Method and device for determining blood glucose management strategy and electronic equipment

Abstract

The invention discloses a blood glucose management strategy determination method and device and electronic equipment. The method comprises the following steps: acquiring physical state data of a target object in a target time period; a fine-grained blood glucose management model is adopted to process the physical state data to obtain a target blood glucose management strategy, the fine-grained blood glucose management model is obtained through training of a first blood glucose management strategy set determined by a coarse-grained blood glucose management model, and the precision of the fine-grained blood glucose management model is higher than that of the coarse-grained blood glucose management model; the coarse-grained blood glucose management model is obtained by training historical physical state data of the target object in a first time period, the first blood glucose management strategy set is determined according to historical physical state data of the target object in a second time period, and the end time of the first time period is before the start time of the second time period. The technical problem that a blood glucose management strategy in the related technology cannot perform personalized management strategy recommendation on a single patient is solved.

Description

Technical Field

[0001] The present application relates to the medical field, and more specifically, to a method, device and electronic device for determining a blood sugar management strategy. Background Art

[0002] Perioperative blood sugar abnormalities mainly include hyperglycemia and hypoglycemia. It is estimated that the incidence of perioperative hyperglycemia in patients with heart disease is as high as 60-90%, which can easily lead to organ dysfunction, abnormal body metabolism, etc., induce a variety of complications, and increase the risk of postoperative infection and death. In addition, perioperative hypoglycemia is also a risk factor for cardiovascular disease and death. Therefore, it is of great significance to formulate a practical strategy for the management of abnormal perioperative blood sugar. However, due to the presence of dynamic factors such as surgical trauma and stress, infection, heart and kidney dysfunction, and dietary changes, the patient's blood sugar data is unstable, so blood sugar management is still a difficult point.

[0003] The insulin sensitivity of patients with perioperative dysglycemia varies greatly from person to person, requiring collaborative management by multiple disciplines such as surgery, endocrinology, and nutrition. However, the blood sugar management strategies specified in clinical guidelines are often based on the average treatment effect of the patient population, which does not fully consider the individual differences and heterogeneity of patients. Therefore, such blood sugar management strategies cannot be regarded as the gold standard. The blood sugar management strategies provided by existing technologies have the problem of being unable to recommend personalized management strategies for individual patients.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present application provide a method, device and electronic device for determining a blood glucose management strategy, so as to at least solve the technical problem in the related art that the blood glucose management strategy cannot recommend a personalized management strategy for a single patient.

[0006] According to one aspect of an embodiment of the present application, a method for determining a blood glucose management strategy is provided, including: obtaining physical state data of a target object in a target time period; processing the physical state data using a fine-grained blood glucose management model to obtain a target blood glucose management strategy, wherein the fine-grained blood glucose management model is trained with a first blood glucose management strategy set determined by a coarse-grained blood glucose management model, the accuracy of the fine-grained blood glucose management model is higher than that of the coarse-grained blood glucose management model, the coarse-grained blood glucose management model is trained with historical physical state data of the target object in a first time period, the first blood glucose management strategy set is determined based on the historical physical state data of the target object in a second time period, and the end time of the first time period is before the start time of the second time period.

[0007] Optionally, the fine-grained blood glucose management model is trained by: obtaining historical basic data of the target object, wherein the historical basic data includes historical physical state data of the first period, historical blood glucose management strategies of the first period, historical physical state data of the second period, reward functions and discharge data, the historical physical state data includes static feature data and time series feature data of the target object, the static feature data includes attribute feature data of the target object, the time series feature data includes historical test data and historical medication data of the target object, wherein the historical test data includes historical blood glucose data; inputting the historical physical state data of the first period into the initial coarse-grained blood glucose Management model, and train with the historical blood sugar management strategies of the first time period as labels to obtain a coarse-grained blood sugar management model; input the historical physical state data of the second time period into the coarse-grained blood sugar management model to obtain a first probability distribution of predicted blood sugar management strategies corresponding to the historical physical state data of the second time period, and after sorting the first probability distribution in order from large to small, the blood sugar management strategies corresponding to the first K probability distributions are determined as a first blood sugar management strategy set, where K is a positive integer; input the historical physical state data of the second time period and the first blood sugar management strategy set into the initial fine-grained blood sugar management model for training to obtain a fine-grained blood sugar management model.

[0008] Optionally, the coarse-grained blood glucose management model includes a first blood glucose management strategy model, and the historical physical state data of the first time period is input into the initial coarse-grained blood glucose management model for training to obtain a coarse-grained blood glucose management model, including: inputting the static feature data in the historical physical state data of the first time period into the deep learning network of the first blood glucose management strategy model to obtain the first feature; inputting the time series feature data in the historical physical state data of the first time period into the neural network model of the first blood glucose management strategy model to obtain the second feature; after splicing the first feature and the second feature, a spliced ​​feature is obtained; the spliced ​​feature is input into the multi-layer perceptron network of the first blood glucose management strategy model to obtain a second probability distribution of the predicted blood glucose management strategy corresponding to the historical physical state data of the first time period; when the cross entropy loss between the second probability distribution and the probability distribution corresponding to the historical blood glucose management strategy of the first time period does not meet the first threshold, adjusting the first parameter of the first blood glucose management strategy model until the cross entropy loss meets the first threshold, stopping the training, and obtaining the coarse-grained blood glucose management model.

[0009] Optionally, the fine-grained blood glucose management model includes a second blood glucose management strategy model, a first network and a second network. The historical physical state data of the second time period and the first blood glucose management strategy set are input into the initial fine-grained blood glucose management model for training to obtain a fine-grained blood glucose management model, including: inputting the historical physical state data of the second time period into the second blood glucose management strategy model to obtain a third probability distribution of predicted blood glucose management strategies corresponding to the historical physical state data of the second time period; determining a candidate set of blood glucose management strategies corresponding to the historical physical state data of the second time period based on the third probability distribution; inputting the historical physical state data of the second time period and the candidate set of blood glucose management strategies into the first network to obtain a fourth probability distribution of predicted blood glucose management strategies corresponding to the historical physical state data of the second time period, and determining the blood glucose management strategy corresponding to the maximum probability of the fourth probability distribution as the recommended blood glucose management strategy; determining a total loss function based on the recommended blood glucose management strategy, and adjusting the parameters of the first network and the second blood glucose management strategy model according to the total loss function, stopping training when the total loss value of the fine-grained blood glucose management model meets the second threshold, and obtaining a fine-grained blood glucose management model, wherein the parameters of the second network are adjusted based on the parameters of the first network.

[0010] Optionally, a candidate set of blood glucose management strategies corresponding to the historical physical state data of the second time period is determined based on a third probability distribution, including: determining the maximum probability in the third probability distribution as the target probability; determining the ratio of each probability in the third probability distribution to the target probability to obtain a ratio set; and determining the blood glucose management strategy corresponding to the probability greater than a third threshold in the ratio set as the candidate set of blood glucose management strategies.

[0011] Optionally, determining the total loss function based on the recommended blood sugar management strategy includes: inputting the historical physical state data of the first time period and the historical blood sugar management strategy of the first time period into the first network to obtain a first result; inputting the historical physical state data of the second time period and the recommended blood sugar management strategy into the second network to obtain a second result; determining the first loss function based on the reward function, the first result and the second result; inputting the historical physical state data of the first time period into the second blood sugar management strategy model to obtain a third result; determining the second loss function based on the historical blood sugar management strategy of the first time period and the third result; determining the total loss function based on the first loss function and the second loss function.

[0012] Optionally, the reward function includes: obtaining the difference between the historical blood sugar data and the preset blood sugar data of the second time period; determining a first reward parameter based on the range of the difference; determining a second reward parameter based on the relationship between the recommended blood sugar management strategy and the first blood sugar management strategy set; determining a reward function based on the first reward parameter and the second reward parameter.

[0013] According to another aspect of an embodiment of the present application, a device for determining a blood glucose management strategy is also provided, including: an acquisition module for acquiring physical state data of a target object in a target time period; a processing module for processing the physical state data using a fine-grained blood glucose management model to obtain a target blood glucose management strategy, wherein the fine-grained blood glucose management model is trained with a first blood glucose management strategy set determined by a coarse-grained blood glucose management model, the accuracy of the fine-grained blood glucose management model is higher than that of the coarse-grained blood glucose management model, the coarse-grained blood glucose management model is trained with the historical physical state data of the target object in the first time period, the first blood glucose management strategy set is determined based on the historical physical state data of the target object in the second time period, and the end time of the first time period is before the start time of the second time period.

[0014] According to another aspect of an embodiment of the present application, there is also provided an electronic device, comprising: a memory and a processor, the memory being used to store program instructions; the processor being connected to the memory and being used to execute a determination method for implementing the above-mentioned blood sugar management strategy.

[0015] According to another aspect of an embodiment of the present application, a non-volatile storage medium is also provided, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned method for determining the blood sugar management strategy by running the computer program.

[0016] According to another aspect of the embodiments of the present application, a computer program product is also provided, including computer instructions, which, when executed by a processor, implement the above-mentioned method for determining the blood sugar management strategy.

[0017] In an embodiment of the present application, the physical state data of the target object in the target time period is obtained; the physical state data is processed by a fine-grained blood glucose management model to obtain a target blood glucose management strategy, wherein the fine-grained blood glucose management model is trained by the first blood glucose management strategy set determined by the coarse-grained blood glucose management model, the accuracy of the fine-grained blood glucose management model is higher than that of the coarse-grained blood glucose management model, the coarse-grained blood glucose management model is trained by the historical physical state data of the target object in the first time period, the first blood glucose management strategy set is determined according to the historical physical state data of the target object in the second time period, and the end time of the first time period is before the start time of the second time period, thereby achieving the purpose of recommending the target blood glucose management strategy to the target object through the fine-grained blood glucose management model, and due to the high accuracy of the fine-grained blood glucose management model, the technical effect of improving the accuracy of blood glucose management strategy recommendations can be achieved when the fine-grained blood glucose management model is used to recommend blood glucose management strategies for individual patients, thereby solving the technical problem in the related art that the blood glucose management strategy cannot recommend personalized management strategies for individual patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 is a flow chart of a method for determining a blood sugar management strategy according to an embodiment of the present application; Figure 2 is a schematic diagram of a blood sugar management model according to an embodiment of the present application; Figure 3 is a structural diagram of a device for determining a blood sugar management strategy according to an embodiment of the present application; Figure 4 It is a hardware structure block diagram of a computer terminal for implementing a method for determining a blood sugar management strategy according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0021] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0022] In the related art, the blood glucose management strategy recommendation method in the medical field can be divided into two types according to the different types of technologies used: supervised learning-based methods and reinforcement learning-based methods. In the supervised learning-based method, a supervised machine learning or deep learning model is constructed based on the doctor's intervention strategy in the electronic medical record, and the doctor's treatment experience is fitted. The shortcoming of this method is that the supervised learning method that fits the treatment experience of clinical doctors is essentially still a blood glucose management recommendation strategy based on clinical guidelines. This type of strategy is based on the average treatment effect of the patient and cannot recommend a personalized blood glucose management strategy for a single patient. The online reinforcement learning-based method is mostly based on simulation data. Recommendations, using online reinforcement learning models such as Policy Gradient, Q-Learning, etc., are based on the simulated environment in the simulation data to give the optimal personalized blood glucose management recommendation strategy. The shortcoming of this method is that the treatment recommendation based on online reinforcement learning is mostly based on the optimization of the blood glucose management recommendation strategy based on the simulation data simulated by the simulation environment. However, there is a large difference between simulation data and real-world medical data, and this type of recommendation strategy may not be effective in real medical scenarios. Therefore, there is an urgent need for an advanced reinforcement learning algorithm that can use real-world data to recommend personalized blood glucose management strategies to patients.

[0023] In order to solve the problems existing in the related art, the embodiments of the present application provide corresponding solutions, which are described below.

[0024] Figure 1 is a flow chart of a method for determining a blood sugar management strategy according to an embodiment of the present application, such as Figure 1 As shown, the method includes: Step S202, obtaining the physical condition data of the target object in the target period.

[0025] In the above step S202, the physical state data includes two parts: static feature data and time series feature data. Specifically, the static feature data includes gender, age, past history, family history, allergy history, etc., and the time series feature data mainly includes the blood sugar test results of the day, blood test data, daily medication data, etc. The medication data may include, for example, the type and dosage of medication.

[0026] Step S204, use the fine-grained blood glucose management model to process the physical state data to obtain a target blood glucose management strategy, wherein the fine-grained blood glucose management model is trained by the first blood glucose management strategy set determined by the coarse-grained blood glucose management model, the accuracy of the fine-grained blood glucose management model is higher than that of the coarse-grained blood glucose management model, the coarse-grained blood glucose management model is trained by the historical physical state data of the target object in the first time period, the first blood glucose management strategy set is determined according to the historical physical state data of the target object in the second time period, and the end time of the first time period is before the start time of the second time period.

[0027] In the above step S204, the coarse-grained blood glucose management model is first trained by the historical physical state data of the target object in the first time period, and then the trained coarse-grained blood glucose management model is used to process the historical physical state data of the target object in the second time period to obtain a first blood glucose management strategy set. In this step, the first time period can be, for example, the same day, and the second time period can be, for example, the next day. For example, the first time period can be from 0:00 to 24:00 on January 1, 2021, and the second time period can be from 0:00 to 24:00 on January 2, 2021.

[0028] In the above-mentioned method for determining the blood sugar management strategy, the fine-grained blood sugar management model is trained in the following manner: obtaining the historical basic data of the target object, wherein the historical basic data includes the historical physical state data of the first period, the historical blood sugar management strategy of the first period, the historical physical state data of the second period, the reward function and the discharge data, the historical physical state data includes the static feature data and the time series feature data of the target object, the static feature data includes the attribute feature data of the target object, the time series feature data includes the historical inspection data and the historical medication data of the target object, wherein the historical inspection data includes the historical blood sugar data; inputting the historical physical state data of the first period into the initial An initial coarse-grained blood glucose management model is constructed, and the historical blood glucose management strategies of the first time period are used as labels for training to obtain a coarse-grained blood glucose management model; the historical physical state data of the second time period is input into the coarse-grained blood glucose management model to obtain a first probability distribution of predicted blood glucose management strategies corresponding to the historical physical state data of the second time period, and after sorting the first probability distribution in descending order, the blood glucose management strategies corresponding to the first K probability distributions are determined as a first blood glucose management strategy set, where K is a positive integer; the historical physical state data of the second time period and the first blood glucose management strategy set are input into the initial fine-grained blood glucose management model for training to obtain a fine-grained blood glucose management model.

[0029] In the embodiment of the present application, the historical physical status data of the first period may be, for example, the daily historical physical status data of the target object (or patient), which may be t The historical blood sugar management strategy for the first period can be, for example, the blood sugar management strategy adopted by the doctor on a daily basis, which can be expressed as a t The historical physical state data of the second period can be, for example, the historical physical state data of the target object on the next day, which can be expressed as s t+1 Representation; the reward function can be used r t Indicates; discharge data available t Indicates that, d t 0 or 1, the day of discharge is 1, and the rest is 0. t 、a t 、st+1 、r t ,d t ) form to construct historical basic data. Among them, s t 、s t+1 It includes two parts: static feature data and time series feature data. Specifically, static features include attribute feature data of the target object, which may include gender, age, medical history, family history, allergy history, etc.; time series feature data includes historical test data and historical medication data of the target object, among which historical test data includes historical blood sugar data, blood test data, etc., such as fasting blood sugar, pre-meal and post-meal blood sugar, creatinine, glycosylated hemoglobin, etc., and historical medication data may include the type and dosage of medication used on the day, such as insulin 8mg. t The recommended blood sugar management strategy may specifically refer to the insulin dose required for the patient. In the embodiment of the present application, the insulin dose is discretized into a fixed range interval (for example, the commonly used clinical recommended insulin range is 0-40 IU. The embodiment of the present application sets the recommended dose to 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21-25, 26-30, 30-35, 36-40, a total of 25 range intervals. The blood sugar management strategy recommended below is one of the categories of the above 25 intervals). t The evaluation of the intervention effect of the same-day blood glucose management strategy mainly considers whether the next-day blood glucose is within the target range, as well as the differences between the coarse-grained blood glucose management model and the fine-grained blood glucose management model described later.

[0030] It should be noted that before training the coarse-grained blood sugar management model and the fine-grained blood sugar management model, the acquired historical basic data can be preprocessed. t and t+1 They are all composed of structured data. Among them, the methods for handling null values ​​in structured data are model fitting result filling and mode filling. t For blood sugar management strategy, the null value processing method is to fill with zero, where null value means no medication was taken on that day, and the medication dosage is 0, so fill with 0. The reward function and discharge data are not preprocessed. Each sample after preprocessing can be expressed as .

[0031] If the historical basic data is preprocessed, the historical physical state data of the target object in the first period is used based on the preprocessed historical basic data. And the historical blood sugar management strategy for the first period As a label to train the coarse-grained blood sugar management model, after the training is completed, the historical physical status data of the target object in the second period Input the trained coarse-grained blood sugar management model to obtain the first probability distribution of predicting blood sugar management strategies, and after sorting the first probability distributions in descending order, the blood sugar management strategies corresponding to the first K probability distributions are determined as the first blood sugar management strategy set. The first blood sugar management strategy set can be used The fine-grained blood glucose management model is trained using the historical physical state data of the second period and the first blood glucose management strategy set.

[0032] In the above steps, the coarse-grained blood glucose management model includes a first blood glucose management strategy model, and the historical physical state data of the first time period is input into the initial coarse-grained blood glucose management model for training to obtain a coarse-grained blood glucose management model, including: inputting the static feature data in the historical physical state data of the first time period into the deep learning network of the first blood glucose management strategy model to obtain the first feature; inputting the time series feature data in the historical physical state data of the first time period into the neural network model of the first blood glucose management strategy model to obtain the second feature; after splicing the first feature and the second feature, a spliced ​​feature is obtained; the spliced ​​feature is input into the multi-layer perceptron network of the first blood glucose management strategy model to obtain a second probability distribution of the predicted blood glucose management strategy corresponding to the historical physical state data of the first time period; when the cross entropy loss between the second probability distribution and the probability distribution corresponding to the historical blood glucose management strategy of the first time period does not meet the first threshold, adjusting the first parameter of the first blood glucose management strategy model until the cross entropy loss meets the first threshold, stopping the training, and obtaining the coarse-grained blood glucose management model.

[0033] Figure 2 is a schematic diagram of a blood sugar management model according to an embodiment of the present application. Figure 2 The training process of the above-mentioned coarse-grained blood glucose management model is described. In the embodiment of the present application, if the historical basic data is preprocessed, the preprocessed historical physical state data of the first period is processed based on the preprocessed historical basic data. Input the coarse-grained and fine-grained blood glucose management model framework, and use the blood glucose management strategy adopted by the doctor as a t As labels, a coarse-grained blood glucose management model based on supervised learning is constructed. Therefore, it can learn doctors' blood glucose management strategies based on doctors' experience, so that the coarse-grained blood glucose management model can be close to doctors' experience. Specifically, the coarse-grained blood glucose management model includes a first blood glucose management strategy model. Figure 3 It includes DNN network, Transformer model and MLP.

[0034] Since the historical physical state data of the first period after preprocessing It contains two parts: static feature data and time series feature data. Therefore, the deep learning network and neural network model are used to represent them in depth, respectively, to obtain the first feature and the second feature. The first feature can be expressed by h static Indicates that the second feature can be used h dynamic Indicates that the deep learning network in this embodiment of the application can be Figure 3 In the DNN network, the neural network model in this embodiment of the application can be Figure 3 The Transformer model in static and the second feature h dynamic The splicing feature can be obtained by splicing. static II dynamic express , the concatenated features are input into the multi-layer perceptron MLP network of the first blood sugar management strategy model, and softmax classification is adopted to obtain the second probability distribution of the predicted blood sugar management strategy corresponding to the historical physical state data of the first period, and the process adopts the multi-classification cross entropy loss function for training. Specifically, when the cross entropy loss of the probability distribution corresponding to the second probability distribution and the historical blood sugar management strategy of the first period does not meet the first threshold, the first parameter of the first blood sugar management strategy model is adjusted until the cross entropy loss meets the first threshold and the training is stopped to obtain the coarse-grained blood sugar management model.

[0035] In the above steps, the fine-grained blood glucose management model includes a second blood glucose management strategy model, a first network and a second network. The historical physical state data of the second time period and the first blood glucose management strategy set are input into the initial fine-grained blood glucose management model for training to obtain a fine-grained blood glucose management model, including: inputting the historical physical state data of the second time period into the second blood glucose management strategy model to obtain a third probability distribution of predicted blood glucose management strategies corresponding to the historical physical state data of the second time period; determining a candidate set of blood glucose management strategies corresponding to the historical physical state data of the second time period based on the third probability distribution; inputting the historical physical state data of the second time period and the candidate set of blood glucose management strategies into the first network to obtain a fourth probability distribution of predicted blood glucose management strategies corresponding to the historical physical state data of the second time period, and determining the blood glucose management strategy corresponding to the maximum probability of the fourth probability distribution as the recommended blood glucose management strategy; determining a total loss function based on the recommended blood glucose management strategy, and adjusting the parameters of the first network and the second blood glucose management strategy model based on the total loss function, stopping the training when the total loss value of the fine-grained blood glucose management model meets the second threshold, and obtaining a fine-grained blood glucose management model, wherein the parameters of the second network are adjusted based on the parameters of the first network.

[0036] In the above steps, a candidate set of blood glucose management strategies corresponding to the historical physical state data of the second time period is determined based on the third probability distribution, including: determining the maximum probability in the third probability distribution as the target probability; determining the ratio of each probability in the third probability distribution to the target probability to obtain a ratio set; and determining the blood glucose management strategy corresponding to the probability greater than the third threshold in the ratio set as the candidate set of blood glucose management strategies.

[0037] In the embodiment of the present application, if the historical basic data is preprocessed, a fine-grained blood glucose management model of offline reinforcement learning is constructed based on the preprocessed historical basic data. The fine-grained blood glucose management model includes a second blood glucose management strategy model, a first network, and a second network. It should be noted that the second blood glucose management strategy model can be used G w It indicates that its network architecture is consistent with the network architecture of the first blood sugar management strategy model in the coarse-grained blood sugar management model, and is pre-trained using the softmax multi-classification cross entropy loss function. The initialization parameters of the first network and the second network are the same, and their architectures both use three-layer DNN. The three-layer DNN refers to a neural network structure that includes an input layer, a hidden layer, and an output layer, wherein the input layer is responsible for receiving input data; the hidden layer is responsible for processing the input data, extracting its features and passing them to the output layer. In a three-layer DNN, there is usually only one hidden layer; the output layer outputs the final result based on the features extracted by the hidden layer.

[0038] Specifically, the training process of the fine-grained blood glucose management model is as follows: The historical physical status data of the second period after preprocessing Input the second blood sugar management strategy model to obtain the third probability distribution of the predicted blood sugar management strategy corresponding to the historical physical state data of the second period, which can be used According to the third probability distribution, a candidate set of blood glucose management strategies corresponding to the historical physical state data of the second period is determined. Specifically, the maximum probability in the third probability distribution is determined as the target probability. The target probability can be Indicates that the ratio of each probability in the third probability distribution to the target probability is determined to obtain a ratio set; the blood glucose management strategy corresponding to the probability greater than the third threshold in the ratio set is determined as a blood glucose management strategy candidate set, and the blood glucose management strategy candidate set can be used In other words, the third threshold can be represented by β, for example, it can be set to 0.3. The formula of the above process is as follows: (1) The historical physical status data of the second period and candidate sets of blood sugar management strategies The fourth probability distribution of the predicted blood sugar management strategy corresponding to the historical physical state data of the second period is input into the first network, and the blood sugar management strategy corresponding to the maximum probability of the fourth probability distribution is determined as the recommended blood sugar management strategy. The recommended blood sugar management strategy can be used The formula is as follows: (2) Among them, θ is the training parameter of the first network, Indicates that the historical physical status data of the second period and candidate sets of blood sugar management strategies Input the training parameter θ into the first network to obtain the fourth probability distribution.

[0039] The total loss function is determined according to the recommended blood sugar management strategy, and the parameters of the first network and the second blood sugar management strategy model are adjusted according to the total loss function. When the total loss value of the fine-grained blood sugar management model meets the second threshold, the training is stopped to obtain the fine-grained blood sugar management model.

[0040] In the above steps, the total loss function is determined according to the recommended blood sugar management strategy, including: inputting the historical physical state data of the first time period and the historical blood sugar management strategy of the first time period into the first network to obtain a first result; inputting the historical physical state data of the second time period and the recommended blood sugar management strategy into the second network to obtain a second result; determining the first loss function according to the reward function, the first result and the second result; inputting the historical physical state data of the first time period into the second blood sugar management strategy model to obtain a third result; determining the second loss function according to the historical blood sugar management strategy of the first time period and the third result; determining the total loss function according to the first loss function and the second loss function.

[0041] In the embodiment of the present application, based on the update mechanism of the reinforcement learning Bellman equation, the embodiment of the present application designs the following total loss function to update the training parameters θ of the first network and the parameters of the second blood glucose management strategy model. The total loss function can be represented by L. The formula is as follows: 3) Among them, the total loss function includes the first loss function and the second loss function. The first loss function It can be expressed as L1, the second loss function It can be expressed as L2, where L1 is the loss of the reinforcement learning Bellman equation; L2 is the cross entropy loss function added in this application, which ensures that the blood sugar management strategy output by reinforcement learning does not deviate too far from the doctor's strategy. The input of the total loss function is the historical physical status data of the first period , historical physical status data for the second period , Historical blood sugar management strategy for the first period , the reward function r t , recommended blood sugar management strategies . Indicates that the historical physical state data of the first period and the historical blood sugar management strategy of the first period are input into the first network with the training parameter θ to obtain a first result; Indicates that the historical physical status data of the second period and the recommended blood sugar management strategy are input into the second network with the training parameter θ' to obtain a second result; To record the historical physical status data of the first period Input to the second blood glucose management strategy model G w The output blood sugar management strategy is the third result. γ is a hyperparameter (generally set to a value less than 1, and is set to 0.99 in the embodiment of the present application).

[0042] In order to avoid excessive fluctuations in the update of the second network parameters and unstable models, which will affect the recommendation performance, the parameters of the second network are updated according to the parameters of the first network using a soft update method, which is expressed as: (4) in, is a parameter for soft update (set to 0.005 in the embodiment of the present application).

[0043] In the above steps, the reward function includes: obtaining the difference between the historical blood sugar data and the preset blood sugar data of the second time period; determining the first reward parameter based on the range of the difference; determining the second reward parameter based on the relationship between the recommended blood sugar management strategy and the first blood sugar management strategy set; determining the reward function based on the first reward parameter and the second reward parameter.

[0044] In the embodiment of the present application, the reward function r t It is composed of the sum of two parts: 1) the difference between the historical blood sugar data in the second period and the preset blood sugar data r t,1 , such as the difference between the next day's blood sugar and the target blood sugar, r t,1 represents the first reward parameter, which can be expressed by Formula 5; 2) determining the second reward parameter according to the relationship between the recommended blood sugar management strategy and the first blood sugar management strategy set, and the second reward parameter can be expressed by r t,2 Indicates, for example, the difference between fine-grained blood glucose management strategy and coarse-grained blood glucose management strategy, for r t,2 Specifically, if the blood glucose management strategy recommended by the fine-grained blood glucose management strategy model based on reinforcement learning is , is one of the K strategies recommended by the coarse-grained blood glucose management strategy model based on supervised learning, namely: , a positive reward of 0.5 is given. tIt not only improves the rate of achieving the target blood sugar level the next day, thus ensuring that the patient's blood sugar level changes from abnormal to normal, but also makes the fine-grained blood sugar management model approach the coarse-grained blood sugar management model that integrates the doctor's strategy, thus ensuring that the effectiveness of the designed blood sugar management model is further improved. t The composition part includes two parts. The first part makes the next day's blood sugar close to the target blood sugar, thereby ensuring that the patient's blood sugar changes from abnormal to normal; the second part constrains the fine-grained blood sugar management strategy to fall within the confidence interval of the K strategies of the coarse-grained blood sugar management strategy, thereby integrating the doctor's experience.

[0045] (5) Wherein, b represents the blood glucose concentration in mg / dL. , the entire training process is an offline process.

[0046] The application phase includes data collection, data preprocessing, and model application. Specifically: The data collection phase is used to collect daily physical status data of patients during hospitalization. t 1. The data preprocessing stage is used to fill in empty values, remove text noise, and perform standardization on the collected patient physical status data. The preprocessed physical status data is expressed as The model application phase is used to use only the fine-grained blood glucose management model to pre-process the patient's physical state , input into the second blood glucose management strategy model, and output the final recommended personalized blood glucose management strategy a t . Because the model uses real-world data from patients, and the recommended blood sugar management strategy takes into account the patient's own physical condition, and incorporates the physician's clinical experience during the training process, the constructed perioperative blood sugar abnormality management method is more effective, more clinically applicable, and safer.

[0047] The following is an example of insulin dosage recommendation for patients with perioperative hyperglycemia. Static feature data is collected on the first day of hospitalization, including demographics, medical history, family history, allergy history, etc. Dynamic data is collected daily during hospitalization, including structured data such as blood glucose meter monitoring data, blood test data, and medication data.

[0048] For example, the static feature data of patient A includes: male, 65 years old, with a history of diabetes and hypertension, and no history of allergies. The dynamic feature data includes: the previous daily 24-hour dynamic blood glucose monitoring data, blood test data, and medication data. The recommended insulin dose for the patient on the first day after surgery is 20 IU, so the recommended insulin dose for the patient is output to the clinician as 20 IU.

[0049] Figure 3 is a structural diagram of a device for determining a blood sugar management strategy according to an embodiment of the present application, such as Figure 3 As shown, the device comprises: An acquisition module 32, used to acquire the physical state data of the target object in the target period; The processing module 34 is used to process the physical state data using a fine-grained blood glucose management model to obtain a target blood glucose management strategy, wherein the fine-grained blood glucose management model is trained by a first blood glucose management strategy set determined by the coarse-grained blood glucose management model, the accuracy of the fine-grained blood glucose management model is higher than that of the coarse-grained blood glucose management model, the coarse-grained blood glucose management model is trained by the historical physical state data of the target object in the first time period, the first blood glucose management strategy set is determined according to the historical physical state data of the target object in the second time period, and the end time of the first time period is before the start time of the second time period.

[0050] The above-mentioned device for determining the blood sugar management strategy also includes a training module 36, which is used to train the fine-grained blood sugar management model. Specifically, the training module is used to obtain the historical basic data of the target object, wherein the historical basic data includes the historical physical state data of the first time period, the historical blood sugar management strategy of the first time period, the historical physical state data of the second time period, the reward function and the discharge data, the historical physical state data includes the static feature data and the time series feature data of the target object, the static feature data includes the attribute feature data of the target object, the time series feature data includes the historical test data and the historical medication data of the target object, wherein the historical test data includes the historical blood sugar data; the historical physical state data of the first time period is converted into the historical physical state data of the target object, the historical physical state data of the target object ... The historical physical state data are input into the initial coarse-grained blood glucose management model, and the historical blood glucose management strategies of the first time period are used as labels for training to obtain a coarse-grained blood glucose management model; the historical physical state data of the second time period are input into the coarse-grained blood glucose management model to obtain a first probability distribution of predicted blood glucose management strategies corresponding to the historical physical state data of the second time period, and after sorting the first probability distribution in descending order, the blood glucose management strategies corresponding to the first K probability distributions are determined as a first blood glucose management strategy set, where K is a positive integer; the historical physical state data of the second time period and the first blood glucose management strategy set are input into the initial fine-grained blood glucose management model for training to obtain a fine-grained blood glucose management model.

[0051] In the training module in the above-mentioned blood sugar management strategy determination device, the coarse-grained blood sugar management model includes a first blood sugar management strategy model, and the training module is also used to input the static feature data in the historical physical state data of the first time period into the deep learning network of the first blood sugar management strategy model to obtain the first feature; input the time series feature data in the historical physical state data of the first time period into the neural network model of the first blood sugar management strategy model to obtain the second feature; after splicing the first feature and the second feature, a spliced ​​feature is obtained; the spliced ​​feature is input into the multi-layer perceptron network of the first blood sugar management strategy model to obtain a second probability distribution of the predicted blood sugar management strategy corresponding to the historical physical state data of the first time period; when the cross entropy loss between the second probability distribution and the probability distribution corresponding to the historical blood sugar management strategy of the first time period does not meet the first threshold, adjust the first parameter of the first blood sugar management strategy model until the cross entropy loss meets the first threshold, stop training, and obtain a coarse-grained blood sugar management model.

[0052] In the training module in the above-mentioned blood sugar management strategy determination device, the fine-grained blood sugar management model includes a second blood sugar management strategy model, a first network and a second network. The training module is also used to input the historical physical state data of the second time period into the second blood sugar management strategy model to obtain a third probability distribution of the predicted blood sugar management strategy corresponding to the historical physical state data of the second time period; determine the blood sugar management strategy candidate set corresponding to the historical physical state data of the second time period based on the third probability distribution; input the historical physical state data of the second time period and the blood sugar management strategy candidate set into the first network to obtain a fourth probability distribution of the predicted blood sugar management strategy corresponding to the historical physical state data of the second time period, and determine the blood sugar management strategy corresponding to the maximum probability of the fourth probability distribution as the recommended blood sugar management strategy; determine the total loss function based on the recommended blood sugar management strategy, and adjust the parameters of the first network and the second blood sugar management strategy model according to the total loss function, stop training when the total loss value of the fine-grained blood sugar management model meets the second threshold, and obtain a fine-grained blood sugar management model, wherein the parameters of the second network are adjusted based on the parameters of the first network.

[0053] In the training module in the above-mentioned blood sugar management strategy determination device, the training module is also used to determine the maximum probability in the third probability distribution as the target probability; determine the ratio of each probability in the third probability distribution to the target probability to obtain a ratio set; and determine the blood sugar management strategy corresponding to the probability greater than the third threshold in the ratio set as the blood sugar management strategy candidate set.

[0054] In the training module in the above-mentioned blood sugar management strategy determination device, the training module is also used to input the historical physical state data of the first time period and the historical blood sugar management strategy of the first time period into the first network to obtain a first result; input the historical physical state data of the second time period and the recommended blood sugar management strategy into the second network to obtain a second result; determine the first loss function based on the reward function, the first result and the second result; input the historical physical state data of the first time period into the second blood sugar management strategy model to obtain a third result; determine the second loss function based on the historical blood sugar management strategy of the first time period and the third result; determine the total loss function based on the first loss function and the second loss function.

[0055] In the training module of the above-mentioned blood sugar management strategy determination device, the reward function includes: obtaining the difference between the historical blood sugar data and the preset blood sugar data of the second time period; determining the first reward parameter based on the range of the difference; determining the second reward parameter based on the relationship between the recommended blood sugar management strategy and the first blood sugar management strategy set; determining the reward function based on the first reward parameter and the second reward parameter.

[0056] It should be noted that Figure 3 The blood glucose management strategy determination device shown is used to perform Figure 1 The method for determining the blood sugar management strategy shown in the figure, therefore the relevant explanations and instructions in the above-mentioned method for determining the blood sugar management strategy are also applicable to the device for determining the blood sugar management strategy, and will not be repeated here.

[0057] The method for determining the blood sugar management strategy provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 4 FIG. 1 shows a hardware structure block diagram of a computer terminal for implementing a method for determining a blood sugar management strategy. Figure 4 As shown, the computer terminal 10 may include one or more (102a, 102b, ..., 102n are used to illustrate) processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module 106 for communication functions connected via a wired and / or wireless network. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. Those skilled in the art will understand that Figure 4 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 4 More or fewer components as shown, or with Figure 4 Different configurations shown.

[0058] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0059] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method for determining the blood glucose management strategy in the embodiment of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the method for determining the blood glucose management strategy described above is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0060] The transmission module 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission module 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0061] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 .

[0062] It should be noted that, in some optional embodiments, the above Figure 4 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. It should be noted that Figure 4This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0063] An embodiment of the present application also provides an electronic device, which includes a memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute a determination method for implementing the above-mentioned blood sugar management strategy.

[0064] An embodiment of the present application also provides a non-volatile storage medium, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned method for determining the blood sugar management strategy by running the computer program.

[0065] An embodiment of the present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the method for determining the blood glucose management strategy in each embodiment of the present application.

[0066] The embodiments of the present application also provide a computer program, which, when executed by a processor, implements the steps of the method for determining the blood glucose management strategy in each embodiment of the present application.

[0067] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0068] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0069] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0070] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0071] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0072] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc., which can store program code.

[0073] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for determining a blood sugar management strategy, characterized in that: include: Obtaining the physical condition data of the target object during the target period; A fine-grained blood glucose management model is used to process the physical state data to obtain a target blood glucose management strategy, wherein the fine-grained blood glucose management model is trained with a first blood glucose management strategy set determined by a coarse-grained blood glucose management model, the accuracy of the fine-grained blood glucose management model is higher than that of the coarse-grained blood glucose management model, the coarse-grained blood glucose management model is trained with the historical physical state data of the target object in a first time period, the first blood glucose management strategy set is determined based on the historical physical state data of the target object in a second time period, and the end time of the first time period is before the start time of the second time period.

2. The method according to claim 1, characterized in that The fine-grained blood glucose management model is trained in the following way: Acquire historical basic data of the target object, wherein the historical basic data includes historical physical state data of a first period, historical blood sugar management strategy of the first period, historical physical state data of a second period, reward function and discharge data, the historical physical state data includes static feature data and time series feature data of the target object, the static feature data includes attribute feature data of the target object, the time series feature data includes historical inspection data and historical medication data of the target object, wherein the historical inspection data includes historical blood sugar data; Inputting the historical physical status data of the first period into an initial coarse-grained blood glucose management model, and training with the historical blood glucose management strategy of the first period as a label to obtain the coarse-grained blood glucose management model; Inputting the historical physical state data of the second period into the coarse-grained blood sugar management model, obtaining a first probability distribution of predicted blood sugar management strategies corresponding to the historical physical state data of the second period, and sorting the first probability distributions in descending order, and determining the blood sugar management strategies corresponding to the first K probability distributions as the first blood sugar management strategy set, where K is a positive integer; The historical physical status data of the second time period and the first blood glucose management strategy set are input into an initial fine-grained blood glucose management model for training to obtain the fine-grained blood glucose management model.

3. The method according to claim 2, characterized in that The coarse-grained blood glucose management model includes a first blood glucose management strategy model. The historical physical state data of the first period is input into the initial coarse-grained blood glucose management model for training to obtain the coarse-grained blood glucose management model, including: Inputting static feature data in the historical physical state data of the first time period into the deep learning network of the first blood sugar management strategy model to obtain a first feature; Inputting the time series feature data in the historical physical state data of the first time period into the neural network model of the first blood sugar management strategy model to obtain a second feature; After splicing the first feature and the second feature, a spliced ​​feature is obtained; Inputting the concatenated features into the multi-layer perceptron network of the first blood sugar management strategy model to obtain a second probability distribution of the predicted blood sugar management strategy corresponding to the historical physical state data of the first time period; When the cross entropy loss between the second probability distribution and the probability distribution corresponding to the historical blood glucose management strategy of the first time period does not meet the first threshold, the first parameter of the first blood glucose management strategy model is adjusted until the cross entropy loss meets the first threshold, and the training is stopped to obtain the coarse-grained blood glucose management model.

4. The method according to claim 2, characterized in that: The fine-grained blood glucose management model includes a second blood glucose management strategy model, a first network, and a second network. The historical body state data of the second period and the first blood glucose management strategy set are input into the initial fine-grained blood glucose management model for training to obtain the fine-grained blood glucose management model, including: Inputting the historical physical state data of the second time period into the second blood glucose management strategy model to obtain a third probability distribution of the predicted blood glucose management strategy corresponding to the historical physical state data of the second time period; determining a candidate set of blood sugar management strategies corresponding to the historical physical state data of the second time period according to the third probability distribution; Inputting the historical physical state data of the second time period and the candidate set of blood sugar management strategies into the first network, obtaining a fourth probability distribution of predicted blood sugar management strategies corresponding to the historical physical state data of the second time period, and determining the blood sugar management strategy corresponding to the maximum probability of the fourth probability distribution as the recommended blood sugar management strategy; Determine a total loss function based on the recommended blood glucose management strategy, and adjust the parameters of the first network and the second blood glucose management strategy model based on the total loss function, stop training when the total loss value of the fine-grained blood glucose management model meets a second threshold, and obtain the fine-grained blood glucose management model, wherein the parameters of the second network are adjusted based on the parameters of the first network.

5. The method according to claim 4, characterized in that Determining a blood sugar management strategy candidate set corresponding to the historical physical state data of the second time period according to the third probability distribution includes: Determine the maximum probability in the third probability distribution as the target probability; Determine a ratio of each probability in the third probability distribution to the target probability to obtain a ratio set; The blood glucose management strategies corresponding to the probabilities in the ratio set that are greater than a third threshold are determined as the blood glucose management strategy candidate set.

6. The method according to claim 4, characterized in that Determining a total loss function according to the recommended blood sugar management strategy includes: Inputting the historical physical state data of the first time period and the historical blood sugar management strategy of the first time period into the first network to obtain a first result; inputting the historical physical status data of the second time period and the recommended blood sugar management strategy into the second network to obtain a second result; Determine a first loss function according to the reward function, the first result and the second result; Inputting the historical physical status data of the first time period into the second blood glucose management strategy model to obtain a third result; Determining a second loss function according to the historical blood sugar management strategy of the first time period and the third result; The total loss function is determined according to the first loss function and the second loss function.

7. The method according to claim 6, characterized in that The reward function includes: Obtaining a difference between the historical blood sugar data and the preset blood sugar data during the second period; Determining a first reward parameter according to a range of the difference; determining a second reward parameter according to a relationship between the recommended blood sugar management strategy and the first blood sugar management strategy set; The reward function is determined according to the first reward parameter and the second reward parameter.

8. A device for determining a blood sugar management strategy, characterized in that: include: An acquisition module, used to acquire the physical status data of the target object during the target period; A processing module is used to process the physical state data using a fine-grained blood glucose management model to obtain a target blood glucose management strategy, wherein the fine-grained blood glucose management model is trained with a first blood glucose management strategy set determined by a coarse-grained blood glucose management model, the accuracy of the fine-grained blood glucose management model is higher than that of the coarse-grained blood glucose management model, the coarse-grained blood glucose management model is trained with the historical physical state data of the target object in a first time period, the first blood glucose management strategy set is determined based on the historical physical state data of the target object in a second time period, and the end time of the first time period is before the start time of the second time period.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute a method for determining a blood glucose management strategy according to any one of claims 1 to 7.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the method for determining the blood glucose management strategy according to any one of claims 1 to 7 by running the computer program.

11. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the method for determining a blood glucose management strategy according to any one of claims 1 to 7 is implemented.

Images