Inpatient medical risk analysis method and system and medium

Abstract

The invention relates to an inpatient medical risk analysis method and system and a medium, and relates to the technical field of medical information. In the scheme, first diagnosis and treatment data and second diagnosis and treatment data of an inpatient are acquired; wherein the first diagnosis and treatment data is diagnosis and treatment data obtained by a doctor, and the second diagnosis and treatment data is diagnosis and treatment data obtained by calculating a large model; the first diagnosis and treatment data and the second diagnosis and treatment data are diagnosis and treatment data of the same type, and the large calculation model is a neural network model obtained by combining loss function training; calculating the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data; and judging whether an unconventional medical risk exists or not based on the obtained similarity. And if the similarity is lower, the risk of the unconventional medical treatment is higher, and if the similarity is higher, the risk of the unconventional medical treatment is lower, so that the unconventional medical treatment risk of the inpatient can be analyzed to reduce the unconventional medical treatment risk of the inpatient.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a method, system and medium for analyzing medical risks of inpatients. Background Art

[0002] Unconventional medical treatment includes over-medical treatment of minor illnesses and under-medical treatment of major illnesses. With the acceleration of the aging process, unconventional medical treatment of minor illnesses with major treatments is increasing, which will not only increase the burden on medical insurance, but also increase the burden on patients. How to reduce unconventional medical treatment of minor illnesses with major treatments has become an urgent problem that needs to be solved. Summary of the invention

[0003] The technical problem to be solved by the present application is to provide a method, system and medium for analyzing medical risks of inpatients, which have the characteristics of reducing unconventional medical risks.

[0004] In a first aspect, an embodiment provides a method for analyzing medical risk of an inpatient, comprising: Obtaining first diagnosis and treatment data and second diagnosis and treatment data of an inpatient; wherein the first diagnosis and treatment data is diagnosis and treatment data obtained by a doctor, and the second diagnosis and treatment data is diagnosis and treatment data obtained by calculating a large model; the first diagnosis and treatment data and the second diagnosis and treatment data are diagnosis and treatment data of the same type, and the large calculation model is a neural network model obtained by training in combination with a loss function; Calculate the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data; Based on the obtained similarities, determine whether there is a risk of unconventional medical treatment.

[0005] In one embodiment, the first diagnosis and treatment data includes a first diagnosis result obtained by a doctor based on the patient's condition, and the second diagnosis and treatment data includes a second diagnosis result obtained by a diagnosis result recognition model based on the patient's condition; the diagnosis result recognition model is a neural network model trained in combination with a first loss function to diagnose the input condition characteristics to obtain a diagnosis result; the calculation of the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data includes: Vectorizing the first diagnosis result and the second diagnosis result respectively to obtain a first diagnosis vector and a second diagnosis vector; Calculating a first similarity between the first diagnostic vector and the second diagnostic vector; The determination of whether there is a risk of unconventional medical treatment based on the obtained similarity includes: It is determined whether the first similarity reaches a first similarity threshold; if not, it is considered that there is a risk of unconventional medical treatment.

[0006] In one embodiment, the first diagnosis and treatment data includes a first treatment plan obtained by a doctor based on a diagnosis result, and the second diagnosis and treatment data includes a second treatment plan obtained by a treatment plan acquisition model based on the diagnosis result; the treatment plan acquisition model is a neural network model obtained by combining training with a second loss function to analyze the input diagnosis result to obtain a treatment plan; the calculation of the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data includes: Vectorizing the first treatment plan and the second treatment plan respectively to obtain a first treatment vector and a second treatment vector; calculating a second similarity between the first treatment vector and the second treatment vector; The determination of whether there is a risk of unconventional medical treatment based on the obtained similarity includes: It is determined whether the second similarity reaches a second similarity threshold; if not, it is considered that there is a risk of unconventional medical treatment.

[0007] In one embodiment, the second diagnosis and treatment data includes a second treatment plan obtained by the treatment plan acquisition model based on the diagnosis result, including: The second diagnosis and treatment data includes a second treatment plan obtained by the treatment plan acquisition model based on the doctor's diagnosis result.

[0008] In one embodiment, the first diagnosis and treatment data includes a first diagnosis result obtained by a doctor based on the patient's condition and a first treatment plan obtained by the doctor based on the first diagnosis result, and the second diagnosis and treatment data includes a second diagnosis result obtained by a diagnosis result recognition large model based on the patient's condition and a second treatment plan obtained by a treatment plan acquisition large model based on the second diagnosis result; the diagnosis result recognition large model is a neural network model trained in combination with a first loss function to diagnose the input condition characteristics to obtain a diagnosis result; the treatment plan acquisition large model is a neural network model trained in combination with a second loss function to analyze the input diagnosis result to obtain a treatment plan; the calculation of the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data includes: Vectorizing the first diagnosis result and the second diagnosis result respectively to obtain a first diagnosis vector and a second diagnosis vector; Calculating a first similarity between the first diagnostic vector and the second diagnostic vector; Vectorizing the first treatment plan and the second treatment plan respectively to obtain a first treatment vector and a second treatment vector; calculating a second similarity between the first treatment vector and the second treatment vector; The determination of whether there is a risk of unconventional medical treatment based on the obtained similarity includes: Determine whether any one of the first similarity and the second similarity fails to reach the corresponding similarity threshold, and if so, it is considered that there is a risk of unconventional medical treatment; or, determine whether both the first similarity and the second similarity fail to reach the corresponding similarity threshold, and if so, it is considered that there is a risk of unconventional medical treatment.

[0009] In one embodiment, the training method for obtaining a large model of treatment plans includes: Collecting first medical data of multiple patients and annotating the first medical data to clarify the diagnosis result; the first medical data includes the basic information, chief complaint information, clinical manifestations and test data of the patients; the diagnosis result includes the disease type, severity of the disease and cause analysis; Preprocess the labeled data and convert it into a format suitable for the Hua Tuo model input to obtain the data set; Dividing the data set into a training set, a validation set, and a test set; Based on the preset training parameters, the training set is used as the input of the Hua Tuo model for training. The performance of the Hua Tuo model is evaluated and the hyperparameters are adjusted based on the validation set during the training process. The generalization ability of the Hua Tuo model is evaluated based on the test set, thereby obtaining a trained and improved Hua Tuo model. The trained improved Hua Tuo model is used as the diagnosis result identification model.

[0010] In one embodiment, the training method for obtaining a large model of treatment plans includes: Collecting second medical data of multiple patients and annotating the second medical data to clarify the treatment plan; the second medical data includes the patient's diagnosis result; the diagnosis result includes the disease type, severity of the disease and cause analysis; Preprocess the labeled data and convert it into a format suitable for the Hua Tuo model input to obtain the data set; Dividing the data set into a training set, a validation set, and a test set; Based on the preset training parameters, the training set is used as the input of the Hua Tuo model for training. The performance of the Hua Tuo model is evaluated and the hyperparameters are adjusted based on the validation set during the training process. The generalization ability of the Hua Tuo model is evaluated based on the test set, thereby obtaining a trained and improved Hua Tuo model. The trained improved Hua Tuo large model is used as the treatment plan to obtain the large model.

[0011] In a second aspect, an embodiment provides a medical risk analysis system for inpatients, which is used to implement the medical risk analysis method for inpatients described in any one of the above embodiments, and the system includes a second diagnosis and treatment data acquisition module, a similarity calculation module and a risk judgment module; The second diagnosis and treatment data acquisition module is configured to obtain the second diagnosis and treatment data based on the calculation of the large model; the similarity calculation module is configured to obtain the first diagnosis and treatment data and the second diagnosis and treatment data and calculate the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data; the risk judgment module is configured to judge whether there is a risk of unconventional medical treatment based on the obtained similarity.

[0012] In one embodiment, the second diagnosis and treatment data acquisition module includes a second diagnosis result acquisition unit, and the similarity calculation module includes a first similarity calculation unit; and / or, the second diagnosis and treatment data acquisition module includes a second treatment plan acquisition unit, and the similarity calculation module includes a second similarity calculation unit; The second diagnosis result acquisition unit is configured to obtain a second diagnosis result based on the diagnosis result identification macro model, and the second treatment plan acquisition unit is configured to obtain a second treatment plan based on the treatment plan acquisition macro model; The first similarity calculation unit is configured to obtain a first diagnosis result and a second diagnosis result and calculate the similarity between the first diagnosis result and the second diagnosis result; the second similarity calculation unit is configured to obtain a first treatment plan and a second treatment plan and calculate the similarity between the first treatment plan and the second treatment plan.

[0013] In a third aspect, an embodiment provides a computer-readable storage medium, in which a program is stored, and the program can be loaded by a processor to execute the medical risk analysis method for inpatients described in any one of the above embodiments.

[0014] Since the judgment is based on the similarity between the first diagnosis and treatment data given by the doctor and the second diagnosis and treatment data given by the large computing model, the lower the similarity, the higher the risk of unconventional medical treatment, and the higher the similarity, the lower the risk of unconventional medical treatment. In this way, the unconventional medical risks of hospitalized patients can be analyzed to reduce the unconventional medical risks of hospitalized patients. Based on this, the unconventional medical risks of minor illnesses can be further identified. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a schematic diagram of the structure of a medical risk analysis system for inpatients according to an embodiment of the present application; Figure 2 This is a flow chart of a medical risk analysis method for inpatients according to an embodiment of the present application; Figure 3 This is a flowchart of a large model training method for diagnosis result recognition according to an embodiment of the present application; Figure 4 It is a flowchart of a large model training method for obtaining a treatment plan according to an embodiment of the present application.

[0016] In the diagram: 01, second diagnosis and treatment data acquisition module, 02, similarity calculation module, 03, risk judgment module, 0101, second diagnosis result acquisition unit, 0102, second treatment plan acquisition unit, 0201, first similarity calculation unit, 0202, second similarity calculation unit. DETAILED DESCRIPTION

[0017] The present invention is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are for making the present application better understood. However, those skilled in the art can easily recognize that some features can be omitted in different situations, or can be replaced by other elements, materials, methods. In some cases, some operations related to the present application are not shown or described in the specification, this is to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations according to the description in the specification and the general technical knowledge in the art.

[0018] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.

[0019] The serial numbers assigned to the components in this article, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning.

[0020] In view of the current unconventional medical problems of treating minor illnesses with major treatments, a method, system and medium for analyzing medical risks of hospitalized patients are provided in the embodiments of the present application. In the scheme, first medical data provided by a doctor and second medical data provided by a large calculation model are obtained, the similarity between the first medical data and the second medical data is calculated, and whether there is a risk of unconventional medical treatment is judged based on the obtained similarity. Since the judgment is based on the similarity between the first medical data provided by the doctor and the second medical data provided by the large calculation model, the lower the similarity, the higher the risk of unconventional medical treatment, and the higher the similarity, the lower the risk of unconventional medical treatment. In this way, the unconventional medical risks of hospitalized patients can be analyzed to reduce the unconventional medical risks of hospitalized patients. Based on this, the unconventional medical risks of treating minor illnesses with major treatments can also be further identified to reduce the unconventional medical risks of hospitalized patients with treating minor illnesses with major treatments.

[0021] In order to more clearly explain the present application scheme, the medical risk analysis system for inpatients will be first explained below.

[0022] In one embodiment of the present application, a medical risk analysis system for inpatients is provided. Figure 1 The system includes a second diagnosis and treatment data acquisition module 01, a similarity calculation module 02 and a risk judgment module 03.

[0023] The second diagnosis and treatment data acquisition module 01 is configured to obtain the second diagnosis and treatment data based on the calculation of the large model, wherein the calculation of the large model is a neural network model obtained by training in combination with the loss function.

[0024] In some embodiments, the second diagnosis and treatment data acquisition module 01 includes a second diagnosis result acquisition unit 0101 and / or a second treatment plan acquisition unit 0102. The second diagnosis result acquisition unit 0101 is configured to obtain a second diagnosis result based on a diagnosis result recognition large model, and the second treatment plan acquisition unit 0102 is configured to obtain a second treatment plan based on a treatment plan acquisition large model. The diagnosis result recognition large model is a neural network model trained in combination with the first loss function, which diagnoses the input disease characteristics to obtain a diagnosis result; the treatment plan acquisition large model is a neural network model trained in combination with the second loss function, which analyzes the input diagnosis result to obtain a treatment plan.

[0025] The similarity calculation module 02 is configured to obtain the first diagnosis and treatment data and the second diagnosis and treatment data and calculate the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data. The first diagnosis and treatment data is the diagnosis and treatment data obtained by the doctor, and the first diagnosis and treatment data and the second diagnosis and treatment data are the same type of diagnosis and treatment data.

[0026] In some embodiments, the similarity calculation module 02 includes a first similarity calculation unit 0201 and / or a second similarity calculation unit 0202. The first similarity calculation unit 0201 is configured to obtain a first diagnosis result and a second diagnosis result and calculate the similarity between the first diagnosis result and the second diagnosis result; the second similarity calculation unit 0202 is configured to obtain a first treatment plan and a second treatment plan and calculate the similarity between the first treatment plan and the second treatment plan. The first diagnosis result is the diagnosis result obtained by the doctor based on the patient's condition, and the first treatment plan is the treatment plan obtained by the doctor based on the diagnosis result.

[0027] The risk judgment module 03 is configured to judge whether there is a risk of unconventional medical treatment based on the obtained similarity.

[0028] in, Figure 1In the embodiment, the second diagnosis and treatment data acquisition module 01 includes a second diagnosis result acquisition unit 0101 and a second treatment plan acquisition unit 0102, and the similarity calculation module 02 includes a first similarity calculation unit 0201 and a second similarity calculation unit 0202. Then, for the risk judgment module 03, it can be determined whether there is a risk of unconventional medical treatment based only on the first similarity, or it can be determined whether there is a risk of unconventional medical treatment based only on the second similarity, or it can be determined whether there is a risk of unconventional medical treatment by combining the first similarity and the second similarity.

[0029] The medical risk analysis system for inpatients based on any of the above-mentioned embodiments makes a judgment based on the similarity between the first diagnosis and treatment data provided by the doctor and the second diagnosis and treatment data provided by the large computing model. Therefore, if the similarity is lower, the risk of unconventional medical treatment is higher, and if the similarity is higher, the risk of unconventional medical treatment is lower. In this way, the medical risks of inpatients can be analyzed to reduce the unconventional medical risks of inpatients. Based on this, the unconventional medical risks of treating minor illnesses with major treatments can be further identified to reduce the unconventional medical risks of treating minor illnesses with major treatments of inpatients.

[0030] The present application provides a method for analyzing medical risk of hospitalized patients, which can be implemented based on the analysis system of any of the above embodiments. Figure 2 , the analysis methods include: Step S10, obtaining the first diagnosis and treatment data and the second diagnosis and treatment data of the inpatient, wherein the first diagnosis and treatment data is the diagnosis and treatment data obtained by the doctor, and the second diagnosis and treatment data is the diagnosis and treatment data obtained by calculating the large model; the first diagnosis and treatment data and the second diagnosis and treatment data are the same type of diagnosis and treatment data, and the large calculation model is a neural network model obtained by training with a loss function.

[0031] The resident doctor will infer, exclude and judge the condition of the inpatient and get a diagnosis result. However, the applicant found in the study that if there is a non-routine medical treatment, there is a high probability that the diagnosis result obtained by the doctor will be significantly different from the actual diagnosis result.

[0032] In view of this, in one embodiment, the first diagnosis and treatment data includes a first diagnosis result obtained by a doctor based on the patient's condition, and the second diagnosis and treatment data includes a second diagnosis result obtained by a diagnosis result recognition model based on the patient's condition. The diagnosis result recognition model is a neural network model trained in combination with the first loss function, which diagnoses the input condition characteristics to obtain a diagnosis result.

[0033] The large model for diagnosis result recognition can be obtained by training based on some neural network training methods in the prior art. A new neural network training method is provided in one embodiment of the present application. Please refer to Figure 3,include: Step S1011, collect the first medical data of multiple patients and annotate the first medical data to clarify the diagnosis result. The first medical data includes the basic information, chief complaint information, clinical manifestations and test data of the patient; the diagnosis result includes the disease type, severity of the disease and cause analysis.

[0034] Collect a large amount of first-line medical data containing disease descriptions and corresponding accurate diagnosis results. These data cover a variety of diseases, different symptoms, etiology analysis, and disease severity to ensure data diversity and representativeness. Label the data so that the disease description corresponds to the diagnosis result one by one, forming a data set that can be used for model training.

[0035] Among them, basic personal information (including gender, age and other basic information), health-related information (including medical history, family medical history, allergy history and other information) and disease-related information (including onset, treatment, etc.). The patient's chief complaint information includes the patient's and / or family's statement on the disease and the course of the disease. Test results may include the test results of various indicators of the patient's body.

[0036] In one embodiment, the corresponding diagnosis result recognition big model can be trained based on the diagnosis and treatment department. For example, the diagnosis result recognition big model for neurosurgery can be obtained. In this way, the first medical data of neurosurgery patients can be collected and the first medical data can be annotated to clarify the diagnosis result.

[0037] Step S1012, pre-process the labeled data and convert it into a format suitable for input of the Hua Tuo large model to obtain a data set.

[0038] In one embodiment, preprocessing includes cleaning the collected data, removing erroneous data, duplicate data, and interference information irrelevant to the condition and diagnosis. At the same time, the condition description and diagnosis results are standardized, such as unifying the text format and encoding method, so that the data meets the requirements of model input and output. Data enhancement can also be performed as needed, such as synonym replacement and sentence rearrangement of the condition description to increase the diversity of the data, but attention should be paid to maintaining semantic consistency.

[0039] The Hua Tuo model, also known as Hua Tuo GPT, is an open source medical model of the existing ChatGPT. It has good natural language understanding of medical information. Patients can interact with Hua Tuo GPT in a conversational manner to achieve department triage, auxiliary diagnosis and other functions, providing assistance to doctors. In the embodiment of the present application, based on the Hua Tuo model, fine-tuning the Hua Tuo model can obtain a better diagnosis result recognition model.

[0040] Step S1013, dividing the data set into a training set, a validation set and a test set.

[0041] The training set is used to train the model, the validation set is used to evaluate the performance of the model and adjust the hyperparameters during the training process, and the test set is used to finally evaluate the generalization ability of the model.

[0042] Step S1014, based on the preset training parameters, the training set is used as the input of the Hua Tuo model for training, the performance of the Hua Tuo model is evaluated and the hyperparameters are adjusted based on the validation set during the training process, and the generalization ability of the Hua Tuo model is evaluated based on the test set, thereby obtaining a trained improved Hua Tuo model.

[0043] You can set appropriate training parameters such as learning rate, batch size, and number of training rounds according to the model requirements and computing resources. The learning rate determines the step size of the model parameter update, the batch size affects the efficiency and stability of the model training, and the number of training rounds determines the number of iterations of the model training.

[0044] During the training process, the first loss function of the model is calculated based on the output of the model and the annotated true results, and the parameters of the model are updated using an optimization algorithm to reduce the value of the first loss function. In one embodiment, during the training process, the weight file of the model can be saved periodically so that the training can be resumed when the training is interrupted or a problem occurs.

[0045] In one embodiment, the first loss function is composed of a cross entropy loss function and a mean square error loss function. For multi-classification problems of disease type and etiology analysis, the cross entropy loss function is used, and for regression problems of disease severity analysis, the mean square error loss function is used.

[0046] In one embodiment, the trained model is evaluated using a validation set, and the accuracy, recall and other indicators of the model on the validation set are calculated to evaluate the performance of the model for entity extraction. Based on the evaluation results, the problems and deficiencies of the model are analyzed.

[0047] In one embodiment, the model can be optimized based on the results of the model evaluation. The training parameters can be adjusted, the amount of training data can be increased, the quality of data annotation can be improved, etc., to improve the performance of the model. If the model has overfitting or underfitting problems, regularization techniques, increasing data diversity, etc. can be used to solve them.

[0048] Step S1015, using the trained improved Hua Tuo large model as the diagnosis result recognition large model.

[0049] In one embodiment, the resident doctor will infer, exclude and judge based on the diagnosis of the inpatient to obtain a treatment plan. However, the applicant found in the study that if there is a situation of unconventional medical treatment, there is a high probability that the treatment plan obtained by the doctor will be significantly different from the appropriate treatment plan. For example, under the same diagnosis, the second treatment plan is a treatment plan based on drug treatment, and the first treatment plan is a treatment plan based on surgical treatment, then there is a great risk of unconventional medical treatment.

[0050] In view of this, in one embodiment, the first diagnosis and treatment data includes a first treatment plan obtained by the doctor based on the diagnosis result, and the second diagnosis and treatment data includes a second treatment plan obtained by the treatment plan acquisition model based on the diagnosis result. The treatment plan acquisition model is a neural network model obtained by combining the second loss function training to analyze the input diagnosis result to obtain the treatment plan.

[0051] Thus, the first diagnosis and treatment data may include only the first diagnosis result obtained by the doctor based on the patient's condition, or only the first treatment plan obtained by the doctor based on the diagnosis result, or both the first diagnosis result and the first treatment plan. In the case where the first diagnosis and treatment data only include the first diagnosis result, the second diagnosis and treatment data only include the second diagnosis result. In the case where the first diagnosis and treatment data only include the first treatment plan, the second diagnosis and treatment data only include the second treatment plan. In the case where the first diagnosis and treatment data include both the first diagnosis result and the first treatment plan, the second diagnosis and treatment data include both the second diagnosis result and the second treatment plan.

[0052] In one embodiment, when the first diagnosis and treatment data only includes the first treatment plan and the second diagnosis and treatment data only includes the second treatment plan, the treatment plan acquisition model obtains the second treatment plan based on the doctor's diagnosis result.

[0053] In one embodiment, when the first diagnosis and treatment data includes both the first diagnosis result and the first treatment plan, and the second diagnosis and treatment data includes both the second diagnosis result and the second treatment plan, the treatment plan acquisition model obtains the second treatment plan based on the second diagnosis result, thereby obtaining a more accurate second treatment plan.

[0054] The treatment plan acquisition model can be obtained by training based on some neural network training methods in the prior art. In one embodiment of the present application, a new training method for obtaining a treatment plan model is provided. Please refer to Figure 4 , which may include: Step S1021, collect second medical data of multiple patients and annotate the second medical data to clarify the treatment plan. The second medical data includes the patient's diagnosis result, and the diagnosis result includes the disease type, severity of the disease and cause analysis.

[0055] In one embodiment, a large model for obtaining a corresponding treatment plan can be trained based on the diagnosis and treatment department. For example, a large model for obtaining a treatment plan for neurosurgery can be obtained. In this way, the second medical data of neurosurgery patients can be collected and the second medical data can be annotated to clarify the treatment plan. For example, for patients with cerebral hemorrhage, the treatment plan comprehensively derived based on the bleeding site, bleeding amount and etiology analysis in the diagnosis results can include general treatment (such as bed rest, monitoring vital signs), blood pressure control (rational use of antihypertensive drugs), reducing intracranial pressure, hemostasis treatment (such as drug hemostasis), surgical treatment and rehabilitation treatment.

[0056] Step S1022, pre-process the labeled data and convert it into a format suitable for input of the Hua Tuo large model to obtain a data set.

[0057] In the embodiments of the present application, based on the Hua Tuo model, fine-tuning the Hua Tuo model can obtain a better treatment plan to obtain a large model.

[0058] Step S1023, dividing the data set into a training set, a validation set and a test set.

[0059] Step S1024, based on the preset training parameters, the training set is used as the input of the Hua Tuo model for training, the performance of the Hua Tuo model is evaluated and the hyperparameters are adjusted based on the validation set during the training process, and the generalization ability of the Hua Tuo model is evaluated based on the test set, thereby obtaining a trained improved Hua Tuo model.

[0060] Step S1025, using the trained improved Hua Tuo large model as a treatment plan to obtain a large model.

[0061] Step S20, calculating the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data.

[0062] In one embodiment, the first diagnosis and treatment data only includes the first diagnosis result obtained by the doctor based on the patient's condition, and the second diagnosis and treatment data only includes the second diagnosis result. Then, calculating the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data includes: Step S2011, vectorizing the first diagnostic result and the second diagnostic result respectively to obtain a first diagnostic vector and a second diagnostic vector.

[0063] First, we can construct a vector space based on one-hot encoding, with each disease corresponding to one dimension. For example, if the diagnosis is a cold, the vector is represented as [1,0,0]; if it is pneumonia, it is [0,1,0]; and bronchitis is [0,0,1]. The advantage of this method is that it is simple and intuitive, and can clearly represent different categories. However, when there are many categories, the vector dimension will become very large.

[0064] For the diagnosis results in text form, in one embodiment, the bag-of-words model can be used to vectorize the diagnosis results, such as "the patient has symptoms of cough and fever, and the preliminary diagnosis is upper respiratory tract infection". First, a vocabulary is constructed, which contains words related to the medical field, such as "cough", "fever", "infection", etc. Then the number of occurrences of each word in the diagnosis text is counted, and these counts are combined into a vector. For example, the vocabulary has three words "cough", "fever", and "infection", and in a diagnosis text, "cough" appears 2 times, "fever" appears 1 time, and "infection" appears 1 time, then the vector can be expressed as [2,1,1]. This method ignores the order of words, but can capture the main information in the diagnosis results.

[0065] Step S2012: Calculate a first similarity between the first diagnostic vector and the second diagnostic vector.

[0066] Based on the obtained vectors, cosine similarity may be used to calculate the similarity between two vectors. In one embodiment, a method for calculating the similarity between two vectors (a first vector A and a second vector B) may include: Step S100, respectively calculating the modulus of the first vector and the modulus of the second vector.

[0067] In one embodiment, step S100 can be expressed as: , , in, represents the magnitude of the first vector, represents the magnitude of the second vector, Indicates the first vector element values, represents the first element values, Represents the total number of elements in the first or second vector, .

[0068] Those skilled in the art can understand that in order to compare the similarity between the first vector and the second vector, the first vector and the second vector have the same form and the same number of elements.

[0069] Step S200: Calculate the product of the modulus of the first vector and the modulus of the second vector as a first product.

[0070] Step S300: Calculate the product of the first vector and the second vector as a second product.

[0071] Step S400, calculating the ratio of the second product to the first product, the closer the ratio is to 1, the higher the similarity is.

[0072] In one embodiment, step S400 can be expressed as: , in, Represents the similarity value.

[0073] For the obtained similarity value, when the value is 1, it is considered that the two vectors are exactly the same, when the value is -1, it is considered that the two vectors are completely opposite, and when the value is 0, it means that the two vectors are orthogonal. The closer to 1, the more similar the two disease names are.

[0074] In one embodiment, the first diagnosis and treatment data only includes the first treatment plan, and the second diagnosis and treatment data only includes the second treatment plan, and calculating the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data includes: Step S2021, vectorizing the first treatment plan and the second treatment plan respectively to obtain a first treatment vector and a second treatment vector.

[0075] Those skilled in the art will appreciate that the vectorization of the first treatment plan and the second treatment plan may adopt a vectorization method similar to the vectorization of the first diagnosis result and the second diagnosis result, which will not be described in detail here.

[0076] Step S2022, calculating the second similarity between the first treatment vector and the second treatment vector.

[0077] Those skilled in the art can understand that the method for calculating the second similarity can be the same as the method for calculating the first similarity, which will not be described in detail here.

[0078] In one embodiment, the first diagnosis and treatment data includes a first diagnosis result and a first treatment plan, and the second diagnosis and treatment data includes a second diagnosis result and a second treatment plan, and calculating the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data includes: Step S2031 , vectorizing the first diagnosis result and the second diagnosis result respectively to obtain a first diagnosis vector and a second diagnosis vector.

[0079] Step S2032: Calculate a first similarity between the first diagnostic vector and the second diagnostic vector.

[0080] Step S2033: vectorize the first treatment plan and the second treatment plan respectively to obtain a first treatment vector and a second treatment vector.

[0081] Step S2034, calculating the second similarity between the first treatment vector and the second treatment vector.

[0082] Step S30, judging whether there is a risk of unconventional medical treatment based on the obtained similarity.

[0083] In one embodiment, judging whether there is a risk of unconventional medical treatment based on the first similarity includes: judging whether the first similarity reaches a first similarity threshold, and if not, it is considered that there is a risk of unconventional medical treatment.

[0084] In one embodiment, judging whether there is a risk of unconventional medical treatment based on the second similarity includes: judging whether the second similarity reaches a first similarity threshold, and if not, it is considered that there is a risk of unconventional medical treatment.

[0085] In some embodiments, the first similarity threshold and the second similarity threshold may be the same or different. For example, the first similarity threshold and the second similarity threshold may both be set to 0.92, or one may be set to 0.92 and the other may be set to 0.93. Those skilled in the art may make settings based on demand.

[0086] In some embodiments, whether there is a risk of unconventional medical treatment is determined based on the first similarity and the second similarity. It can be determined whether any one of the first similarity and the second similarity does not reach the corresponding similarity threshold, and if so, it is considered that there is a risk of unconventional medical treatment; it can also be determined whether both the first similarity and the second similarity do not reach the corresponding similarity threshold, and if so, it is considered that there is a risk of unconventional medical treatment. Based on different situations, those skilled in the art can set different first similarity thresholds and / or second similarity thresholds.

[0087] Based on the medical risk analysis method for hospitalized patients in any of the above-mentioned embodiments, since the judgment is based on the similarity between the first medical data provided by the doctor and the second medical data provided by the large computing model, if the similarity is lower, the risk of unconventional medical treatment is higher, and if the similarity is higher, the risk of unconventional medical treatment is lower. In this way, the unconventional medical risks of hospitalized patients can be analyzed to reduce the unconventional medical risks of hospitalized patients. Based on this, the unconventional medical risks of treating minor illnesses with major treatments can also be further identified to reduce the unconventional medical risks of treating minor illnesses with major treatments of hospitalized patients.

[0088] In one embodiment of the present application, a computer-readable storage medium is provided, on which a program is stored. The stored program includes a method that can be loaded by a processor and process any of the above embodiments.

[0089] Those skilled in the art will appreciate that all or part of the functions of the various methods in the above-mentioned embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to implement the above-mentioned functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the above-mentioned functions can be implemented. In addition, when all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can also be stored in a storage medium such as a server, another computer, disk, optical disk, flash disk or mobile hard disk, and can be downloaded or copied and saved in the memory of the local device, or the system of the local device is updated, and when the program in the memory is executed by the processor, all or part of the functions in the above-mentioned embodiments can be implemented.

[0090] The above specific examples are used to illustrate the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the art, according to the concept of the present invention, some simple deductions, modifications or substitutions can be made.

Claims

1. A method for analyzing medical risk of hospitalized patients, characterized in that: include: Obtaining first diagnosis and treatment data and second diagnosis and treatment data of an inpatient; wherein the first diagnosis and treatment data is diagnosis and treatment data obtained by a doctor, and the second diagnosis and treatment data is diagnosis and treatment data obtained by calculating a large model; the first diagnosis and treatment data and the second diagnosis and treatment data are diagnosis and treatment data of the same type, and the large calculation model is a neural network model obtained by training in combination with a loss function; Calculate the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data; Based on the obtained similarities, determine whether there is a risk of unconventional medical treatment.

2. The medical risk analysis method for hospitalized patients according to claim 1, characterized in that: The first diagnosis and treatment data includes a first diagnosis result obtained by a doctor based on the patient's condition, and the second diagnosis and treatment data includes a second diagnosis result obtained by a diagnosis result recognition model based on the patient's condition; the diagnosis result recognition model is a neural network model trained in combination with the first loss function, which diagnoses the input condition characteristics to obtain a diagnosis result; The calculating the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data comprises: Vectorizing the first diagnosis result and the second diagnosis result respectively to obtain a first diagnosis vector and a second diagnosis vector; Calculating a first similarity between the first diagnostic vector and the second diagnostic vector; The determination of whether there is a risk of unconventional medical treatment based on the obtained similarity includes: It is determined whether the first similarity reaches a first similarity threshold; if not, it is considered that there is a risk of unconventional medical treatment.

3. The medical risk analysis method for hospitalized patients according to claim 1, characterized in that: The first diagnosis and treatment data includes a first treatment plan obtained by the doctor based on the diagnosis result, and the second diagnosis and treatment data includes a second treatment plan obtained by the treatment plan acquisition model based on the diagnosis result; the treatment plan acquisition model is a neural network model obtained by combining training with a second loss function to analyze the input diagnosis result to obtain the treatment plan; The calculating the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data comprises: Vectorizing the first treatment plan and the second treatment plan respectively to obtain a first treatment vector and a second treatment vector; calculating a second similarity between the first treatment vector and the second treatment vector; The determination of whether there is a risk of unconventional medical treatment based on the obtained similarity includes: It is determined whether the second similarity reaches a second similarity threshold; if not, it is considered that there is a risk of unconventional medical treatment.

4. The medical risk analysis method for hospitalized patients according to claim 3, characterized in that: The second diagnosis and treatment data includes a second treatment plan obtained by the treatment plan acquisition model based on the diagnosis result, including: The second diagnosis and treatment data includes a second treatment plan obtained by the treatment plan acquisition model based on the doctor's diagnosis result.

5. The medical risk analysis method for hospitalized patients according to claim 1, characterized in that: The first diagnosis and treatment data includes a first diagnosis result obtained by a doctor based on the patient's condition and a first treatment plan obtained by the doctor based on the first diagnosis result, and the second diagnosis and treatment data includes a second diagnosis result obtained by a diagnosis result recognition large model based on the patient's condition and a second treatment plan obtained by a treatment plan acquisition large model based on the second diagnosis result; the diagnosis result recognition large model is a neural network model trained in combination with a first loss function, which diagnoses the input condition characteristics to obtain a diagnosis result; the treatment plan acquisition large model is a neural network model trained in combination with a second loss function, which analyzes the input diagnosis result to obtain a treatment plan; The calculating the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data comprises: Vectorizing the first diagnosis result and the second diagnosis result respectively to obtain a first diagnosis vector and a second diagnosis vector; Calculating a first similarity between the first diagnostic vector and the second diagnostic vector; Vectorizing the first treatment plan and the second treatment plan respectively to obtain a first treatment vector and a second treatment vector; calculating a second similarity between the first treatment vector and the second treatment vector; The determination of whether there is a risk of unconventional medical treatment based on the obtained similarity includes: Determine whether any one of the first similarity and the second similarity fails to reach the corresponding similarity threshold, and if so, it is considered that there is a risk of unconventional medical treatment; or, determine whether both the first similarity and the second similarity fail to reach the corresponding similarity threshold, and if so, it is considered that there is a risk of unconventional medical treatment.

6. The medical risk analysis method for hospitalized patients according to claim 3 or 5, characterized in that: The training method for obtaining a large model for the treatment plan includes: Collecting first medical data of multiple patients and annotating the first medical data to clarify the diagnosis result; the first medical data includes the basic information, chief complaint information, clinical manifestations and test data of the patients; the diagnosis result includes the disease type, severity of the disease and cause analysis; Preprocess the labeled data and convert it into a format suitable for the Hua Tuo model input to obtain the data set; Dividing the data set into a training set, a validation set, and a test set; Based on the preset training parameters, the training set is used as the input of the Hua Tuo model for training. The performance of the Hua Tuo model is evaluated and the hyperparameters are adjusted based on the validation set during the training process. The generalization ability of the Hua Tuo model is evaluated based on the test set, thereby obtaining a trained and improved Hua Tuo model. The trained improved Hua Tuo model is used as the diagnosis result identification model.

7. The medical risk analysis method for hospitalized patients according to claim 3 or 5, characterized in that: The training method for obtaining a large model for the treatment plan includes: Collecting second medical data of multiple patients and annotating the second medical data to clarify the treatment plan; the second medical data includes the patient's diagnosis result; the diagnosis result includes the disease type, severity of the disease and cause analysis; Preprocess the labeled data and convert it into a format suitable for the Hua Tuo model input to obtain the data set; Dividing the data set into a training set, a validation set, and a test set; Based on the preset training parameters, the training set is used as the input of the Hua Tuo model for training. The performance of the Hua Tuo model is evaluated and the hyperparameters are adjusted based on the validation set during the training process. The generalization ability of the Hua Tuo model is evaluated based on the test set, thereby obtaining a trained and improved Hua Tuo model. The trained improved Hua Tuo large model is used as the treatment plan to obtain the large model.

8. A medical risk analysis system for inpatients, characterized in that: Used to implement the medical risk analysis method for inpatients as described in any one of claims 1 to 7, the system comprises a second diagnosis and treatment data acquisition module, a similarity calculation module and a risk judgment module; The second diagnosis and treatment data acquisition module is configured to obtain the second diagnosis and treatment data based on the calculation of the large model; the similarity calculation module is configured to obtain the first diagnosis and treatment data and the second diagnosis and treatment data and calculate the similarity between the first diagnosis and treatment data and the second diagnosis and treatment data; The risk judgment module is configured to judge whether there is a risk of unconventional medical treatment based on the obtained similarity.

9. The inpatient medical risk analysis system according to claim 8, characterized in that: The second diagnosis and treatment data acquisition module includes a second diagnosis result acquisition unit, and the similarity calculation module includes a first similarity calculation unit; and / or, the second diagnosis and treatment data acquisition module includes a second treatment plan acquisition unit, and the similarity calculation module includes a second similarity calculation unit; The second diagnosis result acquisition unit is configured to obtain a second diagnosis result based on the diagnosis result identification macro model, and the second treatment plan acquisition unit is configured to obtain a second treatment plan based on the treatment plan acquisition macro model; The first similarity calculation unit is configured to obtain the first diagnosis result and the second diagnosis result and calculate the similarity between the first diagnosis result and the second diagnosis result; The second similarity calculation unit is configured to obtain the first treatment plan and the second treatment plan and calculate the similarity between the first treatment plan and the second treatment plan.

10. A computer-readable storage medium, characterized in that: The medium stores a program, which can be loaded by a processor and execute the medical risk analysis method for hospitalized patients as described in any one of claims 1 to 7.

Images