Stroke secondary prevention model and generation method and device thereof

Abstract

The present disclosure provides a stroke secondary prevention model and a generation method and device thereof, and belongs to the technical field of artificial intelligence, which can solve the problem that the existing general large model generation result cannot meet the requirements of the professional medical field. The stroke secondary prevention model generation method of the present disclosure comprises: processing stroke clinical data to construct a training data set; using the training data set to fine-tune a first general large model to obtain a stroke secondary prevention model; using the stroke secondary prevention model to perform reasoning and generate a reasoning result and reasoning basis; and based on the reasoning result and reasoning basis, clinically verifying and evaluating the stroke secondary prevention model.

Description

Technical Field

[0001] This disclosure belongs to the field of artificial intelligence technology, specifically relating to a secondary prevention model for stroke and its generation method and device. Background Technology

[0002] With the rapid development of artificial intelligence, general-purpose large models have been widely used in various fields. Although these general-purpose large models currently possess a certain degree of generalization ability, they are severely lacking in their ability to target specific professional fields, especially in the medical and health field.

[0003] Taking the currently popular open-source general-purpose large model Llama3.1-405B as an example, although Llama3.1-405B has a very large parameter level (405 billion) and strong generalization ability, compared with large models in the medical and health professional field, it still has the problem of giving vague answers or avoiding answers. Summary of the Invention

[0004] This disclosure aims to at least solve one of the technical problems existing in the prior art, and provides a secondary prevention model for stroke and its generation method and device.

[0005] In a first aspect, embodiments of this disclosure provide a method for generating a secondary prevention model for stroke, the method comprising:

[0006] Clinical data on stroke are processed to construct a training dataset;

[0007] Using the training dataset, the first general large model is fine-tuned to obtain a secondary prevention model for stroke.

[0008] Using a secondary prevention model for stroke, reasoning is performed and reasoning results and reasoning basis are generated;

[0009] Based on the reasoning results and the reasoning basis, the secondary prevention model for stroke was clinically validated and evaluated.

[0010] In some embodiments, processing the clinical data of stroke to construct a training dataset includes:

[0011] The unstructured clinical data of stroke were converted into semi-structured data.

[0012] Key information is extracted from the semi-structured data using regular expressions;

[0013] Using the second general large model, multi-round question-and-answer data is constructed around the key information;

[0014] The longest common substring matching algorithm is used to evaluate the degree of merging between the generated question-and-answer data and the original data, so as to select the question-and-answer data that meets the preset degree of merging.

[0015] The question-and-answer data is denoised to construct the training dataset.

[0016] In some embodiments, the step of extracting key information from the semi-structured data using regular expressions further includes:

[0017] If the regular expression fails, the key information is extracted from the semi-structured data using the second general large model.

[0018] In some embodiments, the step of fine-tuning the first general large model using the training dataset to obtain a secondary stroke prevention model includes:

[0019] Based on the training dataset, the first general large model is trained using a small amount of labeled data combined with a transfer learning strategy.

[0020] In some embodiments, the step of fine-tuning the first general large model using the training dataset to obtain a secondary stroke prevention model includes:

[0021] Based on the training dataset, the first general large model is trained using a hinting tuning technique.

[0022] In some embodiments, the step of using the secondary stroke prevention model to perform reasoning and generate reasoning results and reasoning basis further includes:

[0023] Using the aforementioned secondary stroke prevention model, reasoning is performed and reasoning process data is generated.

[0024] In some embodiments, the step of using the secondary stroke prevention model to perform reasoning and generate reasoning results and reasoning basis further includes:

[0025] Using the aforementioned secondary stroke prevention model, inferences are performed and multi-dimensional explanatory data is generated.

[0026] In some embodiments, the multidimensional interpretation data includes: medical data and statistics for doctors, and disease data and prevention recommendations for patients.

[0027] Secondly, this disclosure provides a device for generating a secondary stroke prevention model, the device comprising:

[0028] The dataset building module is configured to process clinical data on stroke to build a training dataset;

[0029] The fine-tuning module is configured to fine-tune the first general large model using the training dataset to obtain a secondary prevention model for stroke.

[0030] The reasoning module is configured to use the secondary prevention model for stroke to perform reasoning and generate reasoning results and reasoning basis;

[0031] The verification and evaluation module is configured to perform clinical verification and evaluation of the secondary stroke prevention model based on the inference results and the inference basis.

[0032] Thirdly, this disclosure provides a secondary stroke prevention model, which is generated by the secondary stroke prevention model generation method provided above.

[0033] Fourthly, embodiments of this disclosure provide an electronic device, the electronic device comprising:

[0034] At least one processor; and

[0035] A memory communicatively connected to the at least one processor; wherein,

[0036] The memory stores one or more computer programs that can be executed by the at least one processor, such that the at least one processor can perform the stroke secondary prevention model generation method provided above.

[0037] Fifthly, embodiments of this disclosure provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for generating a secondary prevention model for stroke as described above. Attached Figure Description

[0038] Figure 1 This is a schematic diagram illustrating the response of a general large model to the field of healthcare.

[0039] Figure 2 This is a schematic diagram illustrating the answers to questions about healthcare from a large-scale professional medical model.

[0040] Figure 3 This is a flowchart illustrating the method for generating a secondary prevention model for stroke provided in an embodiment of this disclosure.

[0041] Figure 4 This is a schematic diagram illustrating the process of constructing a training dataset according to an embodiment of this disclosure.

[0042] Figure 5 This is a schematic diagram of the structure of the stroke secondary prevention model generation device provided in the embodiments of this disclosure.

[0043] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. The components of the embodiments of this disclosure described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this disclosure provided in the accompanying drawings is not intended to limit the scope of the claimed disclosure, but merely represents selected embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure. Without conflict, the various embodiments of this disclosure and the features in the embodiments can be combined with each other.

[0045] Unless otherwise defined, the technical or scientific terms used in this disclosure shall have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms “first,” “second,” and similar terms used in this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an,” “a,” or “the,” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising,” “including,” or “including,” and similar terms mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects.

[0046] In this disclosure, "multiple or several" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0047] To facilitate understanding, the special terms involved in the embodiments of this disclosure will be explained in detail first.

[0048] Artificial Intelligence (AI) refers to the ability of computer systems to perform tasks that typically require human intelligence. These tasks can include, but are not limited to, learning, reasoning, problem-solving, knowledge representation, perception, planning, natural language processing, vision, and motion control.

[0049] Large Language Models (LLMs) are a class of artificial intelligence systems based on deep learning techniques. They are trained on massive amounts of text data to understand and generate natural language. These models typically have millions to billions of parameters and excel in various natural language processing tasks, such as text generation, question answering, translation, summarization, and dialogue systems.

[0050] Fine-tuning is an important concept in machine learning, generally referring to the process of further training a pre-trained model (usually a general model trained on a large amount of data) for a specific task or dataset. This approach can fully utilize the general features learned by the pre-trained model and adjust it to better suit the performance of the specific task.

[0051] Secondary prevention of stroke refers to a series of measures taken for patients who have experienced one or more strokes, aimed at reducing their risk of recurrence.

[0052] General-Purpose Large Models (GPLAMs) refer to AI models trained on a large scale that can handle multiple types of natural language processing tasks without requiring specific training for each task. These models acquire language understanding capabilities through unsupervised or self-supervised learning on massive amounts of data and can adapt to different application scenarios, such as question answering, text generation, translation, and sentiment analysis.

[0053] Stroke is one of the leading causes of death and disability worldwide. According to statistics from the World Health Organization, approximately 15 million people die or become disabled from stroke each year globally. China is one of the countries with a high incidence of stroke, which is closely related to factors such as high-salt diets, rising smoking rates, and an aging population. Secondary prevention of stroke has become an important means of reducing mortality and disability rates. Secondary prevention of stroke involves individuals who have already experienced a stroke or related symptoms, as well as how to prevent future stroke events through interventions.

[0054] For secondary prevention of stroke, general-purpose large-scale artificial intelligence models play an important role. Taking the currently popular open-source general-purpose large-scale model Llama 3.1-405B as an example... Figure 1 This is a schematic diagram illustrating a general large-scale model's response in the field of healthcare, by... Figure 1 It can be seen that the current generalized model only has one relatively general and generalized reasoning method, and it cannot answer the specific cases mentioned in the question. Figure 2 This is a schematic diagram illustrating the answers to questions in the healthcare field from a large-scale professional medical model, such as... Figure 2 As shown, large-scale models in specialized medical fields will present conclusions that are closer to reality (e.g., multicenter, double-blind, placebo-controlled). Therefore, general-purpose large-scale models cannot solve specific problems in specialized medical fields and cannot meet the needs of specialized medical fields.

[0055] To at least solve one of the aforementioned technical problems, this disclosure provides a secondary stroke prevention model and its generation method and apparatus. The following will describe in further detail the secondary stroke prevention model and its generation method and apparatus provided by this disclosure in conjunction with the accompanying drawings and specific embodiments.

[0056] In a first aspect, embodiments of this disclosure provide a method for generating a secondary prevention model for stroke. Figure 3 This is a flowchart illustrating the method for generating a secondary stroke prevention model according to an embodiment of this disclosure, as shown below. Figure 3 As shown, the method for generating a secondary prevention model for stroke includes the following steps S301 to S304.

[0057] S301 processes clinical data on stroke to construct a training dataset.

[0058] In step S301 above, clinical data for stroke patients can be obtained from partner hospitals, covering medical records, diagnostic information, treatment records, laboratory test results, discharge summaries, secondary prevention plans, and publicly available medical knowledge bases. The partner hospitals can be stroke-specialized hospitals, ensuring more comprehensive and accurate clinical data. Furthermore, the obtained clinical data will only be used for medical research to avoid infringing on patient privacy. Additionally, it is unnecessary to widely obtain clinical data from the internet to avoid restrictions imposed by national-level data security controls.

[0059] Since the acquired clinical data on stroke generally includes data in different formats such as CT images, X-ray images, electroencephalograms, and drug prescription texts, the acquired clinical data can be processed to form data that can be recognized by a general large model, and the data that can be recognized by the general large model can be used to build a training dataset.

[0060] S302 uses the training dataset to fine-tune the first general large model to obtain a secondary prevention model for stroke.

[0061] In step S302 above, the first general large model is used as the base model, for example, the Llama-3.1-8B-Instruct general large model can be selected as the base model. The understanding and predictive capabilities of this general large model for a large amount of general knowledge are utilized as the basis for subsequent fine-tuning. It is understandable that other general large models can also be selected as the base model, such as DeepSeek-R1.

[0062] The first general-purpose large model is an AI neural network trained on a large amount of unlabeled datasets, capable of handling various tasks from translating text to analyzing medical images. The obtained training dataset is input into the pre-trained first general-purpose large model to fine-tune it, resulting in a stroke secondary prevention model. This model can then address specific problems in the professional medical field (stroke secondary prevention), meeting the needs of that field.

[0063] S303 utilizes a secondary stroke prevention model to perform reasoning and generate reasoning results and reasoning basis.

[0064] In step 303 above, during the reasoning process of the secondary stroke prevention model, in addition to generating reasoning results, such as stroke diagnosis results and stroke prevention methods, a reasoning evidence generation module is also introduced. This module can automatically generate reasoning evidence based on multi-source data, forming an evidence chain. Each prediction result is accompanied by an evidence chain supporting the conclusion, including references, experimental data, clinical cases, etc., providing a theoretical basis for the reasoning process of the secondary stroke prevention model.

[0065] S304, based on the reasoning results and reasoning basis, conducts clinical validation and evaluation of the secondary prevention model for stroke.

[0066] In step S304 above, a clinical trial is conducted at a cooperating hospital, and clinical trial data is obtained. The clinical trial data is compared with the inference results to determine whether the two are the same, so as to verify the secondary prevention model of stroke. The prediction accuracy and reliability of the secondary prevention model of stroke in real scenarios are verified. The applicability of the model in different populations is tested through multi-center and diversified datasets, and the fine-tuning parameters are adjusted to improve the general applicability of the secondary prevention model of stroke.

[0067] Standard medical evaluation metrics, such as Receiver Operating Characteristic Curve (ROC), Area Under Curve (AUC), and F1 score, are used to evaluate the effectiveness of the secondary stroke prevention model. The performance of the general large model, the unadjusted first general large model, and the finely adjusted secondary stroke prevention model in this embodiment are compared on the same dataset to assess the effectiveness of the secondary stroke prevention model.

[0068] The stroke secondary prevention model generation method provided in this disclosure uses clinical data obtained from cooperating hospitals as a foundation to construct a training dataset, which improves the comprehensiveness and accuracy of the data, thus enabling the construction of a high-quality training dataset. During model training, the high-quality training dataset is used to fine-tune a general-purpose model, resulting in a stroke secondary prevention model. This model not only possesses the understanding and predictive capabilities of a general-purpose model but also has analytical capabilities specific to the medical field (stroke secondary prevention), and it does not require a large amount of data for training, reducing training costs. During the model's inference process, while generating prediction results, corresponding evidence links can also be generated, thereby improving the professional interpretability of the stroke secondary prevention model. Furthermore, based on the inference results and reasoning basis, the stroke secondary prevention model is validated and evaluated, improving its general applicability and effectiveness.

[0069] In some embodiments, Figure 4 This is a schematic diagram illustrating the process of constructing a training dataset according to an embodiment of this disclosure, such as... Figure 4 As shown, step S301 above processes the clinical data of stroke to construct a training dataset, including steps S401 to S405 below.

[0070] S401 converts unstructured clinical data on stroke into semi-structured data.

[0071] In step S401 above, unstructured data refers to data with irregular or incomplete structures, lacking predefined definitions, and inconvenient to represent using a two-dimensional logical table in a database. Structured data refers to data that can be represented and stored using a relational database, appearing in a two-dimensional form. Its general characteristics are that data is organized by rows, with each row representing information about one entity, and all rows sharing the same attributes. Semi-structured data is a form of structured data that does not conform to the data model structure associated with relational databases or other data tables, but it contains relevant tags used to separate semantic elements and to hierarchically organize records and fields.

[0072] S402 uses regular expressions to extract key information from semi-structured data.

[0073] In step S402 above, extracting key information is an important task in natural language processing, involving the identification and extraction of important information elements from text, such as entities, relationships, and events. Regular expressions are text patterns that include ordinary characters (e.g., letters from a to z) and special characters (called "metacharacters"). Regular expressions use a single string to describe and match a series of strings that match a certain syntax rule, and are commonly used to retrieve and replace text that conforms to a certain pattern (rule). Regular expressions can be used to extract key information from semi-structured data, retaining only information effective for secondary stroke prevention, such as drug treatment information, lifestyle intervention information, rehabilitation training information, and regular check-up information. Drug treatment information includes antiplatelet drugs, antihypertensive drugs, lipid-lowering drugs, and hypoglycemic drugs. Lifestyle intervention information includes adjusting dietary structure, increasing exercise, quitting smoking and limiting alcohol consumption, and maintaining emotional stability. Rehabilitation training information includes physical rehabilitation, speech rehabilitation, and cognitive rehabilitation. Regular check-up information includes regular check-ups of cerebrovascular imaging, electrocardiograms, and blood biochemical indicators.

[0074] S403 utilizes the second general large model to construct multi-round question-and-answer data around key information.

[0075] In step S403 above, a second general large model is used to construct multi-round question-and-answer data based on the extracted key information, so that the extracted key information meets the question-and-answer scenario. Specifically, the second general large model can be Chat GPT (Chat Generative Pre-trained Transformer).

[0076] S404 employs the longest common substring matching algorithm to evaluate the degree of merging between the generated question-and-answer data and the original data, in order to select question-and-answer data that meets the preset merging degree.

[0077] In step S404 above, the Longest Common Substring (LCS) is a string matching algorithm used to find the longest common continuous substring between two strings. The LCS matching algorithm can be used to evaluate the degree of merging between the generated question-and-answer data and the original data, selecting question-and-answer data that meets a preset merging degree, thus selectively retaining high-quality question-and-answer data. The degree of data merging refers to the extent to which data from different data sources (such as generated question-and-answer data and original data) are integrated into a unified dataset for better analysis, processing, or reporting. The data merging process typically includes eliminating data redundancy, handling mismatched data, ensuring data consistency and comparability, and integrating data from different sources into a consistent data structure for more in-depth analysis or decision-making.

[0078] S405, denoise the question-and-answer data and construct the training dataset.

[0079] In step S405 above, the question-and-answer data is denoised to remove incomplete or non-compliant data and retain complete or compliant data. The retained complete or compliant data is then used to construct a training dataset to ensure the comprehensiveness and accuracy of the data.

[0080] In some embodiments, such as Figure 4 As shown, step S402 above uses regular expressions to extract key information from semi-structured data. It then includes step S402A: if the regular expression fails, the second general large model is used to extract key information from the semi-structured data.

[0081] In step S402A above, if the regular expression fails when extracting key information, the second general large model is used to extract key information from the semi-structured data again. Specifically, the general large model can be Chat GPT.

[0082] In some embodiments, step S302 above, which involves fine-tuning the first general large model using the training dataset to obtain a secondary prevention model for stroke, includes: training the first general large model using a small amount of labeled data combined with a transfer learning strategy based on the training dataset.

[0083] Fine-tuning can be achieved by using a small amount of labeled data combined with transfer learning strategies. This reduces reliance on large amounts of data and improves the accuracy of the secondary stroke prevention model. Transfer learning is a machine learning method that refers to using the model developed for task A (the first general model) as an initial point and reusing it in the process of developing a model for task B (the secondary stroke prevention model).

[0084] In some embodiments, step S302 above, which involves fine-tuning the first general large model using the training dataset to obtain a secondary prevention model for stroke, includes: training the first general large model using cue-based tuning techniques based on the training dataset.

[0085] Fine-tuning can employ prompt tuning techniques to optimize the ability of stroke secondary prevention models to understand and respond to specific medical questions, especially in scenarios involving complex medical reasoning. Prompt tuning guides the model to complete specific tasks by designing "cue words" or "context." Traditional fine-tuning methods require updating a large number of model parameters, while prompt tuning only needs to introduce some additional cues, enabling the model to understand the task and generate the required results with minimal computational cost.

[0086] In some embodiments, such as Figure 3 As shown, step S303 above uses a secondary stroke prevention model to perform reasoning and generate reasoning results and reasoning basis, and then includes step S305.

[0087] S305 uses a secondary stroke prevention model to perform reasoning and generate reasoning process data.

[0088] In step S305 above, for each suggestion or inference, the secondary stroke prevention model outputs not only the prediction result, but also detailed reasoning process data to ensure the traceability and interpretability of the results.

[0089] In some embodiments, such as Figure 3 As shown, step S303 above uses a secondary stroke prevention model to perform reasoning and generate reasoning results and reasoning basis, and then includes step S306.

[0090] S306 utilizes a secondary stroke prevention model to reason and generate multi-dimensional explanatory data.

[0091] In step S306 above, the stroke secondary prevention model can generate multi-dimensional interpretive data to meet the different needs of doctors and patients. For example, it can provide medical and statistical data for doctors, and disease data and prevention recommendations for patients. Specifically, the stroke secondary prevention model can summarize key information from records generated by doctors during stroke secondary prevention management of patients, providing a reference for doctors to write subsequent clinical notes and improving management efficiency. Simultaneously, based on patient health records, it provides personalized secondary prevention plans, helping doctors to manage stroke secondary prevention more accurately and effectively.

[0092] Secondly, this disclosure provides a device for generating a secondary prevention model for stroke. Figure 5 This is a schematic diagram of the structure of the stroke secondary prevention model generation device provided in the embodiments of this disclosure, as shown below. Figure 5 As shown, the stroke secondary prevention model generation device includes: a dataset construction module 501, a fine-tuning module 502, an inference module 503, and a verification and evaluation module 504.

[0093] The dataset construction module 501 is configured to process stroke clinical data to construct a training dataset. The fine-tuning module 502 is configured to fine-tune the first general model using the training dataset to obtain a secondary stroke prevention model. The inference module 503 is configured to perform inference using the secondary stroke prevention model and generate inference results and inference justification. The validation and evaluation module 504 is configured to perform clinical validation and evaluation of the secondary stroke prevention model based on the inference results and inference justification.

[0094] Clinical data on stroke patients can be obtained from partner hospitals, encompassing medical records, diagnostic information, treatment records, laboratory test results, discharge summaries, secondary prevention plans, and publicly available medical knowledge bases. Partner hospitals can be stroke-specialized hospitals, ensuring more comprehensive and accurate clinical data. Furthermore, the acquired clinical data will be used solely for medical research to avoid infringing on patient privacy. Additionally, it is unnecessary to widely obtain clinical data from the internet to avoid restrictions imposed by national-level data security controls.

[0095] Since the acquired clinical data on stroke generally includes data in different formats such as CT images, X-ray images, electroencephalograms, and drug prescription texts, the acquired clinical data can be processed to form data that can be recognized by a general large model, and the data that can be recognized by the general large model can be used to build a training dataset.

[0096] The first general-purpose large model can be used as the base model, for example, the Llama-3.1-8B-Instruct general-purpose large model can be selected as the base model. The understanding and predictive capabilities of this general-purpose large model for a large amount of general knowledge can be used as the basis for subsequent fine-tuning. Understandably, other general-purpose large models can also be selected as the base model, such as DeepSeek-R1.

[0097] The first general-purpose large model is an AI neural network trained on a large amount of unlabeled datasets, capable of handling various tasks from translating text to analyzing medical images. The obtained training dataset is input into the pre-trained first general-purpose large model to fine-tune it, resulting in a stroke secondary prevention model. This model can then address specific problems in the professional medical field (stroke secondary prevention), meeting the needs of that field.

[0098] In the reasoning process of the secondary stroke prevention model, in addition to generating reasoning results, such as stroke diagnosis results and stroke prevention methods, a reasoning evidence generation module is also introduced. This module can automatically generate reasoning evidence based on multi-source data, forming an evidence chain. Each prediction result is accompanied by an evidence chain supporting the conclusion, including references, experimental data, clinical cases, etc., providing a theoretical basis for the reasoning process of the secondary stroke prevention model.

[0099] Clinical trials were conducted at partner hospitals to obtain clinical trial data. The clinical trial data was compared with the inference results to determine whether they were the same, in order to validate the secondary prevention model of stroke. The predictive accuracy and reliability of the secondary prevention model of stroke in real-world scenarios were verified. The applicability of the model in different populations was tested through multi-center, diverse datasets, and the fine-tuning parameters were adjusted to improve the general applicability of the secondary prevention model of stroke.

[0100] Standard medical evaluation metrics, such as Receiver Operating Characteristic Curve (ROC), Area Under Curve (AUC), and F1 score, are used to evaluate the effectiveness of the secondary stroke prevention model. The performance of the general large model, the unadjusted first general large model, and the finely adjusted secondary stroke prevention model in this embodiment are compared on the same dataset to assess the effectiveness of the secondary stroke prevention model.

[0101] In the stroke secondary prevention model generation device provided in this embodiment, the dataset construction module 501 can construct a training dataset based on clinical data obtained from cooperating hospitals, which can improve the comprehensiveness and accuracy of the data, thus enabling the construction of a high-quality training dataset. During model training, the fine-tuning module 502 can use the high-quality training dataset to fine-tune the general-purpose model, obtaining a stroke secondary prevention model. This allows the stroke secondary prevention model to not only possess the understanding and prediction capabilities of the general-purpose model but also the analytical capabilities specific to the medical field (stroke secondary prevention), and it does not require a large amount of data for model training, reducing training costs. During model inference, the inference module 503 can generate the evidence chain corresponding to the prediction results while generating the prediction results, thereby improving the professional interpretability of the stroke secondary prevention model. Furthermore, the verification and evaluation module 504 can verify and evaluate the stroke secondary prevention model based on the inference results and inference evidence, improving the generality and effectiveness of the stroke secondary prevention model.

[0102] It should be noted that each module in the above-mentioned stroke secondary prevention model generation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module to realize the stroke secondary prevention model generation method provided in any of the above embodiments.

[0103] Thirdly, this disclosure provides a secondary stroke prevention model, which is generated by the secondary stroke prevention model generation method provided in any of the above embodiments. Its implementation principle and beneficial effects are similar to those of the secondary stroke prevention model generation method provided in any of the above embodiments, and will not be repeated here.

[0104] Fourthly, embodiments of this disclosure provide an electronic device, Figure 6This is a schematic diagram of the structure of the electronic device provided in the embodiments of this disclosure, such as... Figure 6 As shown, the electronic device provided in this embodiment includes: at least one processor 601; at least one memory 602; and one or more I / O interfaces 603 connected between the processor 601 and the memory 602; wherein the memory 602 stores one or more computer programs that can be executed by at least one processor 601, and the one or more computer programs are executed by at least one processor 601 to enable at least one processor 601 to execute the stroke secondary prevention model generation method provided in any of the above embodiments.

[0105] Each module in the aforementioned electronic device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module to implement the stroke secondary prevention model generation method provided in any of the above embodiments.

[0106] Fifthly, embodiments of this disclosure provide a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor / processor core, implements the stroke secondary prevention model generation method provided in any of the above embodiments. The computer-readable storage medium may be a volatile or non-volatile computer-readable storage medium.

[0107] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer storage media (or non-transitory media) and communication media (or transient media).

[0108] As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable program instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), flash memory or other memory technologies, portable compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable program instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0109] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0110] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0111] The computer program product described herein can be implemented specifically through hardware, software, or a combination thereof. In one alternative embodiment, the computer program product is specifically embodied in a computer storage medium; in another alternative embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0112] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0113] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0114] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0115] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0116] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in connection with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in connection with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.

Claims

1. A method for generating a secondary prevention model for stroke, characterized in that, The method for generating the secondary prevention model for stroke includes: Clinical data on stroke are processed to construct a training dataset; Using the training dataset, the first general large model is fine-tuned to obtain a secondary prevention model for stroke. Using the aforementioned secondary prevention model for stroke, reasoning is performed, and reasoning results and reasoning basis are generated; Based on the reasoning results and the reasoning basis, the secondary prevention model for stroke was clinically validated and evaluated.

2. The method for generating a secondary prevention model for stroke according to claim 1, characterized in that, The process of processing clinical data on stroke to construct a training dataset includes: The unstructured clinical data of stroke were converted into semi-structured data. Key information is extracted from the semi-structured data using regular expressions; Using the second general large model, multi-round question-and-answer data is constructed around the key information; The longest common substring matching algorithm is used to evaluate the degree of merging between the generated question-and-answer data and the original data, so as to select the question-and-answer data that meets the preset degree of merging. The question-and-answer data is denoised to construct the training dataset.

3. The method for generating a secondary prevention model for stroke according to claim 2, characterized in that, The step of using regular expressions to extract key information from the semi-structured data further includes: If the regular expression fails, the key information is extracted from the semi-structured data using the second general large model.

4. The method for generating a secondary prevention model for stroke according to claim 1, characterized in that, The step of fine-tuning the first general large model using the training dataset to obtain a secondary stroke prevention model includes: Based on the training dataset, the first general large model is trained using a small amount of labeled data combined with a transfer learning strategy.

5. The method for generating a secondary prevention model for stroke according to claim 1, characterized in that, The step of fine-tuning the first general large model using the training dataset to obtain a secondary stroke prevention model includes: Based on the training dataset, the first general large model is trained using a hinting tuning technique.

6. The method for generating a secondary prevention model for stroke according to claim 1, characterized in that, The process of using the aforementioned secondary stroke prevention model to perform reasoning and generate reasoning results and reasoning basis further includes: Using the aforementioned secondary stroke prevention model, reasoning is performed and reasoning process data is generated.

7. The method for generating a secondary prevention model for stroke according to claim 1, characterized in that, The process of using the aforementioned secondary stroke prevention model to perform reasoning and generate reasoning results and reasoning basis further includes: Using the aforementioned secondary stroke prevention model, inferences are performed and multi-dimensional explanatory data is generated.

8. The method for generating a secondary prevention model for stroke according to claim 7, characterized in that, The multi-dimensional data interpretation includes: medical and statistical data for doctors, and disease data and prevention recommendations for patients.

9. A device for generating a secondary prevention model for stroke, characterized in that, The device for generating a secondary prevention model for stroke includes: The dataset building module is configured to process clinical data on stroke to build a training dataset; The fine-tuning module is configured to fine-tune the first general large model using the training dataset to obtain a secondary prevention model for stroke. The reasoning module is configured to use the secondary prevention model for stroke to perform reasoning and generate reasoning results and reasoning basis; The verification and evaluation module is configured to perform clinical verification and evaluation of the secondary stroke prevention model based on the inference results and the inference basis.

10. A secondary prevention model for stroke, characterized in that, The secondary stroke prevention model is generated by the method for generating a secondary stroke prevention model as described in any one of claims 1 to 8.

11. 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 one or more computer programs that can be executed by the at least one processor, the one or more computer programs being executed by the at least one processor to enable the at least one processor to perform the stroke secondary prevention model generation method as described in any one of claims 1 to 8.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method for generating a secondary prevention model for stroke as described in any one of claims 1 to 8.

Images