Medical diagnostic model training methods, medical diagnostic methods and related devices

Abstract

This invention provides a medical diagnostic model training method, a medical diagnostic method, and related apparatus. The medical diagnostic model training method includes: acquiring first medical data and second medical data; using the first medical data as training samples and prior knowledge from medical textbooks to train a common-sense expert diagnostic module; using the second medical data as training samples and employing Retrieval Enhancement Generative Diagnosis (RAG) technology to train the LLM to obtain a case-specific expert diagnostic module; using the first diagnostic result generated by the common-sense expert diagnostic module and the second diagnostic result generated by the case-specific expert diagnostic module during the training phase as training samples and employing CoT (Coding of Thought) technology to train the LLM to obtain an unbiased diagnostic module, thereby obtaining a medical diagnostic model. The method described in this invention improves the diagnostic accuracy of medical diagnostic models for diverse diseases.

Description

Technical Field

[0001] This invention relates to the field of assisted medical diagnostic technology, and in particular to a medical diagnostic model training method, a medical diagnostic method, and related devices. Background Technology

[0002] In the development of a medical multimodal large language model (MLLM), selecting appropriate training data is crucial, as it directly affects the model's performance and application effectiveness.

[0003] In related technologies, existing MLLMs typically rely on publicly available medical literature databases, such as PubMed. However, since real-world medical needs are more extensive, using only cases recorded in medical literature databases to assist in disease diagnosis and analysis can introduce specific biases. This can easily lead to biases in multimodal language models due to the limited availability of data, resulting in unreliable diagnostic results and failing to meet the diverse needs of disease management and treatment. Summary of the Invention

[0004] This invention provides a medical diagnostic model training method, a medical diagnostic method, and related devices to address the shortcomings of existing technologies that use MLLM to assist in disease diagnosis, which are prone to generating specific biases and leading to inaccurate diagnostic results. This invention improves the diagnostic accuracy of medical diagnostic models for diverse diseases.

[0005] This invention provides a method for training a medical diagnostic model, comprising:

[0006] Acquire first medical data and second medical data; wherein the first medical data is determined based on medical literature data, and the second medical data is determined based on medical clinical data;

[0007] Using the first medical data as training samples and prior knowledge from medical textbooks as training material, a large-scale language model (LLM) is trained to obtain a common-sense expert diagnostic module. Using the second medical data as training samples, the LLM is trained using retrieval-enhanced generative algorithm (RAG) to obtain a case-specific expert diagnostic module. Using the first diagnostic result generated by the common-sense expert diagnostic module and the second diagnostic result generated by the case-specific expert diagnostic module during the training phase as training samples, the LLM is trained using CoT (Cooperative Thought Technology) to obtain an unbiased diagnostic module. Based on the common-sense expert diagnostic module, the case-specific expert diagnostic module, and the unbiased diagnostic module, the medical diagnostic model is obtained.

[0008] According to a medical diagnostic model training method provided by the present invention, after acquiring the first medical data and the second medical data, the method further includes:

[0009] The first medical data and the second medical data are sequentially cleaned to obtain cleaned data.

[0010] Generative adversarial networks are used to augment the cleaned data to obtain augmented first medical data and augmented second medical data.

[0011] According to a medical diagnostic model training method provided by the present invention, after obtaining the medical diagnostic model, the method further includes:

[0012] The output data of the medical diagnostic model were collected from multiple benchmark datasets.

[0013] A fairness index is calculated based on multiple output data, and the performance of the medical diagnostic model is evaluated based on the fairness index.

[0014] According to a medical diagnostic model training method provided by the present invention, after obtaining the medical diagnostic model, the method further includes:

[0015] Obtain feedback data from clinicians and patients;

[0016] The network parameters of each module of the medical diagnostic model are adjusted based on the feedback data to update the medical diagnostic model.

[0017] The present invention also provides a medical diagnostic method, comprising:

[0018] Acquire medical image data for diagnosis;

[0019] The medical image data is used to diagnose the medical image data based on a medical diagnostic model to obtain a diagnostic result; wherein the medical diagnostic model is trained based on the medical diagnostic model training method.

[0020] The present invention also provides a medical diagnostic model training device, comprising:

[0021] The first data acquisition module is used to acquire first medical data and second medical data; wherein, the first medical data is determined based on medical literature data, and the second medical data is determined based on medical clinical data;

[0022] The training module is used to train a large language model (LLM) using the first medical data as training samples and prior knowledge from medical textbooks as prior knowledge, to obtain a common sense expert diagnosis module; using the second medical data as training samples and employing retrieval-enhanced generative algorithm (RAG) to train the LLM, to obtain a case expert diagnosis module; using the first diagnosis result generated by the common sense expert diagnosis module and the second diagnosis result generated by the case expert diagnosis module during the training phase as training samples and employing CoT (Cooperative Thought Technology) to train the LLM, to obtain an unbiased diagnosis module; and based on the common sense expert diagnosis module, the case expert diagnosis module, and the unbiased diagnosis module, the medical diagnosis model is obtained.

[0023] The present invention also provides a medical diagnostic device, comprising:

[0024] The second data acquisition module is used to acquire medical image data to be diagnosed;

[0025] The diagnostic module is used to diagnose the medical image data based on a medical diagnostic model and obtain diagnostic results; wherein the medical diagnostic model is trained based on the medical diagnostic model training method.

[0026] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the medical diagnostic model training method or medical diagnostic method as described above.

[0027] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the medical diagnostic model training method or medical diagnostic method as described above.

[0028] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the medical diagnostic model training method or medical diagnostic method as described above.

[0029] The medical diagnostic model training method, medical diagnostic method, and related apparatus provided by this invention acquire first and second medical data; using the first medical data as training samples and prior knowledge and experience acquired from medical textbooks as a priori knowledge, a large language model (LLM) is trained to obtain a common sense expert diagnostic module; using the second medical data as training samples and employing retrieval-enhanced generative algorithm (RAG) technology to train the LLM, a case-specific expert diagnostic module is obtained; using the first diagnostic result generated by the common sense expert diagnostic module and the second diagnostic result generated by the case-specific expert diagnostic module during the training phase as training samples and employing CoT (Cooperative Thought Technology) technology to train the LLM, an unbiased diagnostic module is obtained, thereby obtaining a medical diagnostic model and improving the diagnostic accuracy of the medical diagnostic model for diverse diseases. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0031] Figure 1 This is one of the flowcharts illustrating the medical diagnostic model training method provided by the present invention.

[0032] Figure 2 This is a schematic diagram of the process for training an LLM model using RAG technology provided by the present invention.

[0033] Figure 3 This is a schematic diagram of the process for training an LLM model using CoT technology provided by the present invention.

[0034] Figure 4 This is the second flowchart of the medical diagnostic model training method provided by the present invention.

[0035] Figure 5 This is a flowchart illustrating the medical diagnostic method provided by the present invention.

[0036] Figure 6 This is a schematic diagram of the structure of the medical diagnostic model training device provided by the present invention.

[0037] Figure 7 This is a schematic diagram of the structure of the medical diagnostic device provided by the present invention.

[0038] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0040] The following is combined with Figures 1-7 The present invention describes a medical diagnostic model training method, a medical diagnostic method, and related apparatus.

[0041] Figure 1 This is one of the flowcharts illustrating the medical diagnostic model training method provided by the present invention, such as... Figure 1 As shown, the training method for this medical diagnostic model includes the following steps:

[0042] Step 110: Obtain first medical data and second medical data; wherein, the first medical data is determined based on medical literature data, and the second medical data is determined based on medical clinical data.

[0043] In this step, the first medical data includes literature data such as medical books, papers or journals recorded in publicly available medical databases. This type of medical data includes text, image or video data; the first medical data includes a variety of medical common knowledge data.

[0044] In this step, the second medical data includes the patient's clinical condition, diagnosis, treatment process and outcome data. This type of medical data may also include text, image or video data; the second medical data includes diagnostic and analysis data for a variety of rare diseases; for example, the second medical data can be obtained from PMC (PubMed Central) data.

[0045] Step 120: Using the first medical data as training samples and knowledge and experience acquired from medical textbooks as prior knowledge, train the large language model LLM to obtain the common sense expert diagnosis module; using the second medical data as training samples and employing retrieval augmentation generation (RAG) technology to train the LLM to obtain the case expert diagnosis module; using the first diagnosis result generated by the common sense expert diagnosis module during the training phase and the second diagnosis result generated by the case expert diagnosis module during the training phase as training samples and employing CoT (Cooperative Thought Technology) technology to train the LLM to obtain the unbiased diagnosis module; based on the common sense expert diagnosis module, the case expert diagnosis module, and the unbiased diagnosis module, obtain the medical diagnosis model.

[0046] In this step, the initial framework of the medical diagnostic model includes at least three LLMs (Large Language Models). The network parameter settings of each LLM can be exactly the same. For example, the weight factor ratio of the two LLMs used for disease diagnosis can be 1:1. The network parameter settings of each LLM can be different. For example, the weight factor ratio of the two LLMs used for disease diagnosis can be 2:3 or other ratios.

[0047] In this embodiment, the LLM is trained and fine-tuned by combining adjustments and instructions from medical textbooks. For example, by adding instructions to the input image data according to medical textbook knowledge, the model can be guided to generate diagnostic results based on medical common sense required for a specific task.

[0048] In this embodiment, during the initial stage of LLM training, the model is trained using first medical data to learn the statistical patterns, semantic information, and contextual relationships of language. Pre-training typically employs self-supervised learning methods, such as training the model by predicting masked words (Masked Language Model) or the next sentence (Next Sentence Prediction). Through pre-training, LLM can acquire broad language understanding and generation capabilities. After pre-training, LLM is fine-tuned on data specific to certain tasks to adapt to specific medical common sense diagnosis scenarios. During training, feedback reinforcement learning can also be used to guide the model's learning process by comparing the training output results with real results, continuously adjusting network parameters until the model's prediction performance meets expectations, resulting in a usable common sense expert diagnosis module.

[0049] Figure 2 This is a flowchart illustrating the process of training an LLM model using RAG technology provided by the present invention. Figure 2 In the illustrated embodiment, training an LLM using the RAG (Retrieval-Augmented Generation) technique includes the following steps:

[0050] (1) Create quick queries based on user needs;

[0051] (2) Retrieval stage: Retrieve relevant information from pre-built external knowledge sources based on the user's quick query;

[0052] In this embodiment, the external knowledge source can be a structured database, an unstructured text document, or a document collection specifically indexed for this task; the retrieval tool measures the relevance between the query and the documents in the database to achieve effective information retrieval.

[0053] In this embodiment, the type of retrieval tool can be sparse, such as retrieval based on inverted indexes and term frequencies, or dense, such as semantic retrieval based on vector embeddings. The granularity of the retrieval can be document-level, paragraph-level, term-level, or entity-level, depending on the requirements of the task and the design of the system.

[0054] (3) Enhancement stage: The retrieved relevant information is used to enhance the contextual semantics or context of the LLM, for example, by extracting key facts or summaries from the document through natural language processing techniques;

[0055] (4) Generation stage: LLM uses the enhanced context to generate a response by combining the retrieved information with the original query and then generating a coherent, accurate and informative answer through the model;

[0056] In this embodiment, the case expert diagnosis module can also be iterated and optimized. For example, the retrieval strategy and generation model can be continuously adjusted based on user feedback and performance evaluation results to improve overall performance and user experience; or the retrieval or generator can be fine-tuned to adapt to specific tasks or datasets.

[0057] In this embodiment, the Chain of Thought (CoT) technology simulates the step-by-step reasoning process of humans, allowing the model to deduce a series of intermediate steps or sub-goals before generating the final answer. These intermediate steps constitute a "chain of thought," ultimately guiding the model to the correct result.

[0058] Figure 3 This is a flowchart illustrating the process of training an LLM model using CoT technology provided by the present invention. Figure 3 In the illustrated embodiment, training an LLM model using CoT technology can be achieved through the following steps:

[0059] (1) Problem identification: Identify the problems that need to be solved in the training task;

[0060] (2) Decompose into sub-problems and analyze whether these problems require multi-step reasoning.

[0061] (3) To solve each word problem, the intermediate reasoning steps required to solve the problem are designed manually or automatically generated by the model. These steps can be problem decomposition, logical reasoning, calculation process, etc.

[0062] (4) Integrate the solutions and construct a training dataset containing questions, intermediate reasoning steps and final answers; these data can be manually annotated (Few-shot CoT) or automatically generated with specific prompts (Zero-shot CoT).

[0063] (5) Train the model to realize CoT reasoning and conclusion. Use the training data constructed in step (3) above to train the LLM. The model generates reasonable intermediate steps according to the given question and finally obtains the correct answer.

[0064] In this embodiment, the common sense expert diagnosis module and the case expert diagnosis module can share a common input terminal, and the output terminals of the common sense expert diagnosis module and the case expert diagnosis module are respectively connected to the input terminal of the unbiased diagnosis module.

[0065] For example, the image A to be tested is input into the common sense expert diagnosis module and the case expert diagnosis module respectively to obtain the corresponding common sense expert diagnosis results and pathology expert diagnosis results. Then, the common sense expert diagnosis results and pathology expert diagnosis results are input into the unbiased diagnosis module for unbiased reasoning, and the unbiased diagnosis module outputs the corresponding unbiased diagnosis results.

[0066] The medical diagnostic model training method provided in this invention involves acquiring first and second medical data; using the first medical data as training samples and prior knowledge from medical textbooks as training material to train a large language model (LLM) to obtain a common sense expert diagnostic module; using the second medical data as training samples and employing retrieval-enhanced generative algorithm (RAG) to train the LLM to obtain a case-specific expert diagnostic module; and using the first diagnostic result generated by the common sense expert diagnostic module during the training phase and the second diagnostic result generated by the case-specific expert diagnostic module during the training phase as training samples and employing CoT (Cooperation of Thought) technology to train the LLM to obtain an unbiased diagnostic module, thereby obtaining a medical diagnostic model and improving the diagnostic accuracy of the medical diagnostic model for diverse diseases.

[0067] Figure 4 This is the second flowchart of the medical diagnostic model training method provided by the present invention. Figure 4 In the illustrated embodiment, case data (corresponding to patient cases) is obtained from relevant literature and input into the common sense expert diagnosis module (corresponding to LLMA - generalized knowledge model) and the case expert diagnosis module (corresponding to LLMB - rare case model, trained based on RAG technology) to obtain two corresponding diagnostic results (the validity of each other cannot be determined). The two results are then input into the unbiased diagnosis module (corresponding to LLMC - unbiased summarizer, trained based on thought chain technology) to output the final diagnostic result (corresponding to decision: antiphospholipid syndrome caused by systemic lupus erythematosus).

[0068] According to a medical diagnostic model training method provided by the present invention, after acquiring first medical data and second medical data, the method further includes: sequentially cleaning the first medical data and second medical data to obtain cleaned data; and using a generative adversarial network to augment the cleaned data to obtain augmented first medical data and augmented second medical data.

[0069] In this embodiment, the data cleaning method includes missing value handling, such as deleting records containing missing values, filling with the mean or median, or using more complex interpolation methods to handle missing values; it also includes duplicate value handling, such as identifying and deleting duplicate records in the dataset to ensure that each record is unique; it also includes outlier handling, such as detecting outliers or isolated points in the data; it also includes data format standardization, such as converting the data into a uniform format, such as date format, number format, or text format; and it also includes data type conversion, data normalization, and data dimensionality reduction operations.

[0070] In this embodiment, sample augmentation includes geometric transformations, such as flipping, rotating, scaling, translating, and cropping images. These operations can increase sample diversity without altering the image semantics. It also includes color transformations, such as adjusting image brightness, contrast, and saturation to simulate images under different lighting conditions, improving the model's adaptability to lighting changes. Furthermore, it includes noise addition, such as adding Gaussian noise or salt-and-pepper noise to the image, improving the model's robustness to noise. Finally, it includes operations such as blurring, using generative models like Generative Adversarial Networks (GANs) or Variational Autoencoders (VAEs) to generate new data samples, defining rules to generate new data samples, and synthesizing data.

[0071] The present invention provides a medical diagnostic model training method that improves data quality and thus enhances the predictive performance of LLM by cleaning and augmenting the first and second medical data.

[0072] In some embodiments, after obtaining the medical diagnostic model, the method further includes: collecting the output data of the medical diagnostic model on multiple benchmark datasets; calculating a fairness index based on the multiple output data; and evaluating the performance of the medical diagnostic model based on the fairness index.

[0073] In this embodiment, fairness metrics include perplexity (PPL), accuracy, F1 score, etc.; in addition, a dedicated benchmark set can be used to evaluate the model’s performance on a specific task.

[0074] In this embodiment, different datasets are used to test the trained medical diagnostic model, resulting in multiple test results. The corresponding fairness index is calculated based on the multiple test results, and finally the calculated data is compared with a preset threshold to obtain the performance level of the medical diagnostic model.

[0075] This invention provides a method for training a medical diagnostic model. By collecting output data of the medical diagnostic model on multiple benchmark datasets, calculating a fairness index based on the multiple output data, and evaluating the performance of the medical diagnostic model based on the fairness index, a scientific evaluation of the performance of the medical diagnostic model is achieved, thereby improving the reliability of the model's prediction results.

[0076] In some embodiments, after obtaining the medical diagnostic model, the method further includes: acquiring feedback data from clinicians and patients; and adjusting the network parameters of each module of the medical diagnostic model based on the feedback data to update the medical diagnostic model.

[0077] In this embodiment, the medical diagnostic model is continuously optimized in the following manner:

[0078] (1) Establish a feedback mechanism to collect feedback from clinicians and patients and continuously optimize the model.

[0079] Specifically, this can be achieved by collecting user feedback, performance metrics, and error logs after model deployment; analyzing the feedback data to identify performance issues such as low accuracy, low recall, bias, and unfairness; using appropriate evaluation metrics to quantify model performance; deeply analyzing model errors to understand which types of data the model performs poorly on and the reasons for these errors; adjusting model parameters based on the analysis results; retraining the model using the adjusted parameters; or deploying the retrained model to the production environment to replace the old model; and finally, continuously monitoring the performance of the new model to ensure that the adjusted model achieves the expected results and promptly identifying new problems.

[0080] (2) Use new technologies or new large models to iterate and update the model continuously in order to maintain its advanced nature and effectiveness.

[0081] This invention provides a method for training a medical diagnostic model, which obtains feedback data from clinicians and patients; adjusts the network parameters of each module of the medical diagnostic model based on the feedback data to update the medical diagnostic model, thereby achieving online updates of the medical diagnostic model and ensuring the continuous and reliable performance of the medical diagnostic model.

[0082] The medical diagnostic method provided by this invention is described below. The medical diagnostic method described below can be referred to in correspondence with the medical diagnostic model training method described above.

[0083] Figure 5 This is a flowchart illustrating the medical diagnostic method provided by the present invention, as shown below. Figure 5 As shown, this medical diagnostic method includes the following steps:

[0084] Step 510: Obtain the medical image data to be diagnosed;

[0085] Step 520: Diagnose the medical image data based on the medical diagnostic model to obtain the diagnostic results; wherein, the medical diagnostic model is trained based on the medical diagnostic model training method.

[0086] In this step, the medical image data to be diagnosed can be literature data such as medical books, papers or journals recorded in publicly available medical databases, or it can be data on the patient's clinical condition, diagnosis, treatment process and results.

[0087] In this step, the medical diagnostic model includes a common sense expert diagnostic module, a case expert diagnostic module, and an unbiased diagnostic module. The common sense expert diagnostic module uses first medical data as training samples and acquires prior knowledge from medical textbooks to train the large language model (LLM). The case expert diagnostic module uses second medical data as training samples and employs retrieval-enhanced generative algorithm (RAG) to train the LLM. The unbiased diagnostic module uses the first diagnostic result generated by the common sense expert diagnostic module during the training phase and the second diagnostic result generated by the case expert diagnostic module during the training phase as training samples, and employs CoT (Coding in Transcription) technology to train the LLM.

[0088] In this embodiment, the common sense expert diagnosis module, the case expert diagnosis module, and the unbiased diagnosis module are as described in the above embodiments, and will not be repeated here.

[0089] In this embodiment, the diagnostic result can be the common sense expert diagnostic result output by the common sense expert diagnostic module or the case expert diagnostic result output by the case expert diagnostic module, or it can be a new diagnostic result obtained by integrating and reasoning the common sense expert diagnostic result and the case expert diagnostic result.

[0090] The medical diagnostic method provided in this invention uses a medical diagnostic model to diagnose medical image data and obtain diagnostic results, thereby improving the accuracy and efficiency of diagnosing a variety of diseases.

[0091] The medical diagnostic model training device provided by the present invention is described below. The medical diagnostic model training device described below and the medical diagnostic model training method described above can be referred to in correspondence.

[0092] Figure 6 This is a schematic diagram of the medical diagnostic model training device provided by the present invention, as shown below. Figure 6As shown, the medical diagnostic model training device includes a first data acquisition module 610 and a training module 620.

[0093] The first data acquisition module 610 is used to acquire first medical data and second medical data; wherein, the first medical data is determined based on medical literature data, and the second medical data is determined based on medical clinical data;

[0094] Training module 620 is used to train a large language model LLM using the first medical data as training samples and knowledge and experience acquired from medical textbooks as prior knowledge, to obtain a common sense expert diagnosis module; using the second medical data as training samples and employing retrieval augmentation generation (RAG) technology to train the LLM, to obtain a case expert diagnosis module; using the first diagnosis result generated by the common sense expert diagnosis module and the second diagnosis result generated by the case expert diagnosis module during the training phase as training samples and employing CoT (Cooperative Thought Technology) technology to train the LLM, to obtain an unbiased diagnosis module; and based on the common sense expert diagnosis module, the case expert diagnosis module, and the unbiased diagnosis module, a medical diagnosis model is obtained.

[0095] The medical diagnostic model training device provided in this embodiment of the invention acquires first medical data and second medical data; uses the first medical data as training samples and knowledge and experience acquired from medical textbooks as prior knowledge to train a large language model (LLM) to obtain a common sense expert diagnostic module; uses the second medical data as training samples and employs retrieval-enhanced generative algorithm (RAG) to train the LLM to obtain a case expert diagnostic module; uses the first diagnostic result generated by the common sense expert diagnostic module during the training phase and the second diagnostic result generated by the case expert diagnostic module during the training phase as training samples and employs CoT (Cooperative Thought Technology) to train the LLM to obtain an unbiased diagnostic module, thereby obtaining a medical diagnostic model and improving the diagnostic accuracy of the medical diagnostic model for diverse diseases.

[0096] The medical diagnostic device provided by the present invention is described below. The medical diagnostic device described below and the medical diagnostic method described above can be referred to in correspondence.

[0097] Figure 7 This is a schematic diagram of the structure of the medical diagnostic device provided by the present invention, as shown below. Figure 7 As shown, the medical diagnostic device includes a second data acquisition module 710 and a diagnostic module 720.

[0098] The second data acquisition module 710 is used to acquire medical image data to be diagnosed;

[0099] The diagnostic module 720 is used to diagnose medical image data based on a medical diagnostic model and obtain diagnostic results; wherein, the medical diagnostic model is trained based on a medical diagnostic model training method.

[0100] The medical diagnostic device provided in this invention uses a medical diagnostic model to diagnose medical image data and obtain diagnostic results, thereby improving the accuracy and efficiency of diagnosing a variety of diseases.

[0101] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 8 As shown, the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communications bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other through the communications bus 840. The processor 810 can call logical instructions in the memory 830 to execute a medical diagnostic model training method, which includes: acquiring first medical data and second medical data; wherein the first medical data is determined based on medical literature data, and the second medical data is determined based on medical clinical data; using the first medical data as training samples, and using knowledge and experience acquired from medical textbooks as prior knowledge, training a large language model LLM to obtain a common sense expert diagnostic module; using the second medical data as training samples, and using retrieval-enhanced generative algorithm (RAG) to train the LLM to obtain a case expert diagnostic module; using the first diagnostic result generated by the common sense expert diagnostic module during the training phase and the second diagnostic result generated by the case expert diagnostic module during the training phase as training samples, and using CoT (Coding in Thinking) technology to train the LLM to obtain an unbiased diagnostic module; and obtaining a medical diagnostic model based on the common sense expert diagnostic module, the case expert diagnostic module, and the unbiased diagnostic module.

[0102] Alternatively, a medical diagnostic method may be performed, which includes: acquiring medical image data to be diagnosed; diagnosing the medical image data based on a medical diagnostic model to obtain a diagnostic result; wherein the medical diagnostic model is trained based on a medical diagnostic model training method.

[0103] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0104] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the medical diagnostic model training method provided by the above methods. The method includes: acquiring first medical data and second medical data; wherein the first medical data is determined based on medical literature data, and the second medical data is determined based on medical clinical data; using the first medical data as training samples and using knowledge and experience obtained from medical textbooks as prior knowledge to train a large language model LLM to obtain a common sense expert diagnostic module; using the second medical data as training samples and using retrieval enhancement generation (RAG) technology to train the LLM to obtain a case expert diagnostic module; using the first diagnostic result generated by the common sense expert diagnostic module during the training phase and the second diagnostic result generated by the case expert diagnostic module during the training phase as training samples and using CoT (Coding in Thinking) technology to train the LLM to obtain an unbiased diagnostic module; and obtaining a medical diagnostic model based on the common sense expert diagnostic module, the case expert diagnostic module, and the unbiased diagnostic module.

[0105] Alternatively, a medical diagnostic method may be performed, which includes: acquiring medical image data to be diagnosed; diagnosing the medical image data based on a medical diagnostic model to obtain a diagnostic result; wherein the medical diagnostic model is trained based on a medical diagnostic model training method.

[0106] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a medical diagnostic model training method provided by the above methods. The method includes: acquiring first medical data and second medical data; wherein the first medical data is determined based on medical literature data, and the second medical data is determined based on medical clinical data; training a large language model (LLM) using the first medical data as training samples and prior knowledge acquired from medical textbooks as prior knowledge to obtain a common sense expert diagnostic module; training the LLM using the second medical data as training samples and employing retrieval-enhanced generative algorithm (RAG) to obtain a case expert diagnostic module; using the first diagnostic result generated by the common sense expert diagnostic module during the training phase and the second diagnostic result generated by the case expert diagnostic module during the training phase as training samples and employing CoT (Coding of Thought) technology to train the LLM to obtain an unbiased diagnostic module; and obtaining a medical diagnostic model based on the common sense expert diagnostic module, the case expert diagnostic module, and the unbiased diagnostic module.

[0107] Alternatively, a medical diagnostic method may be performed, which includes: acquiring medical image data to be diagnosed; diagnosing the medical image data based on a medical diagnostic model to obtain a diagnostic result; wherein the medical diagnostic model is trained based on a medical diagnostic model training method.

[0108] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0109] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for training a medical diagnostic model, characterized in that, include: Acquire first medical data and second medical data; wherein the first medical data is determined based on medical literature data, and the second medical data is determined based on medical clinical data; Using the first medical data as training samples and prior knowledge from medical textbooks as training material, a large-scale language model (LLM) is trained to obtain a common-sense expert diagnostic module. Using the second medical data as training samples, the LLM is trained using retrieval-enhanced generative algorithm (RAG) to obtain a case-specific expert diagnostic module. Using the first diagnostic result generated by the common-sense expert diagnostic module and the second diagnostic result generated by the case-specific expert diagnostic module during the training phase as training samples, the LLM is trained using CoT (Cooperative Thought Technology) to obtain an unbiased diagnostic module. Based on the common-sense expert diagnostic module, the case-specific expert diagnostic module, and the unbiased diagnostic module, the medical diagnostic model is obtained.

2. The medical diagnostic model training method according to claim 1, characterized in that, After acquiring the first medical data and the second medical data, the method further includes: The first medical data and the second medical data are sequentially cleaned to obtain cleaned data. Generative adversarial networks are used to augment the cleaned data to obtain augmented first medical data and augmented second medical data.

3. The medical diagnostic model training method according to claim 1, characterized in that, After obtaining the medical diagnostic model, the method further includes: The output data of the medical diagnostic model were collected from multiple benchmark datasets. A fairness index is calculated based on multiple output data, and the performance of the medical diagnostic model is evaluated based on the fairness index.

4. The medical diagnostic model training method according to claim 1, characterized in that, After obtaining the medical diagnostic model, the method further includes: Obtain feedback data from clinicians and patients; The network parameters of each module of the medical diagnostic model are adjusted based on the feedback data to update the medical diagnostic model.

5. A medical diagnostic method, characterized in that, include: Acquire medical image data for diagnosis; The medical image data is used to diagnose based on a medical diagnostic model to obtain a diagnostic result; wherein the medical diagnostic model is trained based on the medical diagnostic model training method as described in any one of claims 1-4.

6. A medical diagnostic model training device, characterized in that, include: The first data acquisition module is used to acquire first medical data and second medical data; wherein, the first medical data is determined based on medical literature data, and the second medical data is determined based on medical clinical data; The training module is used to train a large language model (LLM) using the first medical data as training samples and prior knowledge from medical textbooks as prior knowledge, to obtain a common-sense expert diagnosis module; using the second medical data as training samples and employing retrieval-enhanced generative algorithm (RAG) to train the LLM, to obtain a case expert diagnosis module; using the first diagnosis result generated by the common-sense expert diagnosis module and the second diagnosis result generated by the case expert diagnosis module during the training phase as training samples and employing CoT (Cooperative Thought Technology) to train the LLM, to obtain an unbiased diagnosis module; and based on the common-sense expert diagnosis module, the case expert diagnosis module, and the unbiased diagnosis module, the medical diagnosis model is obtained.

7. A medical diagnostic device, characterized in that, include: The second data acquisition module is used to acquire medical image data to be diagnosed; A diagnostic module is used to diagnose the medical image data based on a medical diagnostic model to obtain a diagnostic result; wherein the medical diagnostic model is trained based on the medical diagnostic model training method as described in any one of claims 1-4.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 5.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 5.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 5.

Images