the scan, marking the classification results after the image list, and marking the classification results in the scan data. The computer device determines the target operation mode and the target result display method in response to the user's selection operation of the operation mode and the result display method. The interface also includes a model deployment confirmation control, which may include a completion control, a save control, and a reset control. The computer device obtains the deployment of the target model in response to the user's click operation on the completion control; the computer device saves the deployment of the target model to the computer device in response to the user's click operation on the save control; the computer device prompts the user to rewrite and set various deployment parameters of the target model in response to the user's click operation on the reset control.

[0121] The model training method framework provided in this embodiment is as follows Fig.12 As shown, the user obtains the current model by customizing the task and model type. The training sample set required for training the current model can be obtained from the imaging device. The current model can be trained using the training sample set to obtain the trained current model required by the user. For the trained current model, intelligent quality control can be used to determine whether the output result of the model meets the preset conditions; if the preset conditions are not met, the user performs an annotation operation on the corresponding training sample set to obtain an annotated training sample set; if the preset conditions are met, the output result of the model is used as the label of the corresponding training sample set, and the annotated training sample set is obtained based on the label of the training sample set and the training sample set. The annotated training sample set is collected and stored in the training sample database. The user can select the required training samples from the training sample database for model training, and use the trained model to update the current model.

[0122] If the model training method provided in this embodiment is applied to three hospitals, namely, Hospital 1, Hospital 2, and Hospital 3, the structures of the three hospitals and the cloud are as follows: Fig.13As shown. Hospital 1 includes local data (training samples and target training sample sets stored in the computer equipment of the local hospital) and a local model (i.e., the target model obtained by training the computer equipment of the local hospital), Hospital 2 includes a local model, Hospital 3 includes local data, an initial model, and a local model, and the cloud includes a cloud server, a public model, a shared model library, and a shared database. Hospital 1 can upload local data and local models to the cloud, and store them in a shared database and a shared model library. The local model in Hospital 2 can be obtained from the shared model library in the cloud. The initial model in Hospital 3 can be obtained from the shared model library.

[0123] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0124] Based on the same inventive concept, the embodiment of the present application also provides a model training device for implementing the model training method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more model training device embodiments provided below can refer to the limitations on the model training method above, and will not be repeated here.

[0125] In one embodiment, Fig.14 As shown, a model training device 10 is provided, comprising: a selection module 11, an acquisition module 12 and a training module 13, wherein:

[0126] The selection module 11 is used to determine an initial model corresponding to the target task in response to a selection operation on the target task.

[0127] The acquisition module 12 is used to determine a target training sample set in response to an acquisition operation of training samples corresponding to the target task; wherein the target training sample set includes an original training sample set and a labeled training sample set.

[0128] The training module 13 is used to, in response to the training operation on the initial model, train the initial model according to the target training sample set to obtain the target model.

[0129] In one embodiment, the acquisition module 12 includes an acquisition unit and a determination unit. The acquisition unit is used to determine the original training sample set in response to the acquisition operation of the training sample corresponding to the target task; the determination unit is used to input the original training sample set into the initial model in response to the use operation of the initial model to obtain the output result; and determine the labeled training sample set according to the output result.

[0130] In one embodiment, the determination unit is specifically used to respond to a confirmation operation on the output result. If it is determined that the output result meets the preset requirements, the output result is determined as the label of the labeled training sample set, and the labeled training sample set is determined according to the label of the labeled training sample set and the original training sample set; if it is determined that the output result does not meet the preset requirements, the labeled training sample set is obtained in response to the labeling operation on the original training sample set.

[0131] In one embodiment, the acquisition unit is used to determine the original training sample set in response to an acquisition operation on the training samples corresponding to the target task; the determination unit is also used to obtain the labeled training sample set in response to a labeling operation on the original training sample set.

[0132] In one embodiment, the acquisition unit is specifically configured to determine the selected stored training sample set as the original training sample set in response to a selection operation on stored training samples in the database; wherein the stored training samples include scan data acquired in different time periods.

[0133] In one embodiment, the model training device 10 further includes an uploading module, and the uploading module is used to upload the target model and / or the target training sample set to the cloud in response to an upload operation on the target model and / or the target training sample set.

[0134] In one embodiment, the selection module 11 is specifically used to respond to the selection operation of multiple different types of models stored in the cloud, and determine the selected model as the initial model corresponding to the target task; the acquisition module 12 is specifically used to respond to the selection operation of the training samples stored in the cloud, and determine the selected training samples as the target training sample set corresponding to the target task.

[0135] Each module in the above-mentioned model training device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0136] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 1 shown.

[0137] Those skilled in the art will understand that Figure 1 The 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.

[0138] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0139] In response to a selection operation of a target task, determining an initial model corresponding to the target task;

[0140] In response to an operation of acquiring training samples corresponding to a target task, a target training sample set is determined; wherein the target training sample set includes an original training sample set and a labeled training sample set;

[0141] In response to the training operation on the initial model, the initial model is trained according to the target training sample set to obtain a target model.

[0142] In one embodiment, when the processor executes the computer program, the following steps are also implemented: in response to the operation of obtaining the training samples corresponding to the target task, the original training sample set is determined; in response to the operation of using the initial model, the original training sample set is input into the initial model to obtain the output result; and the labeled training sample set is determined according to the output result.

[0143] In one embodiment, when the processor executes the computer program, the following steps are also implemented: in response to a confirmation operation on the output result, if it is determined that the output result meets the preset requirements, the output result is determined as the label of the labeled training sample set, and the labeled training sample is determined according to the label of the labeled training sample set and the original training sample set; if it is determined that the output result does not meet the preset requirements, in response to the labeling operation on the original training sample set, a labeled training sample set is obtained.

[0144] In one embodiment, when the processor executes the computer program, the following steps are further implemented: in response to an operation of acquiring training samples corresponding to the target task, the original training sample set is determined; in response to an operation of labeling the original training sample set, a labeled training sample set is obtained.

[0145] In one embodiment, when the processor executes the computer program, the following steps are also implemented: in response to a selection operation on stored training samples in the database, the selected stored training sample set is determined as the original training sample set; wherein the stored training samples include scanning data acquired in different time periods.

[0146] In one embodiment, when the processor executes the computer program, the following steps are further implemented: in response to an upload operation on the target model and / or the target training sample set, the target model and / or the target training sample set is uploaded to the cloud.

[0147] In one embodiment, when the processor executes the computer program, the following steps are also implemented: in response to a selection operation on a plurality of different types of models stored in the cloud, the selected model is determined as an initial model corresponding to a target task; in response to a selection operation on training samples stored in the cloud, the selected training samples are determined as a target training sample set corresponding to the target task.

[0148] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0149] In response to a selection operation of a target task, determining an initial model corresponding to the target task;

[0150] In response to an operation of acquiring training samples corresponding to a target task, a target training sample set is determined; wherein the target training sample set includes an original training sample set and a labeled training sample set;

[0151] In response to the training operation on the initial model, the initial model is trained according to the target training sample set to obtain a target model.

[0152] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in response to the acquisition operation of the training samples corresponding to the target task, the original training sample set is determined; in response to the use operation of the initial model, the original training sample set is input into the initial model to obtain the output result; and the labeled training sample set is determined according to the output result.

[0153] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in response to a confirmation operation on the output result, if it is determined that the output result meets the preset requirements, the output result is determined as the label of the labeled training sample set, and the labeled training sample is determined according to the label of the labeled training sample set and the original training sample set; if it is determined that the output result does not meet the preset requirements, in response to the labeling operation on the original training sample set, a labeled training sample set is obtained.

[0154] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in response to an operation of acquiring training samples corresponding to the target task, the original training sample set is determined; in response to an operation of labeling the original training sample set, a labeled training sample set is obtained.

[0155] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in response to a selection operation on stored training samples in the database, the selected stored training sample set is determined as the original training sample set; wherein the stored training samples include scanning data acquired in different time periods.

[0156] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: in response to an upload operation on the target model and / or the target training sample set, the target model and / or the target training sample set is uploaded to the cloud.

[0157] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in response to a selection operation on a plurality of different types of models stored in the cloud, the selected model is determined as an initial model corresponding to a target task; in response to a selection operation on training samples stored in the cloud, the selected training samples are determined as a target training sample set corresponding to the target task.

[0158] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0159] In response to a selection operation of a target task, determining an initial model corresponding to the target task;

[0160] In response to an operation of acquiring training samples corresponding to a target task, a target training sample set is determined; wherein the target training sample set includes an original training sample set and a labeled training sample set;

[0161] In response to the training operation on the initial model, the initial model is trained according to the target training sample set to obtain a target model.

[0162] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in response to the acquisition operation of the training samples corresponding to the target task, the original training sample set is determined; in response to the use operation of the initial model, the original training sample set is input into the initial model to obtain the output result; and the labeled training sample set is determined according to the output result.

[0163] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in response to a confirmation operation on the output result, if it is determined that the output result meets the preset requirements, the output result is determined as a labeled training sample set; if it is determined that the output result does not meet the preset requirements, in response to a labeling operation on the original training sample set, a labeled training sample set is obtained.

[0164] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in response to an operation of acquiring training samples corresponding to the target task, the original training sample set is determined; in response to an operation of labeling the original training sample set, a labeled training sample set is obtained.

[0165] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in response to a selection operation on stored training samples in the database, the selected stored training sample set is determined as the original training sample set; wherein the stored training samples include scanning data acquired in different time periods.

[0166] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: in response to an upload operation on the target model and / or the target training sample set, the target model and / or the target training sample set is uploaded to the cloud.

[0167] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in response to a selection operation on a plurality of different types of models stored in the cloud, the selected model is determined as an initial model corresponding to a target task; in response to a selection operation on training samples stored in the cloud, the selected training samples are determined as a target training sample set corresponding to the target task.

[0168] 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 the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile 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), magnetoresistive 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. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0169] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, 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 specification.

[0170] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A model training method, characterized in that: The method comprises: In response to a selection operation on a target task, determining an initial model corresponding to the target task; In response to the acquisition operation of the training samples corresponding to the target task, a target training sample set is determined; wherein the target training sample set includes an original training sample set and a labeled training sample set; In response to the training operation on the initial model, the initial model is trained according to the target training sample set to obtain a target model.

2. The method according to claim 1, characterized in that If the initial model is a trained model, determining a target training sample set in response to an operation of acquiring training samples corresponding to the target task includes: In response to an operation of acquiring training samples corresponding to the target task, determining the original training sample set; In response to the use operation of the initial model, the original training sample set is input into the initial model to obtain an output result; and the labeled training sample set is determined according to the output result.

3. The method according to claim 2, characterized in that Determining the labeled training sample set according to the output result includes: In response to a confirmation operation on the output result, if it is determined that the output result meets a preset requirement, the output result is determined as a label of the labeled training sample set, and the labeled training sample set is determined according to the label of the labeled training sample set and the original training sample set; If it is determined that the output result does not meet the preset requirement, the labeled training sample set is obtained in response to a labeling operation on the original training sample set.

4. The method according to claim 1, characterized in that: If the initial model is an untrained model, determining a target training sample set in response to an operation of acquiring training samples corresponding to the target task includes: In response to an operation of acquiring training samples corresponding to the target task, determining the original training sample set; In response to the labeling operation on the original training sample set, the labeled training sample set is obtained.

5. The method according to claim 2 or 4, characterized in that: The step of determining the original training sample set in response to the operation of acquiring the training sample corresponding to the target task includes: In response to a selection operation on stored training samples in a database, the selected stored training sample set is determined as the original training sample set; wherein the stored training samples include scan data acquired in different time periods.

6. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: In response to an upload operation on the target model and / or the target training sample set, the target model and / or the target training sample set is uploaded to the cloud.

7. The method according to any one of claims 1 to 4, characterized in that: In response to the selection operation of the target task, determining the initial model corresponding to the target task includes: In response to a selection operation of a plurality of different types of models stored in the cloud, determining the selected model as an initial model corresponding to the target task; The step of determining a target training sample set in response to an operation of acquiring the training sample corresponding to the target task includes: In response to a selection operation on a training sample stored in the cloud, the selected training sample is determined as a target training sample set corresponding to the target task.

8. A model training device, characterized in that: The device comprises: A selection module, configured to determine an initial model corresponding to the target task in response to a selection operation on the target task; An acquisition module, configured to determine a target training sample set in response to an acquisition operation of the training sample corresponding to the target task; wherein the target training sample set includes an original training sample set and a labeled training sample set; A training operation is used to, in response to a training operation on the initial model, train the initial model according to the target training sample set to obtain a target model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

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 method according to any one of claims 1 to 7 are implemented.

Images