Estimation device, estimation method, estimation program, estimation system

Abstract

To enable patients to simply grasp a proper insulin dosage.SOLUTION: An estimation device 100 comprises: an acquisition unit 103 that acquires state variables including blood sugar information, activity information, and meal information; an estimation unit 1012 that uses an estimation model to estimate an insulin dosage based on the state variables and a target range; and an output unit 104 that outputs the insulin dosage.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to an estimation device, an estimation method, an estimation program, and an estimation system.

Background Art

[0002] Diabetes treatment and quality of life (QOL) are closely related. In all treatment methods, the higher the treatment compliance, the higher the QOL, and the higher the QOL, the higher the compliance, resulting in improved blood glucose control.

[0003] Since patients with type 1 diabetes have depleted endogenous insulin secretion ability, it is essential to receive an appropriate amount of insulin administration by insulin injection to maintain life. Patients with type 1 diabetes are constantly troubled by the hassle of calculating the appropriate insulin dosage. Furthermore, patients with type 1 diabetes are forced to keep their food intake constant and limit unplanned food intake in order to avoid hypoglycemia and severe hyperglycemia. Therefore, patients with type 1 diabetes tend to have a sense of restraint that they cannot eat freely and a sense of alienation that they alone are undergoing a special diet therapy, which is a factor that reduces QOL.

[0004] Japanese Patent Application Laid-Open No. 2015-118516 (Patent Document 1) describes obtaining, as calculation information used for calculating the insulin dosage, first calculation information considering the instructed dosage and blood glucose level, and second calculation information considering the food intake amount, and calculating the insulin dosage for each patient and each time period based on the obtained information.

[0005] As an existing method, the carbohydrate counting method, which self-adjusts the insulin dosage according to the amount of carbohydrates contained in the meal, is known. According to the carbohydrate counting method, the degree of freedom in diet increases, so an improvement in QOL can be expected.

Prior Art Documents

Patent Documents

[0006] [Patent Document 1] Japanese Patent Publication No. 2015-118516 [Overview of the project] [Problems that the invention aims to solve]

[0007] However, the calculation of carbohydrate content in a meal is often based on empirical rules, and discrepancies can occur between the calculated result and the actual amount of carbohydrates, which is one of the causes of poor blood glucose control, such as hypoglycemia. Furthermore, the carbohydrate / insulin ratio fluctuates constantly depending on factors such as the activity level of type 1 diabetic patients, but the calculation method described in Patent Document 1 did not take the activity level of type 1 diabetic patients into consideration.

[0008] This disclosure was made to address these issues, and its purpose is to enable patients to be presented with an appropriate insulin dosage that takes into account their activity level. [Means for solving the problem]

[0009] An estimation device according to a certain aspect of this disclosure is an estimation device for estimating the amount of insulin to be administered in order to bring a patient's blood glucose level into a target range, comprising: an acquisition unit that acquires state variables including blood glucose information indicating the patient's blood glucose level, patient activity information different from the blood glucose information, and dietary information regarding the food the patient consumes; an estimation unit that uses a dosage estimation model to estimate the amount of insulin to be administered based on the state variables and the target range; and an output unit that outputs the amount of insulin to be administered.

[0010] Estimation methods according to other aspects of the present disclosure are estimation methods for estimating an insulin dose to bring a patient's blood glucose level into a target range using a computer, comprising the steps of: obtaining state variables including blood glucose information indicating the patient's blood glucose level, patient activity information separate from the blood glucose information, and dietary information regarding the food the patient consumes; estimating an insulin dose based on the state variables and the target range using a dose estimation model; and outputting an insulin dose.

[0011] Estimation programs according to other aspects of this disclosure are estimation programs for estimating an insulin dose to bring a patient's blood glucose level into a target range, and cause a computer to perform the steps of: acquiring state variables including blood glucose information indicating the patient's blood glucose level, patient activity information different from the blood glucose information, and dietary information regarding the food the patient consumes; estimating an insulin dose based on the state variables and the target range using a dose estimation model; and outputting an insulin dose.

[0012] An estimation system according to other aspects of this disclosure comprises an estimation device, a blood glucose sensor for detecting blood glucose information from a patient, an activity sensor for detecting activity information from a patient, and a portable device carried by the patient that collects blood glucose information and activity information from the blood glucose sensor and the activity sensor, wherein the portable device includes a camera for capturing images of food consumed by the patient, a generator for generating meal information based on the images captured by the camera, a communication unit for communicating with the estimation device via a network, and a display unit, wherein the communication unit transmits blood glucose information, activity information, and meal information to the estimation device, the output unit transmits insulin dosage to the portable device, and the display unit displays the insulin dosage on the display unit. [Effects of the Invention]

[0013] According to this disclosure, state variables including blood glucose information indicating the patient's blood glucose level, patient activity information different from blood glucose information, and dietary information regarding the food the patient consumes are acquired. Using a dosage estimation model, the insulin dosage is estimated based on the state variables and the target range, so that an appropriate insulin dosage that takes activity level into account can be presented to the patient. [Brief explanation of the drawing]

[0014] [Figure 1] This is a diagram illustrating the overall configuration of the estimation system. [Figure 2] This is a block diagram illustrating the general operation of the estimation system. [Figure 3] This figure shows an example of how a patient's blood glucose levels fluctuate. [Figure 4] This is a diagram illustrating the first estimation model (peak value estimation model). [Figure 5] This is a diagram to explain the second estimation model (dosage estimation model). [Figure 6] This is a diagram illustrating the procedure for generating the first estimation model. [Figure 7] This is a diagram illustrating the procedure for generating the second estimation model. [Figure 8] This diagram illustrates the difference in the time periods for acquiring information used to estimate the amount of bolus insulin and the amount of basal insulin. [Figure 9] This flowchart shows the procedure for estimating insulin levels using the first and second estimation models. [Figure 10] This figure shows a modified version of the first estimation model (Modification 1). [Figure 11] This figure shows a modified version of the second estimation model (Modification 1). [Figure 12] This figure shows the third estimated model (modified version 2). [Figure 13] This figure shows a modified version of the second estimation model (Modified Version 2). [Modes for carrying out the invention]

[0015] This embodiment will be described in detail with reference to the drawings. In the drawings, the same or corresponding parts are denoted by the same reference numerals, and the description thereof will not be repeated in principle.

[0016] <Estimation system> FIG. 1 is a diagram for explaining the overall configuration of an estimation system 1. The estimation system 1 presents an optimal insulin dosage for a type 1 diabetic patient who requires treatment to reproduce a physiological insulin secretion pattern by insulin injection. The estimation system 1 can present appropriate amounts of basal insulin and bolus insulin, for example, to a type 1 diabetic patient who requires insulin injection combining basal insulin and bolus insulin.

[0017] The estimation system 1 includes a blood glucose sensor 10 and an activity sensor 20 attached to a type 1 diabetic patient (hereinafter referred to as a patient), a portable device 50 carried by the patient, and an estimation device 100.

[0018] The blood glucose sensor 10 is constituted by, for example, a continuous glucose monitoring sensor. The blood glucose sensor 10 is attached to the skin of the patient's abdomen or the like. The blood glucose sensor 10 constantly detects the patient's glucose value that fluctuates moment by moment. The blood glucose sensor 10 wirelessly transmits the detected glucose value to the portable device 50 as blood glucose information.

[0019] The activity sensor 20 is constituted by, for example, a wearable device that is also used as a wristwatch or the like. The activity sensor 20 constantly detects the patient's activity information that fluctuates moment by moment. The activity sensor 20 detects, for example, blood pressure, blood oxygen concentration, body temperature, body weight, calorie consumption, heart rate, and moving distance as activity information. The activity sensor 20 wirelessly transmits the detected activity information to the portable device 50.

[0020] The portable device 50 is comprised of, for example, a smartphone. The portable device 50 includes a display unit 51, an operation unit 52, a communication unit 53, an imaging unit 54, and a generation unit 55. The patient operates the portable device 50 and, each time they eat, uses the imaging unit 54 to photograph the food they plan to eat. The generation unit 55 generates meal information, such as the calories of the food to be eaten, based on the images captured by the imaging unit 54.

[0021] The portable device 50 and the estimation device 100 are connected via a network 90 so as to be communicative. The estimation device 100 is composed of, for example, a cloud server or a part of a cloud server. For example, the estimation device 100 is a computer. The estimation device 100 may be composed of a desktop PC (Personal Computer) or a laptop PC, etc.

[0022] The portable device 50 transmits the blood glucose information received from the blood glucose sensor 10 and the activity information received from the activity sensor 20 to the estimation device 100. The portable device 50 further transmits the meal information generated by the generation unit 55 to the estimation device 100. The portable device 50 may transmit the information received from the blood glucose sensor 10 and the activity sensor 20 to the estimation device 100 each time it receives information. The portable device 50 may store the blood glucose information received from the blood glucose sensor 10 and the activity information received from the activity sensor 20, and transmit the stored information to the estimation device 100 periodically, for example, every 10 minutes or every hour.

[0023] The estimation device 100 includes an arithmetic unit 101, a memory device 102, an input interface 103, an output interface 104, and a bus 105. The arithmetic unit 101 accesses the memory device 102, the input interface 103, and the output interface 104 via the bus 105.

[0024] The arithmetic unit 101 is a computing entity (computer) that performs various processes according to various programs. The arithmetic unit 101 includes, for example, at least one of the following: CPU (Central Processing Unit), FPGA (Field Programmable Gate Array), GPU (Graphics Processing Unit), and MPU (Multi Processing Unit). Furthermore, the arithmetic unit 101 may include volatile memory such as DRAM (Dynamic Random Access Memory) and SRAM (Static Random Access Memory), and non-volatile memory such as ROM (Read Only Memory) and flash memory. The arithmetic unit 101 may also be composed of processing circuits.

[0025] The storage device 102 includes non-volatile memory such as an HDD (Hard Disk Drive) and an SSD (Solid State Drive).

[0026] The estimation device 100 estimates the appropriate insulin dose for the patient based on information received from the portable device 50. The storage device 102 stores the estimation model 60 and the estimation program 71 used for estimating the insulin dose. The estimation model 60 includes a first estimation model 61 and a second estimation model 62.

[0027] The computing unit 101 executes the estimation program 71 using the first estimation model 61 and the second estimation model 62 to estimate the amount of insulin required by the patient. The computing unit 101 functions as a learning unit 1011 in generating the first estimation model 61 and the second estimation model 62, and functions as an estimation unit 1012 in the estimation process using the learned first estimation model 61 and the second estimation model 62.

[0028] The input interface 103 is an example of an "acquisition unit" and acquires information transmitted from the portable device 50 via the network 90.

[0029] The output interface 104 is an example of an "output unit" that outputs information to the outside according to the control of the arithmetic unit 101. For example, the output interface 104 outputs the amount of insulin estimated by the arithmetic unit 101 to the portable device 50. The patient administers insulin using a syringe, referring to the amount of insulin displayed on the display unit 51 of the portable device 50.

[0030] Figure 2 is a block diagram illustrating the general operation of Estimation System 1. Referring to Figure 2, we will explain the general operation of Estimation System 1 in conjunction with patient behavior. Here, we will explain the operation of Estimation System 1 using the example of Estimation System 1 estimating the amount of bolus insulin, out of the basal insulin dose and the bolus insulin dose.

[0031] The portable device 50 receives blood glucose information and activity information from the blood glucose sensor 10 and activity sensor 20 attached to the patient. The portable device 50 transmits the received blood glucose information and activity information from the communication unit 53 to the estimation device 100.

[0032] When it is time to eat, the patient takes a picture of the food they plan to eat using the camera unit 54 of the portable device 50. The generation unit 55 of the portable device 50 is equipped with an application that calculates meal information from the image of the food. This calculation application may be, for example, an application downloaded from the cloud. The patient uses the calculation application to calculate the meal information of the photographed food. The calculation application calculates the total energy, carbohydrate, fat, and protein content of the food, for example, with the assistance of artificial intelligence. The calculation application may, for example, send the image of the food to a cloud server and receive the calculation results of the meal information from the cloud server. The communication unit 53 transmits the meal information to the estimation device 100.

[0033] The estimation device 100 acquires blood glucose information, activity information, and meal information from the portable device 50 (step S101). The estimation device 100 stores the acquired information in the storage device 102. When meal information is received, the estimation device 100 estimates an appropriate amount of additional insulin based on the blood glucose information, activity information, meal information, and estimation model 60 (step S102).

[0034] The estimation device 100 transmits the estimated amount of additional insulin to the portable device 50 (step S103). The portable device 50 displays the received information on the amount of additional insulin on the display unit 51. At this time, the portable device 50 also displays the patient's current blood glucose level and the amount of carbohydrates in the meal on the display unit 51.

[0035] The patient, referring to the amount of additional insulin displayed on the display unit 51 and taking into account their own measured blood glucose level, determines the actual amount of additional insulin to administer and administers the additional insulin (step S51). After that, the patient operates the control unit 52 and inputs the amount of additional insulin actually administered into the portable device 50 (step S52). The portable device 50 transmits the amount of additional insulin input via the operation of the control unit 52 to the estimation device 100.

[0036] The patient consumes the food captured by the imaging unit 54 (step S53). As the patient eats, their blood glucose and activity information fluctuates constantly. These fluctuations are detected by the blood glucose sensor 10 and the activity sensor 20, and these fluctuations are input to the estimation device 100 as blood glucose and activity information via the portable device 50.

[0037] The estimation device 100 updates the estimation model 60 by training the estimation model using the amount of additional insulin actually administered by the patient, the changes in blood glucose information and activity information after the patient administered the additional insulin, and the patient's dietary information (step S104).

[0038] <Graph showing an example of blood glucose level fluctuations> Figure 3 shows an example of how a patient's blood glucose levels fluctuate. In Figure 3, the horizontal axis represents the time of day, the left vertical axis represents blood glucose levels (mg / dL), and the right vertical axis represents the variance of the acquired blood glucose data. Graph G1 is created based on data obtained by observing a patient's blood glucose levels over several days, and shows the average fluctuation of blood glucose levels within a day. The bands extending above and below Graph G1 represent the variance of the patient's blood glucose data.

[0039] As shown in Figure 3, a patient's blood glucose level fluctuates significantly throughout the day due to factors such as meals. In particular, a patient's blood glucose level after a meal rises sharply to a peak value, then declines, eventually reaching a bottom value, and fluctuates smoothly until the next meal. Hereafter, in this specification, "peak value" refers to the maximum blood glucose level from the time a patient eats until the time of their next meal, and "bottom value" refers to the lowest blood glucose level from the time a patient eats until the time of their next meal.

[0040] In treatment, one of the goals is to keep the patient's blood glucose levels, which fluctuate constantly, within a target range (for example, 70 mg / dL to 180 mg / dL). For this reason, patients may be administered basal insulin to maintain stable blood glucose levels throughout the day and night, and bolus insulin to suppress rapid increases in blood glucose levels after meals.

[0041] In particular, rapid-acting insulin is used as bolus insulin to suppress the rapid rise in blood glucose levels after meals. Compared to long-acting insulin used as basal insulin, rapid-acting insulin can quickly suppress the rapid rise in blood glucose levels. However, if the dosage is too high, there is a risk of lowering blood glucose levels too far below the target range. In this case, the patient may reach a dangerous state known as hypoglycemia. Therefore, when estimating the appropriate amount of bolus insulin using machine learning, it is necessary to take into account not only the peak blood glucose level after the administration of bolus insulin, but also the subsequent bottom blood glucose level.

[0042] Furthermore, blood glucose levels fluctuate not only due to the food consumed, but also due to the patient's activity level, including blood pressure, body temperature, heart rate, and exercise. Therefore, when using machine learning to estimate the appropriate amount of additional insulin, it is important to consider not only the content of the meal the patient ate when administering additional insulin, but also various activity information obtained from the patient up to that point.

[0043] Therefore, in this embodiment, the appropriate amount of additional insulin is estimated by combining a first estimation (peak value estimation) that estimates the peak blood glucose value after the patient has eaten a meal without administering additional insulin, and a second estimation (dosage estimation) that estimates the amount of additional insulin needed to bring the patient's peak and bottom blood glucose values ​​within the target range, based on the first estimation.

[0044] Furthermore, in this embodiment, in the first and second estimations, the appropriate amount of additional insulin is estimated by considering not only the patient's blood glucose information and dietary information, but also the patient's activity information.

[0045] To enable such first and second estimations, in this embodiment, a first estimation model 61 used for the first estimation and a second estimation model 62 used for the second estimation are generated by machine learning using blood glucose information, meal information, and activity information. The first estimation model 61 and the second estimation model 62 thus generated can be used not only to estimate the amount of additional insulin but also the amount of basal insulin.

[0046] <First Estimation Model 61 (Peak Value Estimation Model)> Figure 4 is a diagram illustrating the first estimation model 61 (peak value estimation model). The first estimation model 61 is generated by machine learning using the learning unit 1011 included in the computing unit 101. The first estimation model 61 is generated, for example, by supervised learning.

[0047] The first estimation model 61 is generated with the aim of estimating the peak blood glucose level after a meal in a patient with certain blood glucose and activity levels who has not been administered insulin.

[0048] As shown in Figure 4, the datasets of "Input 1" and "Input 2" are used as training data for the first estimation model 61. The "Input 1" data consists of blood glucose information (ST1), activity information (ST2), and meal information (ST3). All of these ST1 to ST3 function as state variables. For ST1 and ST2, it is desirable to use waveform information acquired retrospectively for a certain period from when the patient eats a meal. For example, if the patient eats a meal at 6 a.m., blood glucose information and activity information acquired from the time of dinner the previous day up to that point may be used as ST1 and ST2, respectively.

[0049] The data in "Input 2" is the ground truth data for the data in "Input 1". This ground truth data is the peak blood glucose value Ap for patients (without insulin administration during meals) identified by ST1-ST3 in the dataset.

[0050] By preparing a large number of such datasets as training data and repeating machine learning, a trained first estimation model 61 is generated. The learning unit 1011 stores the generated first estimation model 61 in the storage device 102. It is desirable that the computing unit 101 be input not only training data collected from specific patients, but also training data collected from a large number of patients.

[0051] <Second Estimation Model 62 (Dosage Estimation Model)> Figure 5 is a diagram illustrating the second estimation model 62 (dosage estimation model). The second estimation model 62 is generated by machine learning using the learning unit 1011 included in the computing unit 101.

[0052] The second estimation model 62 is generated with the aim of estimating the amount of insulin needed to keep the peak and bottom blood glucose levels of a patient within a target range after a meal. The second estimation model 62 is generated, for example, by reinforcement learning.

[0053] In reinforcement learning, an agent (an acting entity) in a given environment observes its current state (state variables) and decides what action to take based on a policy for decision-making. By optimizing the policy, the choice of action is optimized. The environment changes dynamically as a result of the agent's actions, and the agent is given rewards based on reward criteria in response to these environmental changes. The agent repeats these actions and learns the action that yields the most rewards through a series of actions. In Figure 5, the current state is represented by a number of state variables "ST1~ST5", and the agent's actions are represented by "Action".

[0054] The learning unit 1011 includes a reward calculation unit 1013 and a function update unit 1014 related to reinforcement learning. Based on training data including "actions" and state variables "ST1~ST5", and reward criteria stored in the reward calculation unit 1013, the learning unit 1011 advances the learning of the second estimation model 62.

[0055] One example of a reinforcement learning algorithm is Q-learning. In Q-learning, the state s of the agent and the action a that the agent can choose in state s are used as independent variables, and a function Q(s,a) is learned that represents the value of the action a when action a is chosen in state s. In Q-learning, the optimal solution is to choose the action a that maximizes the value function Q in state s.

[0056] The "behavior" involves determining the optimal insulin dose X to administer to the patient based on the state variables "ST1~ST5". Reward criteria are established based on the relationship between the peak blood glucose value Bp and the bottom blood glucose value Bb after administering insulin dose X to the patient, and the target range. For example, as shown in Figure 3, the target range may be set to 70 mg / dL to 180 mg / dL.

[0057] The reward calculation unit 1013 calculates the reward based on the relationship between the peak value Bp and the bottom value Bb and the target range. More specifically, the reward calculation unit 1013 increases the reward when the peak value Bp and the bottom value Bb fall within the target range, and decreases the reward when the peak value Bp and the bottom value Bb fall outside the target range. The reward calculation unit 1013 may also calculate the reward by considering the distance between the peak value Bp and the upper limit of the target range, and the distance between the bottom value Bb and the lower limit of the target range.

[0058] The function update unit 1014 updates the function (behavioral value function) for determining the insulin amount X based on the reward calculated by the reward calculation unit 1013. This generates the second estimation model 62.

[0059] As shown in Figure 5, the state variables input as training data include blood glucose information (ST1), activity information (ST2), and meal information (ST3). These ST1-ST3 values ​​are also input in parallel to the trained first estimation model 61. Based on ST1-ST3, the first estimation model 61 estimates how high blood glucose levels will rise if a patient with the conditions identified by ST1-ST3 does not receive insulin after a meal. The estimation result of the first estimation model 61 is output as a peak value Ap.

[0060] The peak value Ap output from the first estimation model 61 is input to the learning unit 1011 as the state variable ST4. In this way, the second estimation model 62 is learned using the estimation result of the first estimation model 61 as one of its state variables.

[0061] The learning unit 1011 also receives the state variable ST5, which contains the peak blood glucose value Bp and the bottom blood glucose value Bb when the insulin amount X determined by "behavior" is administered to the patient.

[0062] The learning unit 1011 repeatedly trains a second estimation model 62 that infers insulin amount X from state variables using training data that includes "behavior" and "state variables". This generates a trained second estimation model 62. The learning unit 1011 stores the generated second estimation model 62 in the storage device 102. The estimation unit 1012 of the arithmetic unit 101 uses the trained first estimation model 61 and second estimation model 62, blood glucose information, meal information, and activity information to estimate the appropriate insulin dosage to keep the patient's blood glucose level within the target range.

[0063] <Procedure for generating the first estimated model 61> Figure 6 is a diagram illustrating the procedure for generating the first estimation model 61. First, the computing unit 101 acquires the training data necessary for generating the first estimation model 61 (step S1). The training data necessary for generating the first estimation model 61 is as explained using Figure 4. Next, the computing unit 101 performs a deep learning training process based on the training data (step S2). This estimates the peak blood glucose value when no additional insulin is administered to the patient.

[0064] Next, the computing unit 101 updates the first estimation model 61 in accordance with the processing in step S2 (step S3). Next, the computing unit 101 determines whether or not the termination condition has been met (step S4). For example, the computing unit 101 determines that the termination condition has been met when it has completed the learning process based on all the pre-prepared datasets.

[0065] If the arithmetic unit 101 determines that the termination condition is not met, it returns to step S1 and repeats the machine learning process. If the arithmetic unit 101 determines that the termination condition is met, it stores the generated first estimation model 61 in the storage device 102 (step S5) and terminates the process based on this flowchart.

[0066] <Procedure for generating the second estimated model 62> Figure 7 is a diagram illustrating the procedure for generating the second estimation model 62. First, the computing unit 101 acquires the training data necessary for generating the second estimation model 62 (step S21). The training data necessary for generating the second estimation model 62 is as explained using Figure 5. Next, the computing unit 101 determines the relationship between the peak value Bp and the bottom value Bb of blood glucose after insulin administration for the insulin amount X determined by "behavior," and the target range (step S22).

[0067] Next, the arithmetic unit 101 calculates the reward based on the reward criteria (step S23). Next, the arithmetic unit 101 determines whether the calculation result meets the criteria for increasing the reward (step S24). As already explained, the reward increases when the peak value Bp and bottom value Bb fall within the target range, and decreases when the peak value Bp and bottom value Bb fall outside the target range.

[0068] The calculation unit 101 increases the reward if the calculation result matches the criteria for increasing the reward (step S25), and decreases the reward if it does not match (step S26). Next, the calculation unit 101 updates the function (behavioral value function) for determining the amount of insulin X (step S27).

[0069] Next, the arithmetic unit 101 determines whether or not the termination condition has been met (step S28). For example, the arithmetic unit 101 determines that the termination condition has been met when it has completed the learning process based on a predetermined learning period.

[0070] If the arithmetic unit 101 determines that the termination condition is not met, it returns to step S21 and repeats the machine learning process. If the arithmetic unit 101 determines that the termination condition is met, it stores the generated second estimation model 62 in the storage device 102 (step S29) and terminates the process based on this flowchart.

[0071] The methods for generating the first estimated model 61 and the second estimated model 62 have been described above. Here, an example was described in which the first estimated model 61 and the second estimated model 62 are generated by the computing unit 101, which is part of the estimation device 100. However, the first estimated model 61 and the second estimated model 62 may be generated using a computer separate from the estimation device 100, and the generated trained models may be stored in the storage device 102 of the estimation device 100.

[0072] <Estimation of additional insulin dose and basal insulin dose> Figure 8 illustrates the difference in the timeframes for acquiring information used to estimate the bolus insulin dose and the basal insulin dose. Here, we will explain using the example of a patient who receives basal insulin once a day before bedtime and basal insulin three times a day with each meal.

[0073] As shown in Figure 8, when estimating the amount of additional insulin, the estimation device 100 uses the patient's blood glucose and activity information obtained from the time of the previous meal to the time of the current meal, and meal information based on the meal to be consumed. Specifically, when estimating the amount of additional insulin to be administered to the patient at breakfast, the estimation device 100 uses the patient's blood glucose and activity information obtained from the time of the previous evening's dinner to the current breakfast, and the meal information for breakfast. When estimating the amount of additional insulin to be administered to the patient at lunchtime, the estimation device 100 uses the patient's blood glucose and activity information obtained from the time of breakfast to the current lunch, and the meal information for lunch. When estimating the amount of additional insulin to be administered to the patient at dinnertime, the estimation device 100 uses the patient's blood glucose and activity information obtained from the time of lunch to the current dinner, and the meal information for dinner.

[0074] In contrast, when estimating the basal insulin dose, the estimation device 100 uses the patient's blood glucose information, activity information, and meal information obtained within 24 hours prior to the administration of basal insulin. For example, in the example shown in Figure 8, when estimating the basal insulin dose, the estimation device 100 uses meal information for breakfast, lunch, and dinner. Thus, in this embodiment, the acquisition period for the information used for estimation differs depending on whether the additional insulin dose is being estimated or the basal insulin dose is being estimated.

[0075] <Estimation of additional insulin dose and basal insulin dose> Figure 9 is a flowchart illustrating the insulin dosage estimation procedure using the first estimation model 61 and the second estimation model 62. The processing based on this flowchart is performed by the estimation device 100, and more specifically by the calculation unit 101 of the estimation device 100.

[0076] First, the computing device 101 determines whether or not meal information has been acquired (step S31). Meal information is transmitted from the portable device 50 to the estimation device 100 when the patient takes a picture of the food they plan to eat using the portable device 50. If the computing device 101 has acquired meal information, it determines that it is time for the patient to receive additional insulin.

[0077] Therefore, when the computing unit 101 obtains meal information, it proceeds to the process of estimating the amount of additional insulin. To do this, the computing unit 101 reads the patient's blood glucose information and activity information obtained from the time of the last meal to the present from the storage device 102 (step S32).

[0078] Next, the computing device 101 estimates the amount of additional insulin using the meal information acquired in step S31, the blood glucose information and activity information read out in step S32, and the estimation models (first estimation model 61, second estimation model 62) (step S33).

[0079] Next, the computing unit 101 outputs the estimation result to the patient's portable device 50 (step S34).

[0080] If the calculation unit 101 determines in step S31 that it has not obtained meal information, it determines whether or not it has received a request to estimate the basal insulin amount (step S35). The request to estimate the basal insulin amount is input from the portable device 50 to the estimation device 100 in response to the operation of the portable device 50 by the patient.

[0081] If the arithmetic unit 101 has not received a request to estimate the basal insulin dose, it completes the processing based on this flowchart. If the arithmetic unit 101 has received a request to estimate the basal insulin dose, it reads the patient's meal information, blood glucose information, and activity information obtained within the last 24 hours from the storage device 102 (step S36). The arithmetic unit 101 estimates the basal insulin dose using the read meal information, blood glucose information, and activity information, as well as the estimation models (first estimation model 61, second estimation model 62) (step S37). Next, the arithmetic unit 101 outputs the estimation result to the patient's portable device 50 (step S34) and completes the processing based on this flowchart. <Example 1> Next, I will explain Modification 1. Modification 1 concerns the first estimated model 611, which is a modified version of the first estimated model 61 described earlier, and the second estimated model 621, which is a modified version of the second estimated model 62.

[0082] Figure 10 shows a modified version of the first estimation model 61. The first estimation model 61, as explained earlier using Figure 4, is generated with the aim of estimating the peak blood glucose level after a meal in a patient with certain blood glucose and activity levels who has been given insulin. The first estimation model 611 is generated with the aim of estimating the peak blood glucose level Ap' after a meal in a patient with certain blood glucose and activity levels who has been given insulin.

[0083] As shown in Figure 10, the datasets "Input 1" and "Input 2" are used as training data for the first estimation model 611. The "Input 1" data includes blood glucose information (ST1), activity information (ST2), and meal information (ST3), similar to the training data for the first estimation model 61 described earlier. In addition to these, the "Input 1" data for the first estimation model 611 also includes the amount of insulin administered (ST6).

[0084] The data in "Input 2" is the ground truth data for the data in "Input 1". This ground truth data is the peak postprandial blood glucose value Ap' of patients identified by ST1-ST3 and ST6 in the dataset.

[0085] By preparing a large number of such datasets as training data and repeating machine learning, a trained first estimation model 611 is generated. The learning unit 1011 stores the generated first estimation model 611 in the storage device 102.

[0086] Figure 11 shows a modified version of the second estimation model 62. Figure 11 shows the second estimation model 621, which is generated using the first estimation model 611, as a modified version of the second estimation model 62.

[0087] When generating the second estimation model 621 using the first estimation model 611, the amount of insulin administered (ST6), in addition to ST1 to ST3, is input to the learning unit 1011 and the first estimation model 61. Furthermore, ST5 and ST4', which is the estimation result of the first estimation model 611, are input to the learning unit 1011. Here, ST5 is the same state variable input when generating the second estimation model 62, and represents the peak value Bp and bottom value Bb of blood glucose when the patient is administered the amount of insulin X determined by "behavior". ST4' is the peak value Ap' of blood glucose after the patient has eaten after receiving insulin.

[0088] The learning unit 1011 repeatedly trains a second estimation model 621 that infers insulin amount X from state variables using training data including "behavior" and "state variables". This generates a trained second estimation model 621. The learning unit 1011 stores the generated second estimation model 621 in the storage device 102. The estimation unit 1012 of the arithmetic unit 101 uses the trained first estimation model 611 and second estimation model 621, along with blood glucose information, meal information, and activity information, to estimate the appropriate insulin dosage to keep the patient's blood glucose level within the target range. <Modification 2> Next, we will explain Modification 2. Modification 2 concerns an example in which the third estimation model 63 is further applied to Modification 1, which was explained earlier.

[0089] Figure 12 shows the third estimation model 63. The first estimation model 611, which relates to the modified example 1 described earlier using Figure 10, is generated with the aim of estimating the peak blood glucose value Ap' after a meal in a patient with certain blood glucose and activity levels who has been administered insulin. In contrast, the third estimation model 63 is generated with the aim of estimating the bottom blood glucose value Ab' after a meal in a patient with certain blood glucose and activity levels who has been administered insulin. Therefore, by utilizing the estimation results of the first estimation model 611 and the third estimation model 63, it is possible to estimate the peak blood glucose value Ap' and the bottom blood glucose value Ab' after a meal in a patient with certain blood glucose and activity levels who has been administered insulin.

[0090] As shown in Figure 12, the datasets of "Input 1" and "Input 2" are used as training data for the third estimation model 63. The "Input 1" data, like the training data for the first estimation model 611 described earlier, includes blood glucose information (ST1), activity information (ST2), meal information (ST3), and administered insulin amount (ST6).

[0091] The data in "Input 2" is the ground truth data for the data in "Input 1". This ground truth data is the bottom value Ab' of postprandial blood glucose for patients identified by ST1-ST3 and ST6 in the dataset.

[0092] By preparing a large number of such datasets as training data and repeating machine learning, a trained third estimation model 63 is generated. The learning unit 1011 stores the generated third estimation model 63 in the storage device 102.

[0093] Figure 13 shows a modified version of the second estimation model 62. Figure 11 shows the second estimation model 622, which is generated using the first estimation model 611 and the third estimation model 63 as a modified version of the second estimation model 62.

[0094] The second estimation model 622 shown in Figure 13 is a modified version of the second estimation model 621 shown in Figure 11. When creating the second estimation model 622, in addition to the estimation results of the first estimation model 611, the estimation result ST4'' of the third estimation model 63 is input to the learning unit 1011. The estimation result ST4'' of the third estimation model 63 is the bottom blood glucose value Ab' after the patient has been given insulin and eaten.

[0095] The learning unit 1011 repeatedly trains the second estimation model 622, which infers the insulin amount X from the state variables, using training data that includes "behavior" and "state variables". This generates a trained second estimation model 622. The learning unit 1011 stores the generated second estimation model 622 in the storage device 102. The estimation unit 1012 of the arithmetic unit 101 estimates the appropriate insulin dosage to keep the patient's blood glucose level within the target range, using the trained first estimation model 611, the second estimation model 622, and the third estimation model 63, as well as blood glucose information, meal information, and activity information.

[0096] <Other variations> Other variations will be described. While 70 mg / dL to 180 mg / dL was given as an example of a target blood glucose range, this disclosure does not limit the target range to this range, and other numerical ranges may be used to define the target range. Furthermore, the estimation device 100 may be configured to allow users, such as physicians, to set the target range.

[0097] In the update step S104 of the estimation model 60 shown in Figure 2, the estimation device 100 identifies the peak and bottom blood glucose values ​​after the patient has eaten a meal based on the blood glucose information, and further advances the reinforcement learning of the learned second estimation model 62. At this time, if there is a difference between the amount of insulin actually administered by the patient and the amount of insulin presented to the patient by the estimation device 100, the reinforcement learning may be advanced taking that difference into account.

[0098] As described above, the estimation device 100 according to this embodiment estimates and provides to the patient the appropriate insulin dosage to bring the patient's blood glucose level within the target range, based on blood glucose information, meal information, and activity information. Therefore, the estimation device 100 according to this embodiment allows the patient to easily understand the appropriate insulin dosage. This embodiment can be adapted to frequent insulin injection therapy.

[0099] (Appearance) The following describes aspects of this embodiment.

[0100] (1) The estimation device according to paragraph 1 is an estimation device (100) for estimating the amount of insulin to administer in order to keep a patient's blood glucose level within a target range (70 mg / dL to 180 mg / dL), and comprises an acquisition unit (103) for acquiring state variables (ST1 to ST3) including blood glucose information indicating the patient's blood glucose level, patient activity information different from the blood glucose information, and dietary information regarding the food the patient consumes; an estimation unit (1012) for estimating the amount of insulin to administer based on the state variables and the target range using a dosage estimation model (62); and an output unit (104) for outputting the amount of insulin to administer.

[0101] (2) In the estimation device according to paragraph 1, the estimation unit (1012) is capable of estimating the insulin dosage for each meal of the patient (S31), and when the patient has a second meal after the first meal, the estimation unit (1012) estimates the insulin dosage using meal information related to the second meal, as well as blood glucose information and activity information acquired by the acquisition unit during the period from the first meal to the second meal (S33, Figure 8).

[0102] (3) If the estimation unit (1012) relating to paragraph 1 or 2 receives an instruction to estimate the insulin dosage at a time different from the patient's mealtime (S35), it estimates the insulin dosage using meal information, blood glucose information, and activity information acquired by the acquisition unit over a predetermined period of time retrospectively from the time the estimation instruction was received (S37, Figure 8).

[0103] (4) In the estimation device (100) relating to paragraph 1 or 2, the state variable further includes the first peak value (ST4) of the patient's blood glucose, which is estimated when the patient does not receive insulin after consuming food based on dietary information.

[0104] (5) The estimation unit (1012) relating to paragraph 4 estimates the first peak value using the peak value estimation model (61), which is trained to estimate the first peak value based on blood glucose information, activity information, and dietary information (Figure 4), and the estimation unit (1012) inputs the estimation result of the peak value estimation model (61) as one of the state variables into the dosage estimation model (62) (Figure 5).

[0105] (6) In any one of paragraphs 1 to 3, the state variable further includes the patient's second peak blood glucose value (Figure 11, Figure 13: ST4') which is estimated when the patient is administered insulin after consuming food based on dietary information.

[0106] (7) In paragraph 6, the state variable further includes the estimated bottom blood glucose level of the patient when insulin is administered after the patient has consumed food based on dietary information (Figure 13: ST4'').

[0107] (8) The estimation method relating to paragraph 8 is an estimation method for estimating the amount of insulin to be administered to bring a patient's blood glucose level into a target range using a computer (100), and includes the steps of: acquiring state variables including blood glucose information indicating the patient's blood glucose level, patient activity information different from the blood glucose information, and dietary information regarding the food the patient consumes (S101); estimating the amount of insulin to be administered based on the state variables and the target range using a dosage estimation model (S102, S33, S37); and outputting the amount of insulin to be administered (S103, S34).

[0108] (9) The estimation program relating to paragraph 9 is an estimation program (71) for estimating the amount of insulin to administer in order to bring a patient's blood glucose level into a target range, and causes a computer (100) to perform the following steps: acquire state variables including blood glucose information indicating the patient's blood glucose level, patient activity information different from the blood glucose information, and dietary information regarding the food the patient consumes (S101); estimate the amount of insulin to administer based on the state variables and the target range using a dosage estimation model (S102, S33, S37); and output the amount of insulin to administer (S103, S34).

[0109] (10) The estimation system (1) relating to paragraph 10 comprises an estimation device (100) relating to paragraph 1, a blood glucose sensor (10) for detecting blood glucose information from a patient, an activity sensor (20) for detecting activity information from a patient, and a portable device (50) carried by the patient that collects blood glucose information and activity information from the blood glucose sensor and the activity sensor. The portable device (50) has a shooting unit (54) for taking images of food consumed by the patient, a generation unit (55) for generating meal information based on the images taken by the shooting unit, a communication unit (53) for communicating with the estimation device via a network, and a display unit (51). The communication unit (51) transmits blood glucose information, activity information, and meal information to the estimation device (Figures 1 and 2), the output unit (104) transmits insulin dosage to the portable device (Figures 1 and 2), and the display unit (51) displays the insulin dosage on the display unit (Figures 1 and 2).

[0110] The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of this disclosure is indicated by the claims rather than by the description of the embodiments above, and all modifications within the meaning and scope equivalent to the claims are intended to be included. [Explanation of symbols]

[0111] 1 Estimation system, 10 Blood glucose sensor, 20 Activity level sensor (wearable device), 50 Portable device, 51 Display unit, 52 Operation unit, 53 Communication unit, 54 Image capture unit, 55 Generation unit, 60 Estimation model, 61, 611 First estimation model, 62, 621, 622 Second estimation model, 63 Third estimation model, 71 Estimation program, 90 Network, 100 Estimation device, 101 Processing unit, 102 Memory device, 103 Input interface (acquisition unit), 104 Output interface (output unit), 105 Bus, 1011 Learning unit, 1012 Estimation unit, 1013 Reward calculation unit, 1014 Function update unit.

Claims

1. An estimation device for estimating the amount of insulin to administer in order to bring a patient's blood glucose level within a target range, An acquisition unit that acquires state variables including blood glucose information indicating the patient's blood glucose level, activity information of the patient different from the blood glucose information, and dietary information regarding the food consumed by the patient. An estimation unit that estimates the insulin dosage based on the state variables and the target range using a dosage estimation model, The system includes an output unit that outputs the aforementioned insulin dosage, The estimation unit is capable of estimating the insulin dosage for each meal of the patient. The estimation unit is an estimation device that, when the patient has a second meal after a first meal, estimates the insulin dosage using the meal information related to the second meal, as well as the blood glucose information and activity information acquired by the acquisition unit during the period from the first meal to the second meal.

2. An estimation device for estimating the amount of insulin to administer in order to bring a patient's blood glucose level within a target range, An acquisition unit that acquires state variables including blood glucose information indicating the patient's blood glucose level, activity information of the patient different from the blood glucose information, and dietary information regarding the food consumed by the patient. An estimation unit that estimates the insulin dosage based on the state variables and the target range using a dosage estimation model, The system includes an output unit that outputs the aforementioned insulin dosage, If the estimation unit receives an instruction to estimate the insulin dosage at a time different from the patient's mealtime, An estimation device that estimates the insulin dosage using the meal information, blood glucose information, and activity information acquired by the acquisition unit over a predetermined period of time, retrospectively from the time the estimation instruction is received.

3. An estimation device for estimating the amount of insulin to administer in order to bring a patient's blood glucose level within a target range, An acquisition unit that acquires state variables including blood glucose information indicating the patient's blood glucose level, activity information of the patient different from the blood glucose information, and dietary information regarding the food consumed by the patient. An estimation unit that estimates the insulin dosage based on the state variables and the target range using a dosage estimation model, The system includes an output unit that outputs the aforementioned insulin dosage, The estimation device further includes, as state variables, a first peak value of the patient's blood glucose estimated when the patient does not receive insulin after consuming food based on the dietary information.

4. An estimation device for estimating the amount of insulin to administer in order to bring a patient's blood glucose level within a target range, An acquisition unit that acquires state variables including blood glucose information indicating the patient's blood glucose level, activity information of the patient different from the blood glucose information, and dietary information regarding the food consumed by the patient. An estimation unit that estimates the insulin dosage based on the state variables and the target range using a dosage estimation model, The system includes an output unit that outputs the aforementioned insulin dosage, The estimation device further includes, as state variables, a second peak value of the patient's blood glucose, which is estimated when the patient is administered insulin after consuming food based on the dietary information.

5. The estimation unit estimates the first peak value using a peak value estimation model. The peak value estimation model is trained to estimate the first peak value based on the blood glucose information, the activity information, and the meal information. The estimation device according to claim 3, wherein the estimation unit inputs the estimation result of the peak value estimation model as one of the state variables into the dosage estimation model.

6. The estimation device according to claim 4, wherein the state variable further includes the patient's bottom blood glucose level, which is estimated when the patient is administered insulin after ingesting food based on the dietary information.

7. A method for estimating the amount of insulin to administer to a patient to bring their blood glucose level within a target range, using a computer. The steps include obtaining state variables including blood glucose information indicating the patient's blood glucose level, activity information of the patient that is different from the blood glucose information, and dietary information regarding the food the patient consumes. A step of estimating the insulin dose based on the state variable and the target range using a dose estimation model, The step includes outputting the insulin dose, The aforementioned estimation step is, The steps include: estimating the insulin dosage for each meal of the patient; An estimation method comprising the step of estimating the insulin dosage when the patient has a second meal after a first meal, using the meal information relating to the second meal, and the blood glucose information and activity information obtained by the acquisition step during the period from the first meal to the second meal.

8. An estimation program for estimating the insulin dosage required to bring a patient's blood glucose level within a target range, On the computer, The steps include obtaining state variables including blood glucose information indicating the patient's blood glucose level, activity information of the patient that is different from the blood glucose information, and dietary information regarding the food the patient consumes. A step of estimating the insulin dose based on the state variable and the target range using a dose estimation model, The step of outputting the insulin dosage is performed. The aforementioned estimation step is, The steps include: estimating the insulin dosage for each meal of the patient; An estimation program comprising the step of estimating the insulin dosage when the patient has a second meal after a first meal, using the meal information relating to the second meal, and the blood glucose information and activity information obtained by the acquisition step during the period from the first meal to the second meal.

9. The estimation device according to claim 1, A blood glucose sensor for detecting the blood glucose information from the patient, An activity level sensor that detects the activity information from the patient, The device comprises a portable device carried by the patient, which collects blood glucose information and activity information from the blood glucose sensor and the activity sensor, The portable device includes a camera unit that takes images of the food consumed by the patient, A generation unit that generates the meal information based on the image captured by the aforementioned imaging unit, A communication unit that communicates with the estimation device via a network, It has a display unit, The communication unit transmits the blood glucose information, activity information, and meal information to the estimation device. The output unit transmits the insulin dosage to the portable device. The display unit is an estimation system that displays the insulin dosage on the display unit.

Images