Artificial pancreas simulator

Abstract

The present invention relates to an artificial pancreas simulator and, more particularly, may comprise an infusion pump for administering insulin to a user, a blood glucose meter for measuring the user's blood glucose level at predetermined intervals, a user terminal for communicating with the infusion pump and the blood glucose meter, and an organ simulator for simulating bodily functions of the user related to changes in the user's blood glucose level.

Description

Artificial pancreas simulator

[0001] The present invention relates to an artificial pancreas simulator.

[0002] The artificial pancreas is a closed-loop medical device designed to automate and optimize insulin delivery management for diabetic patients. However, the security of this device has been managed by security policies that follow fixed algorithms and is not managed adaptively in response to changes in the external environment. Consequently, because the security system responding to changes in the user's body or external attacks does not change chronologically and relies on fixed algorithms, it fails to provide appropriate responses to risk factors that change over time.

[0003] The present invention aims to solve the problems of the aforementioned prior art. The present invention provides an adaptive security system for a closed-loop medical device capable of responding to the external environment of the closed-loop medical device, which changes over time including the user's physical condition and attacks such as hacking, by detecting and updating operational errors of the closed-loop medical device over time and applying enhanced security policies.

[0004] As a technical means for achieving the above-mentioned technical problem, an artificial pancreas simulator according to one embodiment of the present invention comprises an infusion pump for injecting insulin into a user, a blood glucose meter for measuring the user's blood glucose at a predetermined interval, a user terminal configured to communicate with the infusion pump and the blood glucose meter, and an organ simulator for simulating the user's physical functions related to changes in the user's blood glucose, wherein the user terminal comprises an infusion control unit for determining the insulin provided to the user based on physiological information transmitted from the infusion pump and the blood glucose meter, and an error determination unit for determining an operation error of the artificial pancreas based on the physiological information, and the organ simulator may provide at least simulated physiological information according to the user's physical functions simulated based on the physiological information to the error determination unit.

[0005] In addition, a safety management system for a closed-loop medical device according to one embodiment of the present invention forms a closed-loop medical device comprising an infusion pump for injecting insulin into a user, a blood glucose meter for measuring the user's blood glucose at a predetermined interval, and a user terminal configured to communicate with the infusion pump and the blood glucose meter, wherein the user terminal may include an infusion control unit comprising an artificial intelligence control module that determines the insulin to be provided to the user based on physiological information transmitted from the infusion pump and the blood glucose meter and provides insulin through an artificial intelligence model, and a manual control module that provides insulin according to a pre-stored algorithm or user control, a malfunction detection unit that detects a malfunction of the artificial intelligence control module based on the physiological information, and a malfunction notification unit that provides a notification regarding the content of the malfunction detected by the malfunction detection unit.

[0006] According to the means for solving the problem of the present invention described above, the present invention has the effect of providing an adaptive security system for a closed-loop medical device capable of responding to the external environment of the closed-loop medical device that changes over time, including the user's physical condition and attacks such as hacking, by detecting and updating operational errors of the closed-loop medical device over time and applying an enhanced security policy.

[0007] FIG. 1 is a schematic diagram illustrating a closed-loop medical device (100) according to one embodiment of the present invention.

[0008] FIG. 2 is a schematic diagram illustrating an adaptive security system for a closed-loop medical device according to one embodiment of the present invention.

[0009] FIG. 3 is a block diagram of a user terminal (130) according to one embodiment of the present invention, a block diagram of a malfunction detection unit (133) and a malfunction notification unit (134) according to one embodiment of the present invention, and a schematic block diagram of a closed-loop medical device control unit (135) according to one embodiment of the present invention.

[0010] FIG. 4 is a schematic diagram illustrating the notification content displayed on a user terminal (130) according to one embodiment of the present invention.

[0011] FIG. 5 is a block diagram of a security model generation device (300) according to one embodiment of the present invention.

[0012] Figure 6 is a schematic diagram illustrating the enhancement of physiological information.

[0013] FIG. 7 is a schematic diagram illustrating an adaptive security system for a closed-loop medical device according to another embodiment of the present invention.

[0014] FIG. 8 is a schematic diagram illustrating the simulation of a user's blood glucose-related time-series information in a long-term simulator (600) according to one embodiment of the present invention.

[0015] Throughout the specification of the present invention, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0016] First, referring to FIG. 1, the closed-loop medical device (100) of the present invention may include an infusion pump (110) for injecting insulin into a user, a blood glucose meter (120) for measuring the user's blood glucose at a predetermined interval, and a user terminal (130) configured to communicate with the infusion pump and the blood glucose meter. That is, the infusion pump (110), the blood glucose meter (120), and the user terminal (130) may form the closed-loop medical device (100).

[0017] In one embodiment, a closed-loop medical device (100) may be understood to monitor the user's physiological information to maintain the user's physiological information within a set range and to provide medical measures to the user based on said physiological information. The closed-loop medical device may continuously or periodically measure and collect specific physiological information, such as the patient's blood glucose level, heart rate, blood pressure, or other related indicators, and perform real-time processing through an algorithm or control system. The closed-loop medical device may determine appropriate treatment measures based on the processed physiological information and automatically adjust therapeutic interventions. Additionally, the closed-loop medical device may have a feedback loop to continuously monitor the effect of the therapeutic intervention, and this feedback may be understood as the medical measures provided to the user being improved and adjusted in real time.

[0018] In this specification, physiological information may include information transmitted from an infusion pump (110) and a blood glucose meter (120), such as the user's current blood glucose and the user's blood glucose change pattern over time. That is, physiological information may be understood as information related to the user's blood glucose transmitted from a closed-loop medical device (100) for determining the insulin delivery value. Additionally, supplementary information may be understood as information related to the user's health status transmitted from an external device (500), such as a wearable device, other than the closed-loop medical device (100), or stored in a user terminal (130). For example, supplementary information may be understood as information including the user's blood pressure and exercise information stored in the user terminal, which affects the user's blood glucose change and insulin delivery.

[0019] The infusion pump (110) can transmit and receive data with the user terminal (130) via wireless communication. The infusion pump (110) can also be understood as functioning as a standalone device having its own built-in controller. For example, data transmitted from the infusion pump (110) to the user terminal (130) may include information such as insulin delivery data, blood glucose information, baseline, bolus, insulin-to-carbohydrate ratio, and insulin sensitivity factor.

[0020] A blood glucose meter (120) can be attached to a user's body to measure the user's internal blood glucose. In one embodiment, the blood glucose meter may be equipped as a continuous glucose meter (CGM) and may measure the user's blood glucose at a predetermined interval, preferably in real time. Here, real-time measurement can be understood as measuring blood glucose at intervals of within a few seconds. Physiological information related to the user's blood glucose measured by the blood glucose meter (120) may be transmitted to a user terminal (130).

[0021] The user terminal (130) can receive and verify physiological information including blood glucose information. To this end, after the blood glucose meter is started to operate, a process of establishing a communication connection between the blood glucose meter and the user terminal may be performed first. Typically, the user terminal and the blood glucose meter can perform communication after being paired using Bluetooth communication.

[0022] The infusion pump (110), blood glucose meter (120), and user terminal (130) can communicate with each other via a network. The network refers to a connection structure that enables information exchange between each node, such as terminals and servers, and is not limited to any specific type.

[0023] The user terminal (130) can be any type of wireless communication device.

[0024] Referring to FIG. 2, an artificial pancreas simulator (1) according to one embodiment of the present invention may be disclosed. The artificial pancreas simulator (1) can determine an operation error of a closed-loop medical device, i.e., an artificial pancreas, through an adaptive security policy in response to risk factors for a closed-loop medical device that changes over time.

[0025] Here, an operation error can be understood to refer to software bugs, hardware bugs, hacking, etc., which affect the malfunction of blood glucose or insulin measurement and make normal operation of the artificial pancreas impossible. An operation error may be caused by changes in the user's physical condition, attacks from the outside such as hacking, etc. The artificial pancreas simulator (1) can respond to the external environment of the closed-loop medical device that changes over time (which can be understood to include the user's physical condition, attacks such as hacking) by detecting and updating operation errors of the closed-loop medical device over time and applying enhanced security policies.

[0026] The artificial pancreas simulator (1) may include the above-described closed-loop medical device (100), a gateway (200) that transmits physiological information transmitted from an infusion pump and a blood glucose meter as a closed-loop medical device to a security model generating device (300), and a security model generating device (300) that generates a security model for determining an operation error of the artificial pancreas from the physiological information transmitted from the infusion pump and the blood glucose meter.

[0027] Referring to FIG. 3, the user terminal (130) of the closed-loop medical device (100) can form the closed-loop medical device (100) in conjunction with the infusion pump (110) and the blood glucose meter (120) to provide insulin to the user based on physiological information transmitted from the infusion pump and the blood glucose meter. The user terminal (130) may include an infusion control unit (131), an error determination unit (132), a malfunction detection unit (133), a malfunction notification unit (134), and a closed-loop medical device control unit (135).

[0028] The infusion control unit (131) can be understood as a configuration that determines the insulin provided to the user based on physiological information transmitted from the infusion pump and the blood glucose meter. Here, the insulin provision determination can be understood as a temporal and quantitative determination of the insulin provided to the user, including the amount of insulin provided per dose through the infusion pump (110) and the insulin provision cycle. That is, the infusion control unit (131) determines the insulin provided to the user based on the user's physiological information, for example, the user's current blood glucose and the user's blood glucose change pattern over time, and transmits a signal to the infusion pump (110), and the infusion pump (110) can provide the insulin into the user's body based on the signal from the infusion control unit (131).

[0029] In one embodiment, the injection control unit (131) may include an artificial intelligence control module (1311) that provides insulin through an artificial intelligence model, and a manual control module (1312) that provides insulin according to a pre-stored algorithm or user control.

[0030] The artificial intelligence control module (1311) can determine the insulin to be provided to the user based on an artificial intelligence model trained to derive the amount and / or cycle of insulin provision based on the user's blood glucose changes measured by the blood glucose meter (120). For example, the artificial intelligence control module (1311) can receive blood glucose information from the blood glucose meter (120) and determine the amount of insulin to be administered based on the amount of carbohydrate intake. In addition, the injection rate of insulin can be adjusted according to the blood glucose concentration measured by the blood glucose meter (120).

[0031] Meanwhile, the manual control module (1312) can determine the insulin provided to the user based on a preset algorithm or user control. The manual control module can be understood as the user manually inputting and controlling the delivery of insulin to a certain degree. Although the main feature of the artificial pancreas is a closed-loop automated system that coordinates insulin delivery based on continuous blood glucose monitoring, manual intervention may be required if an unexpected malfunction occurs in the artificial intelligence control module (1311). That is, it may be particularly useful in situations where the automated closed-loop medical device (100) cannot fully capture the user's physical condition or a specific external environment, or when the user wishes to make a choice different from the decision of the closed-loop medical device (100) for a specific purpose. For example, the user can operate the manual control module (1312) to input information regarding planned meals, exercise, or other factors that may affect blood glucose. Control by the manual control module (1312) may vary depending on the artificial pancreas or the user's physical condition.

[0032] The error determination unit (132) described above may be configured to receive a security model generated by the security model generation device (300) described later, and to determine an operation error of the artificial pancreas from the physiological information based on the security model. In one embodiment, the security model distributed to the user terminal (130) and installed or configured in the error determination unit (132) is learned by an artificial intelligence model and can detect operation errors of the artificial pancreas (closed-loop medical device), such as hacking, hardware errors, and communication errors, based on physiological information. As described above, an operation error can be understood as referring to something that makes normal operation of the artificial pancreas impossible, and since it can be caused by changes in the user's physical condition, attacks from the outside such as hacking, etc., the error determination unit (132) can detect operation errors from real-time signals containing physiological information.

[0033] In one embodiment, the error determination unit (132) may classify the type of operation error of the artificial pancreas from physiological information transmitted from the artificial pancreas according to the distributed security model. As described below, the type of operation error may be classified according to at least one of criteria including the cause of the operation error, the configuration of the artificial pancreas where the operation error occurred, and the lethality of the operation error. In a preferred embodiment, the error determination unit (132) may classify the type of operation error based on the lethality of the operation error according to the detection of the security model. Here, lethality may be classified based on how lethal the operation error of the artificial pancreas is to the user's body. For example, the type of operation error may be classified based on lethality into an operation error that may simply cause a temporary abnormal state of blood sugar, an operation error that may cause discomfort or pain to the user, and an operation error that may cause an emergency situation such as shock, but is not limited thereto.

[0034] Based on the operation error determination of the error determination unit (132), the user terminal (130) can generate a notification based on the criticality of the operation error determined by the security model.

[0035] Referring to FIG. 3, the malfunction detection unit (133) can detect a malfunction of the artificial intelligence control module (1311) based on physiological information. Here, unlike an operation error of the artificial pancreas, the malfunction of the artificial intelligence control module (1311) can be understood as a malfunction related to the insulin supply value determined by the artificial intelligence control module (1311) according to an artificial intelligence model trained to derive the amount and / or cycle of insulin supply based on the user's blood sugar change.

[0036] That is, an operation error of the artificial pancreas refers to an abnormal operation of the artificial pancreas related to insulin delivery or blood glucose analysis caused by hacking of the user terminal (130), hardware communication errors or software errors forming the closed-loop medical device (100). Meanwhile, a malfunction of the artificial intelligence control module (1311) refers to a situation where a normal insulin delivery value is not derived due to reasons such as insufficient training data for the artificial intelligence model or overfitting of the training data. If a configuration such as the malfunction detection unit (133) of the present invention, which determines the malfunction, is not provided, such a malfunction of the artificial intelligence control module (1311) cannot be detected, and the user's health may deteriorate due to incorrect insulin delivery.

[0037] The above malfunction detection unit (133) may include a detection module (1331), a switching module (1332), and a classification module (1333).

[0038] The detection module (1331) is configured to detect a malfunction of the artificial intelligence control module (1311) based on physiological information transmitted from the closed-loop medical device (100), and preferably, the malfunction can be detected from the insulin delivery value determined by the artificial intelligence control module (1311) regarding the physiological information according to a set insulin injection-related algorithm. The insulin injection-related algorithm may be a judgment criterion including at least one of the insulin delivery amount per dose, the insulin delivery cycle, and the insulin delivery amount relative to the user's blood glucose.

[0039] In one embodiment, the detection module (1331) detects an unexpected malfunction of the artificial intelligence control module (1311) and can determine whether the insulin delivery value determined by the artificial intelligence control module (1311) deviates from the normal range in relation to the user's blood sugar. The insulin delivery value may be incorrectly determined due to a malfunction in the artificial intelligence control module (1311), for example, when the amount of insulin delivered per dose is excessive or insufficient, when the delivery cycle is too short or too long relative to the user's health condition, or when the amount of insulin delivered is excessive or insufficient relative to the user's blood sugar. The detection module (1331) can detect whether the insulin delivery value was incorrectly determined due to a malfunction in the artificial intelligence control module (1311) based on the user's blood sugar status and insulin-related judgment criteria. The judgment criteria may be stored in the user terminal (130) or transmitted from an external database (400).

[0040] The above switching module (1332) can switch the operation of the injection control unit from operation by the artificial intelligence control module to operation by the manual control module. That is, if the detection module (1331) detects a malfunction of the artificial intelligence control module, the switching module (1332) can switch the insulin delivery operation of the closed-loop medical device (100) from operation by the artificial intelligence control module (1311) to operation by the manual control module (1312).

[0041] Additionally, the classification module (1333) can classify the malfunction of the artificial intelligence control module based on lethality. As described above, lethality refers to how lethal it is to the user's body, and the classification module (1333) can classify the malfunction of the artificial intelligence control module based on how lethal it is to the user's body.

[0042] According to the lethality classification, if the lethality of the malfunction of the artificial intelligence control module (1311) is greater than a set standard, the switching module (1332) may switch the operation of the injection control unit (131) to operation by the manual control module (1312). For example, lethality greater than the set standard may be a level at which it is determined that the user's blood sugar management is practically impossible due to the insulin supply value, and this may be based on a judgment standard that is stored in the user terminal (130) or transmitted from the database (400).

[0043] Additionally, the malfunction detection unit (133) may detect an abnormality in the operation error judgment performed by the aforementioned error judgment unit (132). That is, the malfunction detection unit (133) can detect not only that the provision of insulin controlled by artificial intelligence is unexpectedly performed incorrectly, but also that the operation error judgment of the closed-loop medical device by artificial intelligence is unexpectedly performed incorrectly.

[0044] For example, the error judgment unit (132) determines an operation error based on a security model learned by artificial intelligence (neural network, etc.), and, as with the artificial intelligence control module (1311), may determine normal insulin delivery as an operation error or not determine abnormal insulin delivery as an operation error due to reasons such as a lack of training data or overfitting. In this way, even when the security model by artificial intelligence operates abnormally, it can detect this and report it to the user.

[0045] The above malfunction notification unit (134) can provide a notification regarding the content of the malfunction detected by the above malfunction detection unit (133). In addition, the malfunction notification unit (134) can provide a notification regarding the criticality of the malfunction of the artificial intelligence control module (1311) along with the content of the malfunction, and can also provide an option for the user to choose whether to maintain operation by the artificial intelligence control module (1311). Referring to FIG. 3, the malfunction notification unit (134) may include a report module (1341) and a selection provision module (1342).

[0046] The report module (1341) can report the details of the malfunction of the artificial intelligence control module (1311) detected by the malfunction detection unit (133). The details of the malfunction may include, for example, abnormalities in the amount of insulin provided, abnormalities in the insulin provision cycle, etc. Additionally, the classification module (1333) may report the results of classifying the malfunction of the artificial intelligence control module based on how fatal it is to the user's body, along with the details of the malfunction of the artificial intelligence control module (1311). That is, along with whether there is an abnormality in the insulin provision value, the fatality of such insulin provision value to the user is analyzed and provided to the user.

[0047] The selection provision module (1342) can provide a choice regarding the operation switching of the injection control unit (131). That is, if it is determined that an artificial intelligence control module (1311) has malfunctioned, the report module (1341) reports the details of the malfunction, and the selection provision module (1342) can provide a choice to the user to select whether to switch the operation. At this time, if the classified criticality exceeds a predetermined level, the switching module (1332) can switch the operation of the injection control unit (131) to operation by the manual control module (1312) without the selection provision module (1342) providing a choice to the user.

[0048] As illustrated in FIG. 4, the malfunction notification unit (134) can report the malfunction of the artificial intelligence control module (1311) as an abnormal operation. Additionally, the malfunction notification unit (134) can provide information on whether to switch the operation of the artificial pancreas (closed-loop medical device) from the artificial intelligence control module (1311) to the manual control module (1312) along with the abnormal operation. The user can view the report and select whether to switch the operation accordingly. Furthermore, if it is determined that the artificial intelligence control module (1311) is malfunctioning, but the user is exercising, the user may choose to ignore this and allow the operation by the artificial intelligence control module (1311) to continue.

[0049] Additionally, the malfunction notification unit (134) may receive feedback regarding abnormal operation. That is, the malfunction notification unit (134) may receive feedback in cases where it is determined that there is a malfunction of the artificial intelligence control module (1311) but it cannot actually be considered a malfunction, or in cases where there was a malfunction of the artificial intelligence control module (1311) but the classification of the criticality of the user's blood sugar management or health status is incorrect, and this feedback can be applied to the future detection of malfunctions of the artificial intelligence control module (1311).

[0050] In one embodiment, the above-described error judgment unit (132) can be understood as a primary safety layer linked to the closed-loop medical device (100) in that it detects operational errors in the operation data of the closed-loop medical device caused by hacking of the artificial pancreas (closed-loop medical device) or software or hardware bugs. Meanwhile, the malfunction detection unit (133) can be understood as a secondary safety layer for the closed-loop medical device (100) in that it detects operational abnormalities of the artificial intelligence control module (1311) and switches to the manual control module (1312) in some cases. Additionally, the malfunction notification unit (134) can be understood as a third safety layer for the closed-loop medical device in that it collects various information transmitted from the malfunction detection unit (133) and transmits it so that the user can verify it, and allows the user to switch to manual control (which can be understood as the opposite of automatic control by artificial intelligence) or stop the closed-loop medical device to take measures against danger.

[0051] Referring to FIG. 3, a closed-loop medical device control unit (135) may be configured to control each piece of equipment of the closed-loop medical device (100), preferably an infusion pump (110), a blood glucose meter (120), and a user terminal (130). The closed-loop medical device control unit (135) may generate or transmit control commands for each piece of equipment of the closed-loop medical device (100). Additionally, the closed-loop medical device control unit (135) may control inputs, outputs, and internal operations related to each component of the connected device, such as the infusion pump (110), the blood glucose meter (120), and the user terminal (130). That is, the closed-loop medical device control unit (135) may transmit various commands and / or signals to the connected components including the infusion pump (110), the blood glucose meter (120), and the user terminal (130).

[0052] The closed-loop medical device control unit (135) may include an equipment error detector (1351) that detects equipment errors occurring in each piece of equipment of the closed-loop medical device, an equipment error manager (1352) that manages the equipment errors in response to the equipment errors, an equipment controller (1353) that receives and processes signals generated from the closed-loop medical device, a remote controller (1354) that provides remote control of the closed-loop medical device, and a reporter (1355) that transmits the processing of the equipment controller to the user.

[0053] The equipment error detector (1351) can detect mechanical problems in medical equipment forming a closed-loop medical device, such as an infusion pump (110), a blood glucose meter (120), and a user terminal (130). Here, mechanical problems are a concept distinct from the operation error of the artificial pancreas or the malfunction of the artificial intelligence control module (1311) described above, and may include communication errors, poor connection, or abnormal signals resulting therefrom between the infusion pump (110), the blood glucose meter (120), and the user terminal (130). For example, the equipment error detector (1351) can detect communication failures where physiological information is not smoothly transmitted from the infusion pump (110) or the blood glucose meter (120) to the user terminal (130), the user terminal (130) being turned off, or the suspension of an application for controlling the closed-loop medical device (100). However, mechanical errors detected by the equipment error detector (1351) are not limited to this, and can be understood as detecting errors that occur regardless of insulin injection related to the user's blood sugar, such as the operation error of the artificial pancreas described above or the malfunction of the artificial intelligence control module.

[0054] The equipment error manager (1352) can manage the closed-loop medical device to operate smoothly in response to local events (mechanical problems with the equipment) detected by the equipment error detector (1351). For example, the equipment error manager (1352) can periodically check and connect communication between each piece of equipment forming the closed-loop medical device (100), and can check and operate applications for controlling the closed-loop medical device within the user terminal (130) so that they do not stop during the operation of the closed-loop medical device, but the operation of the equipment error manager (1352) is not limited to this.

[0055] The equipment controller (1353) detects signals from the closed-loop medical device (100) and processes various information to be transmitted to the user. For example, the equipment controller (1353) may be configured to transmit abnormal signals of various levels, such as abnormal insulin injection trends and abnormal blood glucose trends, to the user, and to control each device of the closed-loop medical device accordingly.

[0056] However, it is preferable to understand that the equipment controller (1353) is not configured to detect or derive abnormal insulin injection or abnormal blood sugar levels through artificial intelligence, but rather to control the equipment by organizing and visualizing already detected abnormal signals, transmitting derived abnormal signals, and switching to manual control of the artificial pancreas by user control.

[0057] The remote controller (1354) can provide full control functions for an application for controlling a user terminal (130) or a closed-loop medical device (100) installed on the user terminal (130) so that the user can take action in an emergency situation. For example, in an emergency, the user or medical staff can access the user terminal (130) through the remote controller (1354) to stop the closed-loop medical device or switch to manual control.

[0058] The reporter (1355) can transmit signals to the user that include physiological information transmitted from the closed-loop medical device (100) and mechanical information of each device. Specifically, the reporter (1355) can transmit to the user information regarding various signals processed by the device controller (1353), etc., and the status of each device forming the closed-loop medical device (100) (e.g., power, communication status, etc.).

[0059] Referring again to FIG. 3, the gateway (200) can transmit the user's physiological information transmitted from the closed-loop medical device (100), specifically the infusion pump (110) and blood glucose meter (120), to the security model generation device (300). The gateway (200) can be understood as performing an interface function between different network devices to facilitate communication and data transmission between them, and to enable interoperability and connectivity across heterogeneous equipment. The gateway (200) can be understood as substantially the same as a commonly used gateway. Additionally, for example, the gateway (200) can transmit additional information, including the user's blood pressure and exercise information stored at least in the user terminal (130), to the security model generation device (300). Specifically, additional information stored in the user terminal (130) or an external device (500) can be transmitted to the security model generation device (300). That is, the security model generation device receives additional information including the user's blood pressure and exercise information stored in the user terminal or external device, and can generate or update a security model based on the physiological information and additional information.

[0060] Referring further to FIG. 3, the security model generating device (300) may be configured to generate a security model that determines an operation error of the artificial pancreas from physiological information. The security model generating device (300) may analyze the physiological information in a time-series manner to determine the level of operation error of the artificial pancreas (a closed-loop medical device) in real-time or periodically, thereby deriving the operation error of the closed-loop medical device.

[0061] More specifically, the security model generation device (300) can train a security model to derive an operation error of a closed-loop medical device through learning, and distribute the security model generated as a result of learning to the user terminal (130) to derive an operation error of the closed-loop medical device. Information regarding the derived operation error is transmitted back to the security model generation device (300) for re-learning, and by generating and distributing an updated security model through re-learning, an adaptive security system for the closed-loop medical device (100) can be established. Referring to FIG. 5, the security model generation device (300) may include a preprocessing unit (310), an error classification unit (320), and a learning unit (330).

[0062] In one embodiment, the preprocessing unit (310) may perform preprocessing on physiological information which is time-series information. The preprocessing may include a cleaning technique and a normalization technique. For physiological information which is time-series information, if the degree of need for cleaning or the degree of need for normalization is greater than or equal to a set level, the preprocessing technique may be selectively applied to the physiological information based on the corresponding degree of need.

[0063] Additionally, the preprocessing unit (310) can augment or generate physiological information when there is a lack of physiological information for learning. In particular, the preprocessing unit (310) can augment or generate physiological information when there is a lack of physiological information as time-series information including operational errors of the artificial pancreas (closed-loop medical device). The augmented or generated physiological information can be used as training data in the security model generation device (300) and applied to the derivation of the security model.

[0064] For example, physiological information as time-series information can be generated through generation methods such as GANs or CycleGANs.

[0065] In addition, physiological information as time-series information can be augmented using at least one of a plurality of augmentation techniques.

[0066] Referring to Fig. 6, it can be seen that the Original graph is a graph of unenhanced time-series processed physiological information, such as blood glucose levels over time. Here, if a Jittering technique is applied to the physiological information corresponding to the Original graph in (a) to enhance it, the physiological information can be enhanced as shown in the Jittering graph in (b). Additionally, if a Scaling technique is applied to the physiological information corresponding to the Original graph in (a) to enhance it, it can be seen that the physiological information is enhanced as shown in the Scaling graph in (c). Furthermore, if a Time Warping technique is applied to the physiological information corresponding to the Original graph in (a) to enhance it, it can be seen that the physiological information is enhanced as shown in the Time Warping graph in (d).

[0067] According to one embodiment, a plurality of augmentation techniques may include a first augmentation technique that augments physiological information by adding noise. That is, through the first augmentation technique, augmented physiological information can be generated by adding noise to the physiological information. The first augmentation technique may be implemented by generating a random number from a Gaussian distribution and adding it to the physiological information. The first augmentation technique may include a jittering technique.

[0068] In addition, the multiple augmentation techniques may include a second augmentation technique that augments physiological information by applying a pre-set amount of magnitude change to the variables of the physiological information. If the label for each piece of physiological information can be maintained even if the physiological information changes within a certain level, the physiological information can be augmented by applying a set amount of magnitude change to the variables of the physiological information, for example, blood glucose levels or variables related to blood glucose levels. For example, the second augmentation technique can be implemented by multiplying the blood glucose levels of the physiological information or variables related to blood glucose levels by an arbitrary value. The second augmentation technique can be understood as including a scaling technique. That is, if physiological information, which is time-series data judged to be an operational error of the artificial pancreas, can still be judged to be an operational error of the artificial pancreas even when multiplied by a predetermined value, the physiological information judged to be an operational error can be augmented by multiplying by the predetermined value.

[0069] Furthermore, the multiple augmentation techniques may include a third augmentation technique that augments by changing the time-series reference, i.e., the time point, of the physiological information. If the change in the time point of the physiological information within a predetermined range does not affect the label, the physiological information can be augmented by transforming the time point of the physiological information. The third augmentation technique may be implemented by changing the temporal position of the physiological information by distorting the time interval of the physiological information, which is judged to be an operational error, to a preset degree. For example, the third augmentation technique may include a warping technique.

[0070] However, in the case of the artificial pancreas simulator (1), the preprocessing unit (310) may not perform preprocessing, and a security model may be created by performing training using data related to the operation errors of the artificial pancreas that has been collected or previously stored as training data. That is, the preprocessing unit (310) may preprocess physiological information as needed and use it as training data, and it goes without saying that preprocessing may not be performed if preprocessing of physiological information is not necessary.

[0071] Referring again to FIG. 5, the error classification unit (320) may be configured to classify the types of operating errors of the artificial pancreas. The error classification unit (320) may label the types of errors classified with respect to physiological information related to the operating errors of the artificial pancreas.

[0072] In one embodiment, the type of operation error may be classified according to at least one of the criteria including the cause of the operation error, the configuration of the artificial pancreas where the operation error occurred, and the lethality of the operation error.

[0073] In a preferred embodiment, the error classification unit (320) can classify the type of operation error based on the criticality of the operation error.

[0074] The above-mentioned learning unit (330) may be configured to train a security model that determines an operation error of the artificial pancreas from physiological information. More specifically, the learning unit (330) may train a security model that derives an operation error of the closed-loop medical device based on at least the user's blood glucose level and insulin supply value. Additionally, the learning unit (330) may train a security model that derives an operation error of the closed-loop medical device based on physiological information and additional information.

[0075] In one embodiment, the learning unit (330) can perform learning based on at least a portion of the data related to the operation error of the artificial pancreas that is pre-stored, for example, physiological information related to the operation error of the artificial pancreas generated or augmented by the preprocessing unit (310), physiological information related to the operation error of the artificial pancreas that is pre-stored, and physiological information determined to be an operation error of the artificial pancreas by a security model distributed to the user terminal (130).

[0076] The security model trained to derive an operation error of the artificial pancreas may be generated using one of a Convolutional Neural Network (CNN), a Deep Neural Network (DNN), and a Recurrent Neural Network (RNN), or may be generated by combining at least two of a Convolutional Neural Network, a Deep Neural Network, and a Recurrent Neural Network.

[0077] That is, the learning unit (330) can generate a security model that determines an operation error of the artificial pancreas based on physiological information transmitted from the infusion pump (110) and the blood glucose meter (120) as described above, and the physiological information determined as an operation error of the artificial pancreas by the security model is used again as learning data of the learning unit (330) to periodically update the security model and distribute the security model to the user terminal (130) in an adaptive manner.

[0078] Referring again to FIG. 2, the database (400) may store various data for the operation of the artificial pancreas simulator (1). For example, the database (400) may store physiological information related to the operation errors of the artificial pancreas that is pre-stored as training data for generating a security model in the security model generation device (300). In addition, judgment criteria for detecting malfunctions of the artificial intelligence control module (1311) may be stored. Furthermore, criteria for classifying fatalities caused by, for example, the operation errors of the artificial pancreas or the malfunctions of the artificial intelligence control module (1311) may be stored. It can be understood that the database (400) may store additional information for the operation of the present invention, such as supplementary information, in addition to what has been described above.

[0079] Additionally, the external device (500) may be configured to measure additional information other than physiological information. The external device (500) may be a wearable device for measuring additional information of the user. Additional information such as the user's blood pressure, exercise information, heart rate, and respiration may be stored in the external device (500), and the additional information measured by the external device (500) may be transmitted to a security model generating device (300) or stored in a database (400). Additionally, as described below, the additional information may be transmitted to an organ simulator (600) that simulates the user's physical functions related to changes in blood sugar.

[0080] Referring to FIG. 7, an artificial pancreas simulator (1) according to another embodiment of the present invention may be disclosed. The artificial pancreas simulator (1) may further include an organ simulator (600).

[0081] The organ simulator (600) may be equipped to simulate the user's physical functions related to changes in blood sugar levels. The organ simulator (600) can provide simulated physiological information to the user terminal (130) to perform more precise control over the closed-loop medical device (100). Here, control over the closed-loop medical device (100) may include providing insulin by the artificial intelligence control module (1311), determining an operation error of the artificial pancreas by the error determination unit (132), and detecting a malfunction of the artificial intelligence control module by the malfunction detection unit (133). The organ simulator (600) may simulate the user's physiological information transmitted from the closed-loop medical device (100), the physiological information simulated from additional information transmitted from the external device (500), and / or the simulated additional information.

[0082] Additionally, although not shown, the organ simulator (600) can transmit simulated physiological information to the security model generation device (300) to generate a more precise security model by taking into account the user's blood sugar changes according to the user's physical function.

[0083] Additionally, the organ simulator (600) may transmit simulated additional information along with simulated physiological information to the user terminal (130) and the security model generation device (300).

[0084] Referring to FIG. 8, the organ simulator (600) may include a blood glucose feedback simulation unit (610) that simulates blood glucose feedback according to the user's activity, a hypoglycemia simulation unit (620) that simulates blood glucose-related information according to physical functions for detecting hypoglycemia and suggesting carbohydrates, a diet simulation unit (630) that simulates blood glucose-related information according to physical functions for detecting meals and preventing hyperglycemia, an exercise simulation unit (640) that classifies the user's exercise and simulates blood glucose-related information according to the exercise, and a sleep simulation unit (650) that simulates blood glucose-related information according to physical functions for sleep management.

[0085] That is, the organ simulator (600) can generate simulated physiological information by simulating blood glucose-related information according to physical functions for at least one of blood glucose feedback based on user activity, detection of hypoglycemia and carbohydrate suggestion, detection of meals and prevention of hyperglycemia, exercise classification and sleep management. At this time, the blood glucose-related information can be understood as substantially identical to physiological information and may include the user's blood glucose change pattern and blood glucose level over time, and the simulated physiological information may be time-series information related to the user's blood glucose. In other words, the organ simulator (600) can simulate the user's blood glucose level and blood glucose change pattern over time.

[0086] Additionally, the organ simulator (600) may generate simulated additional information along with simulated physiological information by simulating at least one of the following: blood glucose feedback based on the user's activity, detection of hypoglycemia and carbohydrate suggestion, detection of meals and prevention of hyperglycemia, exercise classification and sleep management. That is, in addition to blood glucose-related information, information that affects the user's blood glucose changes and insulin provision, such as the user's blood pressure, exercise information, and dietary information, can be simulated.

[0087] The simulated physiological information provided by the organ simulator (600) can minimize the malfunction of the artificial intelligence control module (1311), the false positive of the error judgment unit (132) regarding the operation error of the artificial pancreas as a closed-loop medical device, and the false positive of the malfunction detection unit (133).

[0088] For example, based on the insulin supply value determined by the artificial intelligence control module (1311) regarding the simulated physiological information, it can be determined whether the artificial intelligence control module (1311) is functioning normally. Additionally, if training data is insufficient, the simulated physiological information can be used as training data for the artificial intelligence control module (1311) to prevent malfunction of the artificial intelligence control module (1311) to a certain extent.

[0089] As another example, the security model generating device (300) can generate a security model based on simulated physiological information. In this case, the simulated physiological information may include at least a portion of physiological information regarding the operation error of the artificial pancreas. Through this, the security model generating device (300) can learn and distribute the security model to determine the operation error of the artificial pancreas.

[0090] Additionally, the security model generation device (300) may train the security model based on physiological information that is pre-stored or determined to be an operation error of the artificial pancreas, and the security model may determine the operation error of the artificial pancreas from the simulated physiological information transmitted from the organ simulator.

[0091] That is, the security model generating device (300) can train a security model based on physiological information related to an operation error of an artificial pancreas that is not simulated, and distribute the security model to a user terminal (130) to determine the operation error of the artificial pancreas regarding the simulated physiological information. In this case, the simulated physiological information determined to be an operation error of the artificial pancreas among the simulated physiological information is transmitted to the security model generating device (300), and the security model generating device (300) can use the simulated physiological information added to the existing training data for training the security model.

[0092] In addition, more precise learning can be performed by providing simulated additional information along with simulated physiological information.

[0093] In one embodiment, the organ simulator (600) may generate different simulated physiological information according to the progression of diabetes classified according to set criteria. In this case, the progression of diabetes may be classified into a normal level, a suspected diabetes level, and a diabetes onset level based on fasting blood glucose and postprandial blood glucose.

[0094] Depending on the progression of diabetes, a person's blood sugar and physical functions may change. The organ simulator (600) can generate physiological information and additional information that are simulated to reflect the differences in blood sugar-related information and physical functions that change according to the progression of diabetes. As illustrated in FIG. 8, physical functions and physiological information and additional information based on physical functions may change according to the user's progression of diabetes. Accordingly, time-series blood sugar-related information (physiological information) of different users may be simulated according to classifications of normal level, suspected diabetes level, and diabetes onset level.

[0095] The progression of diabetes can be classified based on established values ​​for fasting blood glucose and postprandial blood glucose. For example, normal levels may have criteria of a fasting blood glucose of less than 100 mg / dL and a postprandial blood glucose of less than 140 mg / dL. Additionally, suspected diabetes levels may have criteria of a fasting blood glucose of less than 125 mg / dL and a postprandial blood glucose of less than 200 mg / dL. Furthermore, diabetes onset levels may have criteria of a fasting blood glucose of 126 mg / dL or higher and a postprandial blood glucose of 200 mg / dL or higher. However, these are exemplary criteria, and it goes without saying that different criteria and different stages may be classified depending on the physiological information and additional information intended to be simulated.

[0096] Below, we will briefly examine the operation flow of the present invention based on the details described above.

Claims

1. An infusion pump that injects insulin into a user, A blood glucose meter that measures the user's blood glucose at predetermined intervals, A user terminal configured to communicate with the above-mentioned infusion pump and blood glucose meter, and Includes an organ simulator that simulates the user's physical functions related to changes in blood sugar, The above user terminal is, An infusion control unit that determines insulin provided to the user based on physiological information transmitted from the above infusion pump and blood glucose meter, It includes an error determination unit that determines an operation error of the artificial pancreas from the above physiological information, and The above-mentioned organ simulator is characterized by providing at least simulated physiological information based on the user's body function simulated based on the above-mentioned physiological information to the error determination unit.

2. In Paragraph 1, The above-mentioned organ simulator generates simulated physiological information by simulating blood glucose-related information according to physical functions for at least one of blood glucose feedback based on user activity, detection of hypoglycemia and carbohydrate suggestion, detection of meals and prevention of hyperglycemia, exercise classification and sleep management. An artificial pancreas simulator characterized in that the simulated physiological information is time-series information related to the user's blood sugar.

3. In Paragraph 2, The above-mentioned organ simulator is an artificial pancreas simulator characterized by generating different simulated physiological information according to the progression of diabetes classified according to set criteria.

4. In Paragraph 3, An artificial pancreas simulator characterized by the above-mentioned diabetes progression being classified into normal level, suspected diabetes level, and diabetes onset level based on fasting blood glucose and postprandial blood glucose.

5. In Paragraph 1, The device further includes a security model generating device that generates a security model for determining an operation error of an artificial pancreas from simulated physiological information transmitted from the organ simulator above, The above security model generation device is, Error classification unit for classifying types of operating errors of the above artificial pancreas, It includes a learning unit that trains a security model for determining an operation error of the artificial pancreas from the above physiological information, and The above error classification unit labels the types of errors classified for the pre-stored data related to the operation errors of the artificial pancreas, and An artificial pancreas simulator characterized in that the type of the above-mentioned operating error is classified according to at least one of criteria including the cause of the operating error, the configuration of the artificial pancreas in which the operating error occurred, and the lethality of the operating error.

6. In Paragraph 5, The above error classification unit classifies the types of the above operation errors based on the criticality of the above operation errors, and The above learning unit trains the security model to derive whether an operation error occurs and the criticality of the operation error from the above physiological information, and An artificial pancreas simulator characterized by the user terminal generating a notification based on the criticality of an operation error determined by the security model.

7. In Paragraph 2, The above organ simulator is, Based on the above physiological information and additional information including the user's blood pressure and exercise information stored in the user terminal An artificial pancreas simulator characterized by updating a security model by adding data related to the above-mentioned operational error as training data.

8. In Paragraph 1, The above gateway transmits additional information, including the user's blood pressure and exercise information stored in the user terminal, to the security model generation device, and The above security model generation device is an artificial pancreas simulator characterized by simulating the user's physical functions related to blood sugar changes based on the above physiological information and additional information.

Images