A lesion recognition model training method and device, equipment and storage medium

Abstract

The application discloses a lesion recognition model training method and device, equipment and a storage medium. The method comprises the following steps: obtaining model training information for guiding model training. According to the model training information, training sample data for model training is obtained. According to a preset lesion recognition model and the training sample data, a training task corresponding to the sample label is executed. When the number of training rounds of the preset lesion recognition model meets a preset round threshold, it is determined that the training of the preset lesion recognition model is completed, and a target lesion recognition model corresponding to the model training information is obtained. The technical scheme can automatically execute the corresponding training task according to the training sample label, enhance the personalized model training of the complex analysis scene, and promote the customized training and development and application of the medical model through the code-free configuration mode.

Description

Technical Field

[0001] This invention relates to the field of medical recognition model training technology, and in particular to a training method, apparatus, device and storage medium for a lesion recognition model. Background Technology

[0002] Artificial intelligence-based medical image analysis technology is rapidly developing and is widely used in the auxiliary identification and quantitative analysis of lesions in images. Efficient and accurate models are at its core, and the performance of these models highly depends on high-quality training and the support of scientific processes.

[0003] However, existing model training data preparation processes have significant drawbacks. There is a lack of unified closed-loop management of data labeling, review, and quality control processes, resulting in inconsistent training data quality. The methods for determining training task types are fixed, failing to enable intelligent training that automatically determines training tasks based on user-defined label combinations. The construction of negative datasets is typically isolated from the main process, making it difficult to conveniently incorporate as a configurable standard option.

[0004] These drawbacks collectively limit the in-depth development and reliable application of AI-based medical image analysis technology. Low-quality or unverified training data makes it difficult to guarantee the accuracy and clinical reliability of model outputs. Fixed task adaptation mechanisms hinder personalized training and model development for complex clinical scenarios, increasing the cost of trial and error in research and engineering. These problems make model training processes cumbersome, trial and error costs high, and further raise the technical threshold. Summary of the Invention

[0005] This invention provides a training method, apparatus, device, and storage medium for a lesion recognition model, which enables the automatic execution of corresponding training tasks based on training sample labels, enhances personalized model training for complex analysis scenarios, and effectively promotes the customized training, development, and application of medical models through a no-code configuration method.

[0006] According to one aspect of the present invention, a method for training a lesion recognition model is provided. The method includes: Obtain model training information for guiding model training, wherein the model training information includes at least the model recognition type; Based on the model training information, training sample data for model training is obtained, wherein the sample labels of the training sample data correspond to the model recognition type, and the training sample data includes lesion label results used to determine the model training accuracy; Based on a predetermined lesion recognition model and the training sample data, a training task corresponding to the sample label is executed, wherein the predetermined lesion recognition model is a deep neural network model that has not completed model training; When the number of training rounds of the preset lesion recognition model meets the preset round threshold, the training of the preset lesion recognition model is determined to be completed, and the target lesion recognition model corresponding to the model training information is obtained.

[0007] According to another aspect of the present invention, a training apparatus for a lesion recognition model is provided. The apparatus includes: The training information acquisition module is used to acquire model training information for guiding model training, wherein the model training information includes at least the model recognition type; The training sample acquisition module is used to acquire training sample data for model training based on the model training information, wherein the sample labels of the training sample data correspond to the model recognition types, and the training sample data includes lesion label results used to determine the model training accuracy. The training task execution module is used to execute a training task corresponding to the sample label according to a pre-determined preset lesion recognition model and the training sample data, wherein the preset lesion recognition model is a deep neural network model that has not completed model training; A training task completion module is used to determine that the training of the preset lesion recognition model has ended when the number of training rounds of the preset lesion recognition model meets a preset round threshold, and to obtain the target lesion recognition model corresponding to the model training information. According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the training method of the lesion recognition model according to any embodiment of the present invention.

[0008] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the training method of the lesion recognition model according to any embodiment of the present invention.

[0009] The technical solution of this invention involves acquiring model training information for guiding model training, wherein the model training information includes at least a model recognition type; acquiring training sample data for model training based on the model training information, wherein the sample labels of the training sample data correspond to the model recognition type, and the training sample data includes lesion label results for determining the model training accuracy; executing a training task corresponding to the sample label based on a pre-determined preset lesion recognition model and the training sample data, wherein the preset lesion recognition model is a deep neural network model that has not completed model training; determining that the training of the preset lesion recognition model has ended when the number of training rounds of the preset lesion recognition model meets a preset round number threshold, and obtaining the target lesion recognition model corresponding to the model training information. This achieves automatic execution of training tasks corresponding to sample labels based on the sample labels of the training sample data, reducing the complexity of model training and the professionalism of trainers, enhancing personalized model training for complex analysis scenarios, and promoting the customized training development and application of medical models.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of a training method for a lesion recognition model according to an embodiment of the present invention; Figure 2 This is a flowchart of a training method for a lesion recognition model according to an embodiment of the present invention; Figure 3 This is a structural diagram of a training device for a lesion recognition model according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device that implements the training method of the lesion recognition model in the embodiments of the present invention. Detailed Implementation

[0013] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0014] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] Figure 1 This is a flowchart illustrating a training method for a lesion recognition model provided in an embodiment of the present invention. This embodiment is applicable to situations requiring customized training of medical models. The method can be executed by a training device for the lesion recognition model, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method includes: S101. Obtain model training information for guiding model training. The model training information includes at least the model recognition type.

[0016] It should be noted that this invention is applied to a model training system, which provides configurable model training functions. The model training system is a no-code training platform. Training users (e.g., model users include doctors, patients, and model training professionals) can configure the data for a single model training session through a configuration interface provided by the system, by checking boxes and inputting parameters, so that the model training system can automatically execute the corresponding model training task.

[0017] Among them, model training information can refer to the training needs of model users, such as model recognition accuracy, model training cycle, and model recognition type.

[0018] Specifically, based on the training needs of model users, model training information is constructed to guide model training, and this model training information is loaded into the model training system so that the model training system can obtain the model training information.

[0019] S102. Based on the model training information, obtain training sample data for model training.

[0020] It should be noted that the training sample data in this invention is pre-defined sample data, and different sample labels and lesion labels are configured for the sample data during the sample data definition process. The sample labels are used to indicate different model training tasks, and the lesion labels are used to indicate the lesion results present in the training sample data. All types of training sample data are stored in a storage device, and the storage device has a communication connection with the model training system, allowing the model training system to access the training sample data in the storage device for model training.

[0021] It is worth noting that the sample labels of the acquired training sample data correspond to the model recognition types, and the training sample data includes lesion label results used to determine the model training accuracy. For example, the model recognition types include classification recognition types, segmentation recognition types, and detection recognition types; the sample labels include classification labels, segmentation labels, and detection labels, etc.

[0022] Specifically, based on the model training information, the model training system retrieves training sample data that meets the model training information from the storage device, so that the model training system can train the model based on the training sample data.

[0023] S103. Based on the pre-determined preset lesion recognition model and the training sample data, execute the training task corresponding to the sample label.

[0024] The preset lesion recognition model is a deep neural network model that has not yet completed model training, such as a CNN model or a Transformer model. The preset lesion recognition model can be constructed in advance in the model training system.

[0025] Specifically, based on the model training system, a pre-defined lesion recognition model is trained using predetermined training sample data. It should be noted that the model training system can automatically execute training tasks corresponding to the sample labels in the training sample data, thereby effectively reducing the complexity of model training and the professionalism required of the trainers. This enables personalized model training for complex analysis scenarios based on user needs.

[0026] For example, the training sample data may contain terms like "malignant pulmonary nodule" or "benign pulmonary nodule." The sample labels for the training sample data are classification labels. The model training system performs a classification training task based on the training sample data, enabling the pre-defined lesion identification model to identify lesions in the training sample data and output whether a malignant nodule exists or the probability of its malignancy.

[0027] For example, the training sample data contains a complete outline of each nodule that has been accurately depicted, and the sample labels of this training sample data are segmentation labels. The model training system performs a segmentation training task based on the training sample data, so that the preset lesion recognition model can identify lesions in the training sample data and output the location bounding boxes of each nodule in the training sample data.

[0028] For example, the training sample data has already outlined the approximate location and range of nodules, and the sample label of this training sample data is the detection label. The model training system performs a detection training task based on the training sample data, so that the preset lesion recognition model can identify lesions in the training sample data, output the nodules in the training sample data, and mark them with rectangular boxes.

[0029] For example, the step of performing a training task corresponding to the model's recognition type based on a pre-determined preset lesion recognition model and the training sample data includes: inputting the training sample data into the preset lesion recognition model to perform a recognition decision corresponding to the model's recognition type, and obtaining a lesion recognition result based on the output of the preset lesion recognition model; determining a training error based on the lesion recognition result and the lesion label result, and backpropagating the training error to the preset lesion recognition model to adjust the network parameters in the preset lesion recognition model. Here, the lesion recognition result can refer to the recognition result output by the preset lesion recognition model during the training process.

[0030] Specifically, in each training round, training sample data is sequentially input into a preset lesion recognition model. This allows the model to make recognition decisions corresponding to the model's recognition type based on the sample labels, and output lesion recognition results based on these decisions. The lesion recognition results and the lesion label results are compared and analyzed based on a pre-determined loss function to determine the training error of the preset lesion recognition model. This training error is then backpropagated to the preset lesion recognition model to adjust its network parameters, thus completing a single training round.

[0031] S104. When the number of training rounds of the preset lesion recognition model meets the preset round threshold, the training of the preset lesion recognition model is determined to be completed, and the target lesion recognition model corresponding to the model training information is obtained.

[0032] The preset round number threshold can be obtained through training according to actual needs, and this invention does not impose specific limitations on it.

[0033] Specifically, one training epoch is defined as the process of training all training samples once by the preset lesion recognition model. When the number of training epochs for the preset lesion recognition model reaches a preset epoch threshold, the training of the preset lesion recognition model can be considered complete. The preset lesion recognition model that has completed training is then designated as the target lesion recognition model corresponding to the model training information.

[0034] Optionally, after obtaining the target lesion recognition model, the method further includes: acquiring target image data of the lesion to be identified; inputting the target image data into the target lesion recognition model for lesion recognition; and obtaining the lesion recognition result based on the output of the target lesion recognition model.

[0035] Specifically, after obtaining the target lesion recognition model, the target image data to be identified is input into the target lesion recognition model to perform lesion recognition, thereby obtaining the lesion recognition result. Training the target lesion recognition model based on training sample data can effectively improve the accuracy and efficiency of lesion recognition results, further enhancing the practical application of medical models.

[0036] The technical solution of this invention involves acquiring model training information for guiding model training, wherein the model training information includes at least a model recognition type; acquiring training sample data for model training based on the model training information, wherein the sample labels of the training sample data correspond to the model recognition type, and the training sample data includes lesion label results for determining the model training accuracy; executing a training task corresponding to the sample label based on a pre-determined preset lesion recognition model and the training sample data, wherein the preset lesion recognition model is a deep neural network model that has not completed model training; determining that the training of the preset lesion recognition model has ended when the number of training rounds of the preset lesion recognition model meets a preset round number threshold, and obtaining the target lesion recognition model corresponding to the model training information. This achieves automatic execution of training tasks corresponding to sample labels based on the sample labels of the training sample data, reducing the complexity of model training and the professionalism of trainers, enhancing personalized model training for complex analysis scenarios, and promoting the customized training development and application of medical models.

[0037] Figure 2 This is a flowchart illustrating a training method for a lesion recognition model provided in an embodiment of the present invention. This embodiment further refines the acquisition of training sample data for model training based on the aforementioned embodiments. For example... Figure 2 As shown, the method includes: S201. Obtain model training information to guide model training.

[0038] S202. Determine the sample label corresponding to the model recognition type based on the model recognition type in the model training information.

[0039] Specifically, by conducting a requirements analysis on the model training information, the model's lesion recognition type is determined. Based on the model recognition type, the sample labels corresponding to the required training sample data are determined.

[0040] For example, the sample labels of the training sample data correspond to the model recognition type, including: when the model recognition type is a classification recognition type, the sample label is a classification label; when the model recognition type is a segmentation recognition type, the sample label is a segmentation label; when the model recognition type is a detection recognition type, the sample label is a segmentation label, or a segmentation label and a detection label.

[0041] In other words, sample labels are a crucial parameter for model training. Data with classification labels only supports training tasks for classification and recognition; data with segmentation labels supports training tasks for segmentation and recognition or detection; and data with detection labels only supports training tasks for detection and recognition. Furthermore, data with both segmentation and detection labels only supports training tasks for detection and recognition.

[0042] S203. Select multimodal lesion images corresponding to the sample labels from the pre-constructed quality control data pool.

[0043] The data source for the quality control data pool is data that has completed the labeling-review-quality control process and has passed the quality control by the quality control personnel.

[0044] Users can access the quality control data pool through the model training system and select multimodal lesion images for model training from the pool based on sample labels. These multimodal lesion images contain sample data with lesion labeling already completed.

[0045] In this invention, the pre-quality-controlled training data may include single-modal lesion images or multimodal lesion images. Multimodal images may include data from different modalities such as CT and MRI.

[0046] For example, the construction process of the quality control data pool includes: A variety of desensitized medical images of different modalities are acquired for medical research, and desensitized lesion images with lesion structures are selected from the desensitized medical images; the desensitized lesion images are subjected to lesion annotation processing to obtain lesion image data, wherein the lesion annotation processing includes at least lesion location annotation processing and lesion label annotation processing; the lesion image data is stored in a database to construct the quality control data pool.

[0047] Specifically, the construction of the quality control data pool requires standardized medical image annotation and data entry. Anonymized and desensitized medical images (such as CT and MRI sequences) are obtained from hospital PACS systems, imaging equipment, and other sources. Image data containing lesion structures are selected from these desensitized medical images as desensitized lesion images. Lesion annotation is performed on these desensitized lesion images, followed by multi-level sampling or full-scale review to ensure the accuracy of the lesion image data. High-quality lesion image data that has completed the review is then pushed to the quality control data pool for further processing.

[0048] It is worth noting that lesion annotation processing includes at least lesion location annotation processing and lesion label annotation processing. In the quality control data pool, each lesion image data is accompanied by a sample label, so that intelligent filtering and combination can be performed according to label type during the training sample data screening stage.

[0049] S204. Based on the multimodal lesion images, construct training sample data for model training.

[0050] Specifically, multimodal lesion images are directly identified as training sample data for model training.

[0051] For example, constructing training sample data for model training based on the multimodal lesion images includes: acquiring multimodal normal images corresponding to the model recognition type, and determining the multimodal normal images as a negative dataset that does not contain lesion information; determining the multimodal lesion images as a positive dataset that contains lesion information; and constructing training sample data for model training based on the positive dataset and the negative dataset.

[0052] In this invention, a positive dataset can refer to lesion training data containing lesion information. A negative dataset can refer to normal training data that does not contain lesion information. To improve the accuracy of the training model in identifying lesions, a certain number of normal samples are included as a negative dataset, which is beneficial for model training. A separate negative sample group can be constructed at the research sample site to store the negative dataset.

[0053] Specifically, multimodal lesion images are identified as the positive dataset containing lesion information. Multimodal normal images corresponding to the sample labels are obtained from the negative sample group and used as the negative dataset. The positive and negative datasets are then merged according to a pre-defined ratio to construct training sample data for model training.

[0054] S205. Based on the pre-determined preset lesion recognition model and the training sample data, execute the training task corresponding to the sample label.

[0055] S206. When the number of training rounds of the preset lesion recognition model meets the preset round threshold, the training of the preset lesion recognition model is determined to be completed, and the target lesion recognition model corresponding to the model training information is obtained.

[0056] The technical solution of this invention, by selecting multimodal lesion images corresponding to the sample labels from a pre-constructed quality control data pool, strictly limits the training data to originating from a quality control data pool that has undergone multi-level review, ensuring the high quality of the positive dataset from the source and directly improving the reliability of the model. Secondly, it automatically associates and selects multimodal images with corresponding labels based on the selected recognition type, achieving precise matching between data preparation and task objectives. Most importantly, it transforms the construction of the negative dataset into a standard option that can be easily configured within the workflow, greatly enhancing the convenience and repeatability of robust model training, significantly improving configuration efficiency and reducing the risk of operational errors, and increasing the practicality of customized model training.

[0057] Figure 3 This is a schematic diagram of the structure of a training device for a lesion recognition model provided in an embodiment of the present invention. Figure 3 As shown, the device includes: The training information acquisition module 301 is used to acquire model training information for guiding model training, wherein the model training information includes at least the model recognition type; The training sample acquisition module 302 is used to acquire training sample data for model training based on the model training information, wherein the sample labels of the training sample data correspond to the model recognition type, and the training sample data includes lesion label results used to determine the model training accuracy. The training task execution module 303 is used to execute a training task corresponding to the sample label according to a pre-determined preset lesion recognition model and the training sample data, wherein the preset lesion recognition model is a deep neural network model that has not completed model training. The training task completion module 304 is used to determine that the training of the preset lesion recognition model has ended when the number of training rounds of the preset lesion recognition model meets the preset round number threshold, and to obtain the target lesion recognition model corresponding to the model training information.

[0058] The technical solution of this invention involves acquiring model training information for guiding model training, wherein the model training information includes at least a model recognition type; acquiring training sample data for model training based on the model training information, wherein the sample labels of the training sample data correspond to the model recognition type, and the training sample data includes lesion label results for determining the model training accuracy; executing a training task corresponding to the sample label based on a pre-determined preset lesion recognition model and the training sample data, wherein the preset lesion recognition model is a deep neural network model that has not completed model training; determining that the training of the preset lesion recognition model has ended when the number of training rounds of the preset lesion recognition model meets a preset round number threshold, and obtaining the target lesion recognition model corresponding to the model training information. This achieves automatic execution of training tasks corresponding to sample labels based on the sample labels of the training sample data, reducing the complexity of model training and the professionalism of trainers, enhancing personalized model training for complex analysis scenarios, and promoting the customized training development and application of medical models.

[0059] Optionally, the training task execution module 303 is specifically used for: The training sample data is input into a preset lesion recognition model to execute a recognition decision corresponding to the recognition type of the model, and the lesion recognition result is obtained based on the output of the preset lesion recognition model; The training error is determined based on the lesion identification result and the lesion label result, and the training error is backpropagated to the preset lesion identification model to adjust the network parameters in the preset lesion identification model.

[0060] Optionally, the model recognition type includes classification recognition type, segmentation recognition type, and detection recognition type; the sample label includes classification label, segmentation label, and detection label; The sample labels of the training sample data correspond to the model recognition types, including: When the model recognition type is a classification recognition type, the sample label is a classification label; When the model recognition type is segmentation recognition type, the sample label is a segmentation label; When the model recognition type is the detection recognition type, the sample label is a segmentation label, or a segmentation label and a detection label.

[0061] Optionally, the training sample acquisition module 302 includes: The sample label determination unit is used to determine the sample label corresponding to the model recognition type based on the model recognition type in the model training information; The lesion image screening unit is used to screen out multimodal lesion images corresponding to the sample labels from a pre-constructed quality control data pool; The sample data acquisition unit is used to construct training sample data for model training based on the multimodal lesion images.

[0062] Optional, the sample data acquisition unit is specifically used for: Acquire multimodal normal images corresponding to the model recognition type, and determine the multimodal normal images as a negative dataset that does not contain lesion information; The multimodal lesion images are identified as a positive dataset containing lesion information; Based on the positive dataset and the negative dataset, training sample data for model training is constructed.

[0063] Optionally, the process of constructing the quality control data pool includes: Acquire desensitized medical images of various modalities for medical research, and screen out desensitized lesion images with lesion structures from the desensitized medical images; The desensitized lesion images are subjected to lesion annotation processing to obtain lesion image data, wherein the lesion annotation processing includes at least lesion location annotation processing and lesion label annotation processing; The lesion image data is processed and stored in a database to construct the quality control data pool.

[0064] Optionally, the device further includes a recognition model application module. The model application module is used for: After obtaining the target lesion recognition model, the target image data of the lesion to be identified is acquired. The target image data is input into the target lesion recognition model for lesion recognition, and the lesion recognition result is obtained based on the output of the target lesion recognition model.

[0065] The training device for the lesion recognition model provided in the embodiments of the present invention can execute the training method for the lesion recognition model provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0066] Figure 4A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0067] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0068] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0069] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the training methods for lesion recognition models.

[0070] In some embodiments, the method for training the lesion recognition model can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for training the lesion recognition model described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the method for training the lesion recognition model by any other suitable means (e.g., by means of firmware).

[0071] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0072] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0073] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0074] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0075] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0076] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0077] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0078] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A training method for a lesion recognition model, characterized in that, Model training systems applied to no-code training include: Obtain model training information for guiding model training, wherein the model training information includes at least the model recognition type; Based on the model training information, training sample data for model training is obtained, wherein the sample labels of the training sample data correspond to the model recognition type, and the training sample data includes lesion label results used to determine the model training accuracy; Based on a predetermined lesion recognition model and the training sample data, a training task corresponding to the sample label is executed, wherein the predetermined lesion recognition model is a deep neural network model that has not completed model training; When the number of training rounds of the preset lesion recognition model meets the preset round threshold, the training of the preset lesion recognition model is determined to be completed, and the target lesion recognition model corresponding to the model training information is obtained.

2. The method according to claim 1, characterized in that, The step of performing a training task corresponding to the model's recognition type based on a pre-determined preset lesion recognition model and the training sample data includes: The training sample data is input into a preset lesion recognition model to execute a recognition decision corresponding to the recognition type of the model, and the lesion recognition result is obtained based on the output of the preset lesion recognition model; The training error is determined based on the lesion identification result and the lesion label result, and the training error is backpropagated to the preset lesion identification model to adjust the network parameters in the preset lesion identification model.

3. The method according to claim 1, characterized in that, The model recognition types include classification recognition type, segmentation recognition type, and detection recognition type; the sample labels include classification labels, segmentation labels, and detection labels; The sample labels of the training sample data correspond to the model recognition types, including: When the model recognition type is a classification recognition type, the sample label is a classification label; When the model recognition type is segmentation recognition type, the sample label is a segmentation label; When the model recognition type is the detection recognition type, the sample label is a segmentation label, or a segmentation label and a detection label.

4. The method according to claim 1, characterized in that, The step of obtaining training sample data for model training based on the model training information includes: Based on the model recognition type in the model training information, determine the sample label corresponding to the model recognition type; Multimodal lesion images corresponding to the sample labels are selected from a pre-constructed quality control data pool; Based on the multimodal lesion images, training sample data for model training is constructed.

5. The method according to claim 4, characterized in that, The step of constructing training sample data for model training based on the multimodal lesion images includes: Acquire multimodal normal images corresponding to the model recognition type, and determine the multimodal normal images as a negative dataset that does not contain lesion information; The multimodal lesion images are identified as a positive dataset containing lesion information; Based on the positive dataset and the negative dataset, training sample data for model training is constructed.

6. The method according to claim 4, characterized in that, The process of constructing the quality control data pool includes: Acquire desensitized medical images of various modalities for medical research, and screen out desensitized lesion images with lesion structures from the desensitized medical images; The desensitized lesion images are subjected to lesion annotation processing to obtain lesion image data, wherein the lesion annotation processing includes at least lesion location annotation processing and lesion label annotation processing; The lesion image data is processed and stored in a database to construct the quality control data pool.

7. The method according to claim 1, characterized in that, After obtaining the target lesion identification model, the method further includes: Acquire target image data for lesion identification; The target image data is input into the target lesion recognition model for lesion recognition, and the lesion recognition result is obtained based on the output of the target lesion recognition model.

8. A training device for a lesion recognition model, characterized in that, Model training systems applied to no-code training include: The training information acquisition module is used to acquire model training information for guiding model training, wherein the model training information includes at least the model recognition type; The training sample acquisition module is used to acquire training sample data for model training based on the model training information, wherein the sample labels of the training sample data correspond to the model recognition types, and the training sample data includes lesion label results used to determine the model training accuracy. The training task execution module is used to execute a training task corresponding to the sample label according to a pre-determined preset lesion recognition model and the training sample data, wherein the preset lesion recognition model is a deep neural network model that has not completed model training; The training task completion module is used to determine that the training of the preset lesion recognition model has ended when the number of training rounds of the preset lesion recognition model meets the preset round number threshold, and to obtain the target lesion recognition model corresponding to the model training information.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the training method of the lesion recognition model according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the training method of the lesion recognition model according to any one of claims 1-7.

Images