Closed-loop infusion system and method based on food image recognition

Abstract

This invention discloses a closed-loop infusion system and method based on food image recognition. An imaging module acquires food images, and a food image recognition model analyzes these images to obtain food data. An insulin algorithm determines the patient's postprandial insulin infusion volume based on the food data and simulates the patient's postprandial blood glucose. While the artificial pancreas infuses insulin, real-time blood glucose data is monitored. The real-time blood glucose data is compared with the simulated blood glucose data, and the comparison results are used to correct the parameters of the insulin algorithm, making it more suitable for patients. This improves the accuracy and adaptability of the insulin algorithm and benefits the treatment of diabetes.

Description

Closed-loop infusion system and method based on food image recognition

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of and priority to the following patent application: PCT patent application number PCT / CN2023 / 126418 filed on October 25, 2023. Technical Field

[0003] The present invention mainly relates to the field of medical devices, and in particular to a closed-loop infusion system and method based on food image recognition. Background Art

[0004] In a healthy individual, the pancreas automatically secretes insulin and glucagon based on blood sugar levels, maintaining a healthy blood sugar range. However, in diabetics, pancreatic function is abnormal, preventing the body from producing the necessary insulin. Diabetes is a metabolic disease, a lifelong condition. Current medical technology cannot cure diabetes; instead, it can control the onset and progression of diabetes and its complications by stabilizing blood sugar levels.

[0005] Diabetic patients need to monitor their blood sugar before injecting insulin. Currently, most monitoring methods use in-vivo blood sugar monitoring devices to continuously monitor blood sugar. These devices use disposable transcutaneous sensors inserted into the skin to measure the blood sugar concentration in the interstitial fluid and transmit this data in real time to an external device via a transmitter for easy viewing by the patient. This monitoring method is called continuous glucose monitoring (CGM). Based on the blood sugar value detected by the CGM, an insulin pump automatically adjusts the patient's required insulin infusion volume and delivers the insulin subcutaneously, thus forming a closed-loop artificial pancreas.

[0006] When patients administer insulin before meals, they need to consider the impact of food on their blood sugar levels. Different food types and weights have varying effects on blood sugar levels, making accurate insulin delivery difficult. Current technologies lack a suitable method for accurately estimating blood sugar changes and insulin dosage based on food intake.

[0007] Therefore, the prior art urgently needs a method and system that can accurately estimate blood sugar changes and insulin infusion amounts based on food during mealtimes.

[0008] Summary of the Invention

[0009] An embodiment of the present invention discloses a closed-loop infusion system and method based on food image recognition. An imaging module acquires food images, and a food image recognition model analyzes the food images to acquire food data. An insulin algorithm determines a patient's postprandial insulin infusion amount based on the food data and simulates the patient's postprandial blood sugar. A closed-loop artificial pancreas detects real-time blood sugar data while performing insulin infusion, and compares the real-time blood sugar data with the simulated blood sugar data. The comparison results are used to correct the parameters of the insulin algorithm, making the insulin algorithm more suitable for the patient, thereby improving the accuracy and adaptability of the insulin algorithm and benefiting the patient's diabetes treatment.

[0010] The present invention discloses a closed-loop artificial pancreas system, comprising an imaging module for acquiring food images before a meal; a food image recognition model for recognizing food images to acquire food data, wherein the food data includes at least the type and weight of the food; an insulin algorithm for determining a patient's postprandial insulin infusion amount based on the food data; and a closed-loop artificial pancreas for completing postprandial insulin infusion and real-time blood glucose testing to acquire actual blood glucose data; after determining the postprandial insulin infusion amount, simulating the patient's postprandial blood glucose data based on the postprandial insulin infusion amount to obtain simulated blood glucose data, comparing the simulated blood glucose data with the actual blood glucose data, and correcting the parameters of the insulin algorithm based on the comparison result.

[0011] According to one aspect of the present invention, the insulin algorithm after parameter correction is a narrow insulin algorithm, and the narrow insulin algorithms of different patients are aggregated to form a narrow insulin algorithm cluster.

[0012] According to one aspect of the present invention, in a cluster of narrow insulin algorithms, narrow insulin algorithms are classified into groups based on at least one identifiable characteristic of a patient, and the groups have the identifiable characteristic in common.

[0013] According to one aspect of the invention, the groups are independent of each other.

[0014] According to one aspect of the invention, two groups having at least one identical identifiable characteristic are subsets of each other.

[0015] According to one aspect of the present invention, the identifiable feature is at least associated with a physiological feature or a life habit of the patient.

[0016] According to one aspect of the present invention, parameters of the image recognition model are modified based on the comparison results.

[0017] According to one aspect of the present invention, the insulin algorithm includes a logical operation or a lookup table to determine the amount of post-meal insulin infusion based on the type and weight of food.

[0018] According to one aspect of the invention, an insulin algorithm is integrated into an image recognition model.

[0019] According to one aspect of the present invention, a food nutrition database is further included, and the food data further includes the content of nutrients in the food. The food nutrition database is searched based on the type and weight of the food to determine the content of nutrients in the food.

[0020] According to one aspect of the present invention, an insulin algorithm determines the amount of post-meal insulin infusion based on the content of nutrients.

[0021] According to one aspect of the present invention, a cloud server is further included, and the cloud server establishes communication with the imaging module or the closed-loop artificial pancreas.

[0022] According to one aspect of the present invention, the insulin algorithm is located in the imaging module, the closed-loop artificial pancreas, or a cloud server.

[0023] According to one aspect of the present invention, the image recognition model is located in the imaging module, the closed-loop artificial pancreas or the cloud server.

[0024] According to one aspect of the present invention, the food nutrient library is located in the imaging module, the closed-loop artificial pancreas or the cloud server.

[0025] According to one aspect of the present invention, the imaging module is an image acquisition device independent of the closed-loop artificial pancreas.

[0026] According to one aspect of the present invention, the imaging module is a submodule of the closed-loop artificial pancreas.

[0027] The present invention also discloses a closed-loop insulin infusion method, which includes providing: an imaging module, a closed-loop artificial pancreas, a food image recognition model and an insulin algorithm. The closed-loop insulin infusion method includes the following steps:

[0028] Ⅰ. Before eating, use the imaging module to obtain food images;

[0029] II. The food image recognition model recognizes the food image and obtains food data related to the food;

[0030] Ⅲ. Based on food data, the insulin algorithm determines the patient's postprandial insulin infusion amount and simulates the postprandial blood glucose data to obtain

[0031] to simulated blood glucose data;

[0032] IV. A closed-loop artificial pancreas delivers post-meal insulin infusion and monitors and records actual blood sugar data;

[0033] V. Compare the simulated blood glucose data with the actual blood glucose data, and modify the parameters of the insulin algorithm based on the comparison results.

[0034] According to one aspect of the present invention, the method further includes a step of patient confirmation of food data. If the food data is confirmed by the patient, step III is performed; otherwise, step I is returned to, or the food image is re-identified by the food image recognition model.

[0035] According to one aspect of the present invention, in step II, the food data obtained includes at least the type and weight of the food.

[0036] According to one aspect of the present invention, the method further comprises providing a food nutrition library, and determining the nutritional content of the food based on the type and weight of the food.

[0037] According to one aspect of the present invention, in step III, the amount of post-meal insulin infusion is determined by an insulin algorithm based on the nutritional content of the food.

[0038] According to one aspect of the present invention, in step III, the amount of post-meal insulin infusion is determined by an insulin algorithm based on the type and weight of food.

[0039] According to one aspect of the present invention, a cloud server is provided, and the image recognition model, insulin algorithm or food nutrition library is stored in the cloud server.

[0040] According to one aspect of the present invention, step II further includes uploading the food image to a cloud server and completing food image recognition in the cloud server.

[0041] According to one aspect of the present invention, step III also includes uploading food data to a cloud server and completing the calculation of post-meal insulin in the cloud server.

[0042] According to one aspect of the present invention, the method further includes simulating the patient's postprandial blood sugar in a cloud server based on the postprandial insulin infusion amount to obtain simulated blood sugar data.

[0043] According to one aspect of the present invention, step IV further includes transmitting the post-meal insulin infusion amount data to the closed-loop artificial pancreas by the cloud server.

[0044] According to one aspect of the present invention, step V further includes transmitting the actual blood glucose data to a cloud server, and the cloud server performs comparison between the simulated blood glucose data and the actual blood glucose data.

[0045] According to one aspect of the present invention, in step IV, the time of the actual blood glucose data and the simulated blood glucose data are overlapped.

[0046] According to one aspect of the present invention, in step V, the comparison result is also used to modify the image recognition model parameters.

[0047] According to one aspect of the present invention, in step V, the parameter-corrected insulin algorithm is used to calculate the patient's next post-meal insulin infusion amount.

[0048] According to one aspect of the present invention, in step V, the insulin algorithm after parameter correction is a narrow insulin algorithm, and the narrow insulin algorithms of different patients are aggregated into a narrow insulin algorithm cluster.

[0049] According to one aspect of the present invention, in the cluster of narrow insulin algorithms, the narrow insulin algorithms are classified into groups according to at least one identifiable characteristic of the patient, and have a common identifiable characteristic within the group.

[0050] According to one aspect of the present invention, the postprandial insulin infusion amount includes a basal amount and a large dose.

[0051] Compared with the prior art, the technical solution of the present invention has the following advantages:

[0052] In the closed-loop infusion system and method based on food image recognition disclosed in the present invention, an imaging module acquires food images, and a food image recognition model analyzes the food images to acquire food data. An insulin algorithm determines the patient's postprandial insulin infusion amount based on the food data and simulates the patient's postprandial blood sugar. An artificial pancreas detects real-time blood sugar data while performing insulin infusion, and the real-time blood sugar data is compared with the simulated blood sugar data. The comparison results are used to correct the parameters of the insulin algorithm, making the insulin algorithm more suitable for the patient, thereby improving the accuracy and adaptability of the insulin algorithm and benefiting the patient's diabetes treatment.

[0053] Furthermore, determining the patient's postprandial insulin infusion amount based on the type and weight of food can more accurately control the patient's postprandial blood sugar level, which is helpful for the patient's diabetes treatment.

[0054] Furthermore, determining the patient's postprandial insulin infusion amount based on the nutrient content of the food can more accurately control the patient's postprandial blood sugar level, which is helpful for the patient's diabetes treatment.

[0055] Furthermore, after the image recognition model identifies the food data, it still needs to be confirmed by the patient before calculating the amount of post-meal insulin infusion, so as to avoid misjudgment or recognition errors of the image recognition model, improve the recognition accuracy of the image recognition model, and ensure the safety of the patient's infusion.

[0056] Furthermore, the insulin algorithm with corrected parameters can be used by the patient for the next post-meal insulin calculation. As the insulin algorithm parameters are corrected and iterated, the insulin algorithm will become more and more suitable for the patient, improving the accuracy of the insulin algorithm, and being able to more accurately control the patient's post-meal blood sugar level, which will help the patient's diabetes treatment.

[0057] Furthermore, the insulin algorithms with corrected parameters are aggregated into an insulin algorithm cluster. In the cluster, the insulin algorithms are classified into groups according to the patient's identifiable characteristics. Patients with the identifiable characteristics can call the insulin algorithms in the group for their own use. The insulin algorithms in the group are more suitable for patients with the same identifiable characteristics, and can more accurately control the patient's postprandial blood sugar level, thereby helping the patient's diabetes treatment.

[0058] Furthermore, patients can search for insulin algorithms with the same identifiable features as themselves in the insulin algorithm cluster. They can use the insulin algorithm that suits them without having to go through multiple insulin algorithm parameter corrections. This makes the calculation of insulin dosage more accurate, and can more precisely control the patient's postprandial blood sugar level, which is helpful for the patient's diabetes treatment.

[0059] Furthermore, the cloud server can be connected to the closed-loop artificial pancreas system. With the help of the powerful computing and storage capabilities of the cloud server, it can better provide services for food image recognition, insulin calculation, insulin algorithm correction, and storage and reading of insulin algorithm clusters, and can optimize the closed-loop artificial pancreas, smart device computing power, and insufficient storage capacity.

[0060] Furthermore, identifiable features can be added, modified or deleted. Due to the large base of diabetic patients, the types and number of identifiable features must be huge. It is impossible to store all identifiable features in the system or server in an exhaustive manner. As a large number of patients use it, new features can be added, modified or deleted at any time, and identifiable features can be continuously optimized to meet the needs of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] FIG1 is a schematic diagram showing the relationship between modules of a general closed-loop artificial pancreas insulin infusion control system;

[0062] FIG2 is a schematic structural diagram of an integrated CGM according to an embodiment of the present invention;

[0063] FIG3 is a schematic structural diagram of a split-type CGM according to an embodiment of the present invention;

[0064] FIG4 a is a schematic structural diagram of an integrated insulin pump according to an embodiment of the present invention;

[0065] FIG4 b is a schematic structural diagram of a split-type insulin pump according to an embodiment of the present invention;

[0066] FIG5a is a schematic diagram of a main interface of a control system in a first working mode according to an embodiment of the present invention;

[0067] FIG5 b is a schematic diagram of the main interface of the control system according to an embodiment of the present invention when the control system is in the second working mode;

[0068] 6a-6c are schematic diagrams of different operations of enabling the insulin pump function of a control system according to an embodiment of the present invention;

[0069] 7a-7c are schematic diagrams of different operations of turning on the automatic mode function of a control system according to an embodiment of the present invention;

[0070] 8a-8b are schematic diagrams of the APP interface before and after the automatic mode function is turned on in the control system according to an embodiment of the present invention;

[0071] FIG9 a is a schematic diagram of an interface when a system turns on a feast mode according to an embodiment of the present invention;

[0072] 9b and 9c are schematic diagrams of different interfaces when selecting regular and feast mode in the system according to an embodiment of the present invention;

[0073] FIG9 d is a schematic diagram of the interface when the system turns on the regular meal mode according to an embodiment of the present invention;

[0074] FIG10 is a schematic diagram of the process of the infusion strategy of pre-infusion and supplementary infusion according to an embodiment of the present invention;

[0075] FIG11a is a schematic diagram of a closed-loop infusion system based on food image recognition according to an embodiment of the present invention;

[0076] FIG11 b is a schematic diagram of a closed-loop infusion method based on food image recognition according to one embodiment of the present invention;

[0077] FIG11c is a schematic diagram of a closed-loop infusion method based on food image recognition according to another embodiment of the present invention;

[0078] FIG12 is a schematic diagram of another closed-loop infusion method based on food image recognition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0079] As previously mentioned, when patients administer insulin during mealtimes, they need to consider the impact of food on their blood sugar levels. Different food types and weights have varying effects on blood sugar levels, making accurate insulin delivery difficult. Current technologies lack a suitable method for accurately estimating blood sugar changes and insulin dosage based on food intake.

[0080] In order to solve this problem, the present invention provides a closed-loop infusion system and method based on food image recognition, in which an imaging module acquires food images, and a food image recognition model analyzes the food images to acquire food data. The insulin algorithm determines the patient's postprandial insulin infusion amount based on the food data and simulates the patient's postprandial blood sugar. The artificial pancreas detects real-time blood sugar data while infusing insulin, and compares the real-time blood sugar data with the simulated blood sugar data. The comparison results are used to correct the parameters of the insulin algorithm, making the insulin algorithm more suitable for the patient, improving the accuracy and adaptability of the insulin algorithm, and benefiting the patient's diabetes treatment.

[0081] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments should not be construed as limiting the scope of the present invention.

[0082] In addition, it should be understood that for ease of description, the sizes of the various components shown in the drawings are not necessarily drawn according to actual proportional relationships. For example, the thickness, width, length or distance of certain units may be enlarged relative to other structures.

[0083] The following description of exemplary embodiments is merely illustrative and is not intended to limit the present invention, its application, or use in any sense. Technologies, methods, and apparatus known to those skilled in the art may not be discussed in detail herein, but to the extent applicable, such technologies, methods, and apparatuses should be considered part of this specification.

[0084] It should be noted that like reference numerals and letters denote like items in the following figures, and thus, once an item is defined or described in one figure, it will not need to be further discussed in the subsequent figure descriptions.

[0085] FIG1 is a schematic diagram showing the relationship between modules of a general closed-loop artificial pancreas insulin infusion control system.

[0086] The closed-loop artificial pancreas insulin infusion control system disclosed in the embodiment of the present invention mainly includes a detection module 100 , a program module 101 and an infusion module 102 .

[0087] The detection module 100 is used to continuously monitor the patient's current blood glucose level. Typically, the detection module 100 is a continuous glucose monitoring (CGM), which can detect the patient's current blood glucose level in real time, monitor blood glucose changes, and transmit the current blood glucose information to the program module 101. The CGM includes an implantable sensor connected to a transmitter. The transmitter also includes a memory, a processor, a communication interface, etc. The transmitter is used to transmit at least the blood glucose data monitored by the CGM and the CGM identifier information.

[0088] Infusion module 102 includes the necessary mechanical structures and electronic control unit for insulin infusion, such as a drug reservoir, a drive mechanism, an infusion line and needle, a power supply, and a circuit board. It is controlled by program module 101. Typically, infusion module 102 is an insulin pump, and the electronic control unit includes a memory, a processor, and a communication interface. Based on the current insulin infusion volume data transmitted by program module 101, infusion module 102 infuses the patient's body with the currently required insulin. Simultaneously, the infusion status of infusion module 102 is also fed back to program module 101 in real time.

[0089] Program module 101 is used to control the operation of detection module 100 and infusion module 102. Based on at least the blood glucose level detected by detection module 100, program module 101 generates insulin infusion instructions and controls infusion module 102 to perform infusion. The program module 101 includes memory, a processor, a communication interface, a display, a patient interface, and other components. The memory stores programming instructions, and the processor executes the programming instructions in the memory. Program module 101 is connected to detection module 100 and infusion module 102, respectively. This connection can include conventional electrical or wireless connections.

[0090] The embodiments of the present invention do not limit the specific positions and connection relationships of the detection module 100, the program module 101 and the infusion module 102, as long as the aforementioned functional conditions are met.

[0091] In one embodiment of the present invention, the three modules are electrically connected to form a single integrated structure. Therefore, they can be applied to the same location on the patient's skin. By connecting the three modules into a single unit and applying them to the same location, the number of devices applied to the patient's skin is reduced, thereby minimizing interference with patient activity caused by multiple devices. Furthermore, this effectively resolves the issue of wireless communication reliability between separate devices, further enhancing the patient experience.

[0092] In another embodiment of the present invention, the program module 101 and the infusion module 102 are interconnected to form an integrated structure, while the detection module 100 is separately provided in another structure. In this case, the detection module 100 and the program module 101 transmit wireless signals to each other to achieve mutual connection. Thus, the program module 101 and the infusion module 102 are attached to a certain location on the patient's skin, while the detection module 100 is attached to another location on the patient's skin.

[0093] In another embodiment of the present invention, program module 101 and detection module 100 are interconnected to form a single device, while infusion module 102 is located in a separate structure. Infusion module 102 and program module 101 transmit wireless signals to each other to achieve mutual connection. Thus, program module 101 and detection module 100 can be attached to a specific location on a patient's skin, while infusion module 102 can be attached to another location on the patient's skin.

[0094] In another embodiment of the present invention, the three modules are disposed in separate structures. Thus, the three modules are attached to different locations on the patient's skin. In this case, the program module 101 transmits wireless signals to the detection module 100 and the infusion module 102 to establish connections.

[0095] In another embodiment of the present invention, the three modules are arranged in separate structures. Thus, the detection module 100 and the infusion module 102 are attached to different locations on the patient's skin, while the program module 101 does not need to be attached to the skin. Instead, the detection module 100 and the infusion module 102 are controlled by a handheld or portable device, such as a PDM or a smartphone. In this case, the program module 101 transmits wireless signals to and from the detection module 100 and the infusion module 102, respectively, to establish a connection between them.

[0096] The wireless described in the foregoing embodiments may be achieved through, for example, but not limited to, radio frequency (RF) communication (e.g., radio frequency identification (RFID), Zigbee communication protocol, WiFi, infrared, wireless universal serial bus (USB), ultra-wideband (UWB), Communication protocols and cellular communications, such as Code Division Multiple Access (CDMA) or Global System for Mobile Communications (GSM).

[0097] Figure 2 is a schematic diagram of the structure of an integrated CGM according to an embodiment of the present invention. Figure 3 is a schematic diagram of the structure of a split CGM according to an embodiment of the present invention.

[0098] CGM includes a sensor and a transmitter, which are installed on the patient through an auxiliary installation device and inserted subcutaneously. The sensor is used to collect blood sugar content in the human body and transmit the collected blood sugar content information. The transmitter is connected to the sensor and is used to receive blood sugar data information transmitted by the sensor implanted subcutaneously and convert it into a wireless signal output. Each CGM has a unique identifier, such as a device identifier, hardware identifier, universally unique identifier, serial number, communication protocol-based identifier (such as BLE ID), manufacturer's identifier, etc. The identifier is formed by a combination of multiple randomly combined numbers and letters, which can be set on the CGM shell or packaging, and can also have different settings for different types of CGM.

[0099] Figure 2 is a schematic diagram of the structure of an integrated CGM. This means that the CGM's sensor and transmitter are integrated before use. This CGM is a single-use product that is discarded after use. As shown in Figure 2, the integrated CGM includes a sensor 201, a housing 202, and a transmitter (not shown) housed within housing 202. Sensor 301 monitors the patient's blood glucose data and transmits this data to the transmitter via internal circuitry. The transmitter then transmits this data to a receiver. The identifier can be placed on the CGM's outer housing, packaging, or within the CGM.

[0100] Figure 3 is a schematic diagram of the structure of a split-type CGM. Prior to use, the CGM's sensor and transmitter are two separate components, packaged separately. They are integrated together during use. The split-type CGM includes a base housing 301 and a transmitter 302. The base housing is provided with a sensor 3011, while the transmitter 302 has a separate housing. The base housing 301 and transmitter 302 are provided with snap-fit ​​structures 3012 and 3022, respectively. During use, the base housing 301 and transmitter 302 are snapped together into a single unit via the snap-fit ​​structures. The sensor 3011 is electrically connected to the transmitter 302 via an electrical connector 3013. The sensor 301 monitors a patient's blood glucose data and transmits this data to the transmitter 302 via the electrical connector 3013. The transmitter 302 then transmits this data to the receiver.

[0101] In one embodiment of the present invention, both the sensor and transmitter of a split-type CGM are disposable products that are discarded after use. Therefore, the identifier can be placed on the housing or outer packaging of the sensor or transmitter. In another embodiment of the present invention, only the sensor of the split-type CGM is a disposable product, while the transmitter is a reusable product. Therefore, in this embodiment, preferably, the identifier is placed on the housing or outer packaging of the transmitter. This can reduce the frequency of binding patient information and the identifier, thereby improving the patient experience. This will be described in detail below.

[0102] When the identifier is provided on the housing or outer packaging of the CGM or transmitter, it may be provided in the form of, but not limited to, a QR code, a barcode, or an NFC tag.

[0103] FIG4 a is a schematic structural diagram of an integrated insulin pump according to an embodiment of the present invention; FIG4 b is a schematic structural diagram of a split insulin pump according to an embodiment of the present invention.

[0104] In an embodiment of the present invention, the insulin pump is a patch-type insulin pump, that is, an insulin pump that does not include a long catheter, includes an infusion structure and a control structure, and is adhered as a whole to the patient's skin surface by a same adhesive patch. The drug is directly infused from the drug storage cartridge along the infusion needle into the subcutaneous tissue.

[0105] Each insulin pump has a unique identifier, such as a device identifier, hardware identifier, universally unique identifier, serial number, communication protocol-based identifier, manufacturer's identifier, etc. The identifier is formed by a multi-digit random combination of numbers and letters, which can be set on the housing or packaging of the insulin pump, and can also have different settings for different types of insulin pumps.

[0106] Figure 4a is a schematic diagram of the structure of an integrated insulin pump, that is, the infusion structure 410 and the control structure 400 of the insulin pump are arranged inside the same housing 10, the two are connected by a wire, and are affixed to a certain position on the patient's skin through an adhesive patch 420, and are discarded as a whole after one-time use; the identifier can be set on the outer housing or outer packaging of the insulin pump or inside the insulin pump.

[0107] Figure 4b is a schematic diagram of the structure of a split-type insulin pump, in which the infusion mechanism 410 and control mechanism 400 of the insulin pump are housed in separate housings, connected by a waterproof plug or directly snapped together and electrically connected to form a single unit. The identifier can be placed on the outer housing, packaging, or inside the insulin pump itself.

[0108] In one embodiment of the present invention, both the infusion structure and the control structure of a split-type insulin pump are disposable products that are discarded after use. Therefore, the identifier can be set on the housing or outer packaging of the infusion structure and / or the control structure. In another embodiment of the present invention, only the infusion structure of the split-type insulin pump is a disposable product, while the control structure is a reusable product. Therefore, preferably, in this embodiment, the identifier is set on the housing or outer packaging of the control structure. This can reduce the frequency of binding patient information and the identifier, thereby improving the patient experience. This will be described in detail below.

[0109] When the identifier is provided on the housing or outer packaging of the insulin pump or control structure, it may be provided in the form of, but not limited to, a QR code, a barcode, or an NFC tag.

[0110] According to the different degrees of illness of the patients and the patients' personal physical health conditions, some patients may only need CGM for continuous blood glucose monitoring, while some patients may not only need CGM for continuous blood glucose monitoring, but also need insulin pump for drug infusion. When the doctor determines that the patient only needs to use CGM for continuous blood glucose monitoring, since CGM only involves monitoring the patient's blood glucose, the patient's use of CGM on his own will not pose a risk to the patient's life safety. Therefore, the patient can purchase CGM on his own. Before the CGM is installed on the patient's skin surface, the patient can search and download a special APP for controlling CGM in the application store of the smartphone, and create a new account on the special APP, pair the patient's personal information with the CGM information to be worn, thereby achieving pairing and control of the smartphone and CGM. The CGM and insulin pump in the embodiment of the present invention are developed and produced by the same manufacturer, so they can be controlled by the same special APP in the smartphone. Since the insulin pump is not required by all patients, only CGM-related content is involved in the default main screen of the special APP, as shown in Figure 5a. On the one hand, the interface of the APP can be simplified, providing the patient's visual experience, and on the other hand, preventing the patient from misoperating the insulin pump function and affecting the normal use of the CGM function.

[0111] In one embodiment of the present invention, when the doctor determines that the patient needs to use an insulin pump for drug infusion, as shown in Figure 6a, the doctor sends an application to the backend administrator, requesting that the patient's account be added to the whitelist, allowing the patient to use the insulin pump function. The backend administrator receives the whitelist addition application sent by the doctor, adds the patient's account to the whitelist list, and sends feedback to the doctor that the whitelist addition has been completed. Furthermore, the backend administrator directly opens the insulin pump function of the APP interface used by the patient. At this time, the APP interface changes from Figure 5a to Figure 5b. Compared with the interface of Figure 5a, the interface of Figure 5b adds two function keys related to insulin infusion, "Insulin Delivery" and "Easyloop". In another embodiment of the present invention, the backend administrator does not directly open the insulin pump function of the APP interface used by the patient, but sends a security code to the patient account. The patient can use the security code to open the insulin pump function of the APP interface when needed or convenient.

[0112] In another embodiment of the present invention, when a doctor determines that a patient requires an insulin pump for medication infusion, as shown in Figure 6b, the patient can directly submit a request to enable the insulin pump function to the backend administrator. Upon receiving the request, the backend administrator will verify whether the patient's account is on a whitelist. If the patient's account is on the whitelist, the backend administrator will enable the insulin pump function in the patient's app interface, and the app interface will change from Figure 5a to Figure 5b. If the patient's account is not on the whitelist, the backend administrator will send a feedback message to the patient's account, reminding the patient to ask the doctor to submit a request to add the patient to the whitelist. After the doctor sends the whitelist request to the backend administrator, the backend administrator can directly enable the insulin pump function in the patient's app interface. If the doctor does not receive a response from the backend administrator within a certain period of time (e.g., 1 minute, 2 minutes, or 5 minutes), the backend administrator can submit another request to enable the insulin pump function to the backend administrator, or ask the doctor to submit a whitelist request to the backend administrator. In another embodiment of the present invention, the backend administrator does not directly enable the insulin pump function in the patient's app interface. Instead, the backend administrator sends a security code to the patient's account, which the patient can use to enable the insulin pump function in the app interface when needed or convenient.

[0113] In another embodiment of the present invention, when the doctor determines that the patient needs to use an insulin pump for drug infusion, as shown in Figure 6c, the doctor sends an application to the backend administrator, requesting that the patient's account be added to the whitelist, allowing the patient to use the insulin pump function. The backend administrator receives the whitelist addition application sent by the doctor and adds the patient's account to the whitelist list. At the same time, feedback is sent to the doctor that the whitelist addition has been completed. Further, the doctor notifies the patient that he or she can apply to use the insulin pump function; after receiving the doctor's notification, the patient sends an application to open the insulin pump function to the backend administrator. After the backend administrator receives the application to open the insulin pump function sent by the patient, the backend administrator directly opens the insulin pump function on the APP interface used by the patient. In another embodiment of the present invention, after the backend administrator receives the application to open the insulin pump function sent by the patient, it can also first verify whether the patient's account exists in the whitelist. If it is determined that the patient's account is in the whitelist list, the backend administrator opens the insulin pump function on the APP interface used by the patient. In another embodiment of the present invention, the backend administrator does not directly open the insulin pump function on the APP interface used by the patient, but sends a security code to the patient's account. The patient can use the security code to open the insulin pump function on the APP interface when needed or convenient.

[0114] When the doctor determines that the patient no longer needs to have the insulin pump function enabled, the patient can disable the insulin pump function on the app. The app will automatically send a message to the backend administrator, who will then delete the patient account from the whitelist. The doctor and / or patient can also send a request to the backend administrator to disable the insulin pump function. The backend administrator will disable the insulin pump function on the patient's app and delete the patient account from the whitelist. If the patient needs to re-enable the insulin pump function, the patient account will need to be re-added to the whitelist using one of the methods described in Figures 6a-6c.

[0115] It should be noted that when turning on the insulin pump function of the app, it is necessary to ensure that the insulin pump is correctly installed on the skin and to pair the patient's personal information with the insulin pump information, thereby achieving pairing and control of the smartphone and the insulin pump. The patient's personal information includes name, age, gender, mobile phone number, etc., and the information of the CGM and / or insulin pump worn includes the identifier information of the CGM and / or insulin pump. At the same time, the smartphone uploads the patient's personal information and the identifier information of the CGM and / or insulin pump to a remote server. The remote server can store the information uploaded by the smartphone and verify whether the identifier information of the CGM and / or insulin pump is valid. If a certain identifier information already exists in the remote server, the remote server will send a prompt to the smartphone to remind the patient that the CGM or insulin pump has been used and needs to be replaced. When the CGM and / or insulin pump is installed on the patient's skin and successfully activated, the CGM and / or insulin pump begins to work. The CGM transmitter sends the monitored blood sugar information to the smartphone and further uploads it to the remote server. The control structure of the insulin pump receives the insulin infusion information and controls the infusion structure to infuse insulin. At the same time, the infusion status is sent to the smartphone and further uploaded to the remote server.

[0116] It should be noted that the CGM and insulin pump in the embodiment of the present invention are developed and produced by the same manufacturer, and therefore can be controlled by the same dedicated APP in the smartphone. When the patient needs both CGM and insulin pump and CGM at the same time, even if the CGM or insulin pump is produced by other manufacturers, the CGM or insulin pump can be directly controlled by the dedicated APP, which can also avoid the inconvenience caused to the patient by using different APPs to control the CGM and insulin pump respectively, thereby improving the patient experience.

[0117] When the CGM and / or insulin pump worn by the patient needs to be replaced due to reaching the usage cycle or failure, the unique identifier information of the new CGM and / or insulin pump also needs to be paired and updated with the patient's personal information through the smartphone and further uploaded to the remote server. The patient's personal information is entered manually, and the identifier information of the CGM and / or insulin pump can also be entered manually or by scanning the QR code, barcode, or NFC tag on the shell or outer packaging of the CGM and / or insulin pump.

[0118] When the CGM has a split structure and the transmitter is reusable, the CGM identifier is set on the transmitter's outer shell or packaging. When the patient replaces the CGM, he only needs to replace the sensor without replacing the transmitter. The CGM identifier also remains unchanged. Therefore, there is no need to update the pairing of the CGM identifier and the patient's personal information through a smartphone, nor is there any need to upload it to a remote server. This can reduce the number of operating steps and improve the patient experience.

[0119] When the insulin pump has a split structure and the control structure is reusable, the identifier of the insulin pump is set on the outer shell or packaging of the control structure. When the patient replaces the insulin pump, he only needs to replace the infusion structure without replacing the control structure. The identifier of the insulin pump also remains unchanged. Therefore, there is no need to update the pairing of the insulin pump identifier and the patient's personal information through a smartphone, nor is there any need to upload it to a remote server. This can reduce the number of operating steps and improve the patient experience.

[0120] The smartphone and the CGM and / or insulin pump, as well as the remote server, communicate wirelessly via, for example, but not limited to, radio frequency (RF) communication (e.g., radio frequency identification (RFID), Zigbee communication protocol, WiFi, infrared, wireless universal serial bus (USB), ultra-wide band (UWB), Communication protocols and cellular communications, such as Code Division Multiple Access (CDMA) or Global System for Mobile Communications (GSM). Preferably, the smartphone and the remote server communicate via WiFi and / or cellular, and the smartphone and the CGM and / or insulin pump communicate via Communication protocol communication.

[0121] When the doctor determines that the patient can enable automatic mode, the app reads the current blood glucose level monitored by the CGM and the insulin information infused by the insulin pump, calculates the future blood glucose trend, and controls the insulin pump infusion based on the calculated blood glucose trend, including increasing, decreasing, or stopping insulin infusion, to achieve the purpose of influencing the blood glucose level, forming an automated closed-loop control cycle. As shown in Figure 7a, the doctor sends a request to the backend administrator to add the patient's account to the whitelist, allowing the patient to use the automatic mode function. The backend administrator receives the doctor's whitelist addition request and adds the patient's account to the whitelist list. At the same time, it sends feedback to the doctor that the whitelist addition has been completed. Furthermore, the automatic mode function of the app interface used by the patient is directly enabled. At this time, the app interface changes from Figure 8a to Figure 8b. The interface of Figure 8b has an "Auto Mode" function key associated with the automatic mode compared to the interface of Figure 8a. In another embodiment of the present invention, the backend administrator does not directly enable the automatic mode function of the app interface used by the patient, but instead sends a security code to the patient's account. The patient can use the security code to enable the automatic mode function of the app interface when needed or convenient.

[0122] In another embodiment of the present invention, when the doctor determines that the patient can enable automatic mode, as shown in Figure 7b, the patient can directly send a request to enable the automatic mode function to the backend administrator. After receiving the request, the backend administrator verifies whether the patient's account is on the whitelist. If the patient's account is on the whitelist, the backend administrator enables the automatic mode function on the patient's app interface, and the app interface changes from Figure 8a to Figure 8b. If the patient's account is not on the whitelist, the backend administrator sends a feedback message to the patient's account, reminding the patient to ask the doctor to send a whitelisting request to the backend administrator. After the doctor sends the whitelisting request to the backend administrator, the backend administrator can directly enable the automatic mode function on the patient's app interface. If the doctor does not receive a response from the backend administrator within a certain period of time (e.g., 1 minute, 2 minutes, or 5 minutes), the backend administrator can send another request to enable the automatic mode function to the backend administrator, or ask the doctor to send a whitelisting request to the backend administrator. In another embodiment of the present invention, the backend administrator does not directly enable the automatic mode function on the patient's app interface, but instead sends a security code to the patient's account. The patient can use the security code to enable the automatic mode function on the app interface when needed or convenient.

[0123] In another embodiment of the present invention, when the doctor determines that the patient needs to use the automatic mode, as shown in Figure 7c, the doctor sends a request to the backend administrator to add the patient's account to the whitelist, allowing the patient to use the automatic mode function. The backend administrator receives the doctor's whitelist addition request and adds the patient's account to the whitelist list. At the same time, it sends feedback to the doctor that the whitelist addition has been completed. Further, the doctor notifies the patient that he or she can apply to use the automatic mode function. After receiving the doctor's notification, the patient sends a request to enable the automatic mode function to the backend administrator. After receiving the patient's request to enable the automatic mode function, the backend administrator directly enables the automatic mode function on the APP interface used by the patient. In another embodiment of the present invention, after the backend administrator receives the patient's request to enable the automatic mode function, it can also first verify whether the patient's account is on the whitelist. If it is determined that the patient's account is on the whitelist, the backend administrator enables the automatic mode function on the APP interface used by the patient. In another embodiment of the present invention, the backend administrator does not directly enable the automatic mode function on the APP interface used by the patient, but instead sends a security code to the patient's account. The patient can use the security code to enable the automatic mode function on the APP interface when needed or convenient.

[0124] When the doctor determines that the patient no longer needs automatic mode, the patient can turn it off on the app. The app will automatically send a message to the backend administrator, who will then remove the patient account from the whitelist. The doctor and / or patient can also send a request to the backend administrator to turn off automatic mode. The backend administrator will turn off automatic mode on the patient's app and remove the patient account from the whitelist. If the patient needs to turn automatic mode back on, the patient account will need to be re-added to the whitelist using one of the methods shown in Figures 7a-7c.

[0125] In an embodiment of the present invention, the security code sent by the backend when requesting the insulin pump function and the automatic mode function can be any number or combination of a series of numeric characters, alphabetic characters, and other symbols. It can also be a series of taps, a series of inputs, complex or simple gestures (e.g., sliding or other movements on a touch screen, drawing an image), etc. In some cases, the security code can also include a quiz or set of questions. The security code sent by the backend each time is random.

[0126] Because dietary habits vary significantly across regions and age groups, a unified blood sugar control plan may result in poor blood sugar control for some individuals. Therefore, when the automatic mode function is enabled, patients are required to enter a passcode to access different automatic mode interfaces. For individuals with high carbohydrate consumption, after entering the passcode, the automatic mode interface appears as shown in Figure 9a. A large meal option appears on the automatic mode interface, allowing patients to choose whether to enable the large meal module. After enabling the large meal mode, the infusion screen displays two options: "Regular" and "Meal," as shown in Figures 9b and 9c. Selecting "Regular" delivers insulin at a standard carbohydrate intake; selecting "Meal" delivers insulin at a higher carbohydrate intake. For individuals with low carbohydrate consumption, after entering the passcode, the automatic mode interface appears as shown in Figure 9d. The large meal option is not available, and the default setting is "Regular," which delivers insulin at a standard carbohydrate intake.

[0127] In embodiments of the present invention, the passcode can be a set of short questions, such as "Are you a carb lover?", "Is your age within the AB range?", "What is your gender?", "Where do you live?", "What are your fitness hobbies?", "Do you have any special medical conditions?", "Have you used non-automatic mode before turning on automatic mode?", etc. Based on the patient's answers, the system automatically determines whether the patient is a high carb consumer. In other embodiments of the present invention, the passcode is communicated to the patient in advance by the doctor after diagnosis, or the doctor sends information indicating whether the patient is a high carb consumer to the backend administrator at the same time as submitting a request for automatic mode whitelisting. The backend administrator then automatically assigns the patient a corresponding passcode. The passcode can be any number or combination of numeric characters, alphabetic characters, and other symbols, or a series of taps, inputs, complex or simple gestures (e.g., swiping or other movements on a touch screen, drawing an image), etc. The backend administrator can send the passcode to the patient when helping them turn on automatic mode, send the passcode to the patient at the same time as sending the security code, or send the passcode to the patient after confirming that the patient has turned on the automatic mode function using the security code. In an embodiment of the present invention, the security code and the pass code are both randomly generated, and their generation rules may be the same or different. Preferably, the generation rules of the security code and the pass code are different to avoid confusion and thus trouble for the patient.

[0128] In an embodiment of the present invention, the system or doctor may determine whether a patient has a high carbohydrate consumption based on a comprehensive judgment of the patient's age, diet, exercise habits, health status, and non-automatic mode usage results. When the patient has a non-automatic mode usage record, the non-automatic mode usage record is used as the main basis for judgment.

[0129] After the patient selects a meal, including "regular" and "large meal", the automatic mode adopts the drug infusion strategy of pre-infusion and supplementary infusion, as shown in Figure 10. During pre-infusion and supplementary infusion, the pre-infusion amount and supplementary infusion amount are divided into different levels for regular meals and large meals, as shown in Table 1 below:

[0130] The amount of insulin infused during pre-infusion is at least related to the actual blood glucose level or blood glucose rate of change during pre-infusion, as well as the estimated meal size. In other embodiments of the present invention, the pre-infusion amount may also be related to the body's IOB. Similarly, the amount of insulin infused during supplemental infusion is at least related to the actual blood glucose level or blood glucose rate of change during supplemental infusion, as well as the estimated meal size. In other embodiments of the present invention, the supplemental infusion amount may also be related to the body's IOB. In this embodiment of the present invention, meal size refers to the carbohydrate content of the meal.

[0131] The large, medium and small meal sizes corresponding to pre-infusion and the large, medium and small meal sizes corresponding to supplementary infusion are all independent parameters. The meal size corresponding to pre-infusion of the same level is larger than the meal size corresponding to supplementary infusion, but there is no fixed correspondence between the two. The system can be set according to actual needs. For large meals, the minimum meal size corresponding to pre-infusion is not less than the maximum meal size corresponding to regular meals, and the minimum meal size corresponding to supplementary infusion is not less than the maximum meal size corresponding to regular meals.

[0132] In step 1001, the patient selects the meal type at time T0 and pre-infuses a default insulin infusion amount. The default insulin infusion amount for pre-infusion can be the pre-infusion insulin infusion amount corresponding to any small, medium or large meal size. Preferably, the default insulin infusion amount for pre-infusion is the insulin infusion amount corresponding to a small meal size. Selecting a small amount of supplementary insulin infusion can prevent excessive insulin infusion and reduce the risk of hypoglycemia.

[0133] Step 1002, at time T1, compare the current blood glucose value monitored by the CGM with a preset blood glucose threshold, such as 140, 150, 160, 170, 180, 190, 200 mg / mL, etc. If the current blood glucose value is greater than the preset blood glucose threshold, the default supplemental insulin infusion amount is infused; otherwise, no supplemental infusion amount is infused, wherein the default supplemental infusion amount is the supplemental insulin infusion amount corresponding to any small, medium, or large meal size. Preferably, the default supplemental insulin infusion amount is the insulin infusion amount corresponding to a small meal size. Selecting a small initial supplemental insulin infusion amount can prevent excessive insulin infusion and reduce the risk of hypoglycemia. Time T1 may be 1 hour, 1.5 hours, 2 hours, 2.5 hours, 3 hours, etc. after time T0. In other embodiments of the present invention, whether to perform supplemental infusion can also be determined in conjunction with the blood glucose change rate at time T1.

[0134] Step 1003, within the ΔT0 time after the T0 moment, such as 3h, 4h, 5h, etc., if the patient has hyperglycemia, the next pre-infusion is upgraded, that is, the insulin infusion amount corresponding to the larger estimated meal size is infused, while considering the actual blood glucose value or blood glucose change rate, or the IOB in the body at the next pre-infusion moment; if the patient has hypoglycemia, the next pre-infusion is downgraded, that is, the insulin infusion amount corresponding to the smaller estimated meal size is infused, while considering the actual blood glucose value or blood glucose change rate, or the IOB in the body at the next pre-infusion moment; if the patient has neither hyperglycemia nor hypoglycemia, the next pre-infusion level remains unchanged, that is, the insulin infusion amount corresponding to the same meal size is infused, while considering the actual blood glucose value or blood glucose change rate, or the IOB in the body at the next pre-infusion moment.

[0135] It should be noted that, in the embodiment of the present invention, the change in the infusion level means the change in the level of the estimated meal size during infusion, while taking into account the actual blood glucose value or blood glucose change rate at the time of infusion, or the IOB in the body. Therefore, in the embodiment of the present invention, the change in the estimated meal size level also means the change in the infusion level, that is, the infusion level and the estimated meal size level can be understood to be consistent.

[0136] Step 1004: within the ΔT1 time after time T1, such as 3h, 4h, 5h, etc., if the patient has hyperglycemia, the next supplementary infusion is upgraded, that is, the supplementary insulin infusion amount corresponding to the larger estimated meal size is infused, while considering the actual blood glucose value or blood glucose change rate, or the IOB in the body at the time of the next supplementary infusion; if the patient has hypoglycemia, the next supplementary infusion is downgraded, that is, the insulin infusion amount corresponding to the smaller estimated supplementary meal size is infused, while considering the actual blood glucose value or blood glucose change rate, or the IOB in the body at the time of the next supplementary infusion; if the patient has neither hyperglycemia nor hypoglycemia, the next supplementary infusion level remains unchanged, that is, the insulin infusion amount corresponding to the same supplementary estimated meal size is infused, while considering the actual blood glucose value or blood glucose change rate, or the IOB in the body at the time of the next supplementary infusion.

[0137] In step 1005, the patient selects the meal type at time T2 and pre-infuses the insulin according to the result of step 1003, i.e., pre-infuses a larger, smaller, or unchanged insulin infusion amount corresponding to the estimated meal size, while taking into account the actual blood glucose value or blood glucose change rate at time T2, or the body's IOB.

[0138] At step 1006, at time T3, the current blood glucose level monitored by the CGM is compared with a preset blood glucose threshold, such as 140, 150, 160, 170, 180, 190, 200 mg / mL, etc. If the current blood glucose level is greater than the preset blood glucose threshold, a supplemental infusion is performed based on the result of step 1004, i.e., a supplemental insulin infusion amount corresponding to the supplemental meal amount is increased, decreased, or maintained, while also taking into account the actual blood glucose level or blood glucose rate of change, or the body's 10B at time T3. If the current blood glucose level is not greater than the preset blood glucose threshold, no supplemental infusion is performed. In other embodiments of the present invention, the determination of whether to perform a supplemental infusion may also be made in conjunction with the blood glucose rate of change at time T3.

[0139] Step 1007, within the △T0 time after time T2, such as 3h, 4h, 5h, etc., if the patient has hyperglycemia, the next pre-infusion is upgraded, that is, the insulin infusion amount corresponding to the larger estimated meal size is infused, and the actual blood glucose value or blood glucose change rate, or the IOB in the body at the next pre-infusion time is taken into consideration; if the patient has hypoglycemia, the next pre-infusion is downgraded, that is, the insulin infusion amount corresponding to the smaller estimated meal size is infused, and the actual blood glucose value or blood glucose change rate, or the IOB in the body at the next pre-infusion time is taken into consideration; if the patient has neither hyperglycemia nor hypoglycemia, the next pre-infusion level remains unchanged, that is, the insulin infusion amount corresponding to the same meal size is infused, and the actual blood glucose value or blood glucose change rate, or the IOB in the body at the next pre-infusion time is taken into consideration.

[0140] Step 1008, within the ΔT1 time after time T3, such as 3h, 4h, 5h, etc., if the patient has hyperglycemia, the next supplementary infusion is upgraded, that is, a larger supplementary insulin infusion amount corresponding to the estimated meal size is infused, while considering the actual blood glucose value or blood glucose change rate, or the IOB in the body at the time of the next supplementary infusion; if the patient has hypoglycemia, the next supplementary infusion is downgraded, that is, a smaller supplementary insulin infusion amount corresponding to the estimated meal size is infused, while considering the actual blood glucose value or blood glucose change rate, or the IOB in the body at the time of the next supplementary infusion; if the patient has neither hyperglycemia nor hypoglycemia, the next supplementary infusion level remains unchanged, that is, an insulin infusion amount corresponding to the same supplementary meal size is infused, while considering the actual blood glucose value or blood glucose change rate, or the IOB in the body at the time of the next supplementary infusion.

[0141] When pre-infusion and supplemental infusion are required at the next moment, steps 1005-1008 are repeated.

[0142] Generally, in order to maintain the stability of the patient's blood sugar level, the patient chooses the same meal type at T2 and T0, that is, if the patient chooses a regular meal at T0, then he will also choose a regular meal at T2. If he chooses a large meal at T0, then he will also choose a large meal at T2. Therefore, the next pre-infusion insulin selection and supplementary insulin infusion selection can depend on the results of the previous pre-infusion insulin selection and supplementary insulin infusion selection. However, if the patient's next meal selection is inconsistent with the previous meal selection, the patient will restore to the initial default pre-infusion amount and default supplementary infusion amount at the next pre-infusion and supplementary infusion time to prevent inaccurate insulin infusion amount due to changes in meal patterns.

[0143] It should be noted here that if the meal size corresponding to the current pre-infusion amount is already the largest meal size in the selected meal type, if hyperglycemia still occurs within the △T0 time after T0, the next pre-infusion amount will not be upgraded, and the insulin infusion amount corresponding to the large estimated meal size will still be pre-infused, while taking into account the actual blood sugar value or blood sugar change rate at the next pre-infusion time, or the in-vivo IOB. Similarly, if the meal size corresponding to the current pre-infusion amount is already the smallest meal size in the selected meal type, if hypoglycemia still occurs within the △T0 time after T0, the next pre-infusion amount will not be downgraded, and the insulin infusion amount corresponding to the small estimated meal size will still be pre-infused, while taking into account the actual blood sugar value or blood sugar change rate at the next pre-infusion time, or the in-vivo IOB.

[0144] Similarly, if the meal size corresponding to the current supplementary infusion amount is already the largest meal size in the selected meal type, if hyperglycemia still occurs within the △T0 time after T0, the next supplementary infusion amount will not be upgraded, and the insulin infusion amount corresponding to the large supplementary estimated meal size will still be pre-infused, while taking into account the actual blood sugar value or blood sugar change rate, or the IOB in the body at the time of the next supplementary infusion. Similarly, if the meal size corresponding to the current supplementary infusion amount is already the smallest supplementary meal size in the selected meal type, if hypoglycemia still occurs within the △T0 time after T0, the next supplementary infusion amount will not be downgraded, and the insulin infusion amount corresponding to the small supplementary estimated meal size will still be pre-infused, while taking into account the actual blood sugar value or blood sugar change rate, or the IOB in the body at the time of the next supplementary infusion.

[0145] In other embodiments of the present invention, the meal sizes corresponding to pre-infusion and supplementary infusion are not necessarily graded, that is, the estimated meal size during pre-infusion may only be a default size, while during supplementary infusion, the estimated meal size may be at three different levels: large, medium, and small. Therefore, during the pre-infusion stage, only the insulin infusion amount corresponding to the default meal size is infused for each pre-infusion; or the estimated meal size during pre-infusion may be at three different levels: large, medium, and small, while during supplementary infusion, only the estimated meal size may be at the default size. Therefore, during supplementary infusion, only the insulin infusion amount corresponding to the default meal size is infused for each supplementary infusion.

[0146] Similarly, the settings for the estimated meal sizes corresponding to pre-infusion and supplementary infusion in large meals and regular meals are not necessarily consistent. That is, for each meal mode, the estimated meal sizes corresponding to pre-infusion and supplementary infusion can be selected as graded or ungraded (default amount), and can be set according to the patient's actual needs.

[0147] In an embodiment of the present invention, the system is also provided with a small meal mode, namely, a snack mode. When the patient selects the snack mode, since the carbohydrate content range in snacks is relatively small, the system performs insulin infusion based on the default estimated meal amount, while taking into account the patient's actual blood sugar value or blood sugar change rate when eating snacks, as well as the IOB in the body. Moreover, the default estimated meal amount in the snack mode is smaller than the lowest level estimated meal amount in the regular mode.

[0148] Figure 11a is a schematic diagram of a closed-loop infusion system based on food image recognition according to an embodiment of the present invention. Figure 11b is a schematic diagram of a closed-loop infusion method based on food image recognition according to one embodiment of the present invention. Figure 11c is a schematic diagram of a closed-loop infusion method based on food image recognition according to another embodiment of the present invention.

[0149] In other embodiments of the present invention, quantitative analysis of food is performed to determine the content of nutrients such as carbohydrates, fats, and proteins in the food, which can further optimize the amount of insulin infusion required during meals and obtain a more accurate insulin infusion amount.

[0150] In an embodiment of the present invention, an imaging module 103 is incorporated into the system. This imaging module 103 can be used to capture food images, which are then input into the program module 101 or the cloud server 104. The food images are then quantitatively analyzed to determine the type and weight of the food, for example, 500g potatoes, 340g pasta, 220g beef, and 450g milk. The nutritional composition of the food is then determined based on the type of food, and the nutritional content of each food item is determined based on its weight. The cloud server 104 can be a large server located on a public network for storing and computing data.

[0151] Specifically, in some embodiments of the present invention, the imaging module 103 is an image acquisition device independent of the closed-loop artificial pancreas. With the development and popularization of closed-loop artificial pancreas technology, smart devices can be connected to the closed-loop artificial pancreas to assist in completing the patient's blood sugar testing and insulin infusion. Smart devices such as mobile phones, tablets, head-mounted augmented reality devices, etc. generally have camera functions. Patients can use the camera function of the smart device to take pictures of food before meals to obtain food images. Since smart devices can be iterated frequently, their image acquisition software and hardware meet the needs of daily food image acquisition, have superior performance, and the captured food images are clear, which is conducive to quantitative analysis of food images.

[0152] In other embodiments of the present invention, the imaging module 103 may be a submodule of the closed-loop artificial pancreas. For example, a camera function may be added to the PDM to obtain food images, thereby eliminating the need for additional image acquisition equipment and saving usage costs for patients.

[0153] Regardless of which device is used to photograph food, it is referred to as an imaging module 103 in the embodiment of the present invention.

[0154] In an embodiment of the present invention, the imaging module 103 is connected to the closed-loop artificial pancreas system via a wired or wireless method, and performs data interaction with at least one of the detection module 100, the program module 101, and the infusion module 102 to transmit food image data to the closed-loop artificial pancreas. Alternatively, after acquiring the food image, the imaging module 103 directly performs data analysis on the food image to determine the type and weight of the food, and then transmits the type and weight data of the food to the closed-loop artificial pancreas to determine the nutrient content. The system determines the amount of insulin infusion based on the blood glucose data and infuses it. Alternatively, after acquiring the food image, the imaging module 103 performs data analysis on the food image to determine the type and weight of the food, and then determines the nutrient content based on the type and weight of the food, and then transmits the nutrient data to the closed-loop artificial pancreas. The system determines the amount of insulin infusion based on the blood glucose data and infuses it. Alternatively, after acquiring the food image, the imaging module 103 performs data analysis on the food image to determine the type and weight of the food, and then determines the nutrient content based on the type and weight of the food, and determines the insulin infusion amount in combination with the blood glucose data, and then transmits the insulin infusion amount data to the closed-loop artificial pancreas, and the system infuses according to the insulin infusion amount data.

[0155] In an embodiment of the present invention, after acquiring the food image, the imaging module 103 can directly perform quantitative analysis on the food image in the imaging module 103, or transmit the food image to the closed-loop artificial pancreas, and the system can perform quantitative analysis on the food image, or transmit the food image to the cloud server 104, and the cloud server 104 can perform quantitative analysis on the food image.

[0156] In an embodiment of the present invention, a food image recognition model based on deep learning can be used to quantitatively analyze food images. For example, the color (R\G\B channels), shape, layer, texture, projection and other features in the food image can be identified, and the type and weight of the food can be determined by combining the shooting distance characteristics of the imaging module 103 relative to the food. In order to more accurately identify food images, a food image library can be accessed, and each food in the food image can be extracted into the food image library, and compared with the images in the food image library to determine the type of food. After determining the type and weight of the food, the type and weight data of the food are transmitted to the food nutrition library to determine the nutrients and content of the food.

[0157] In an embodiment of the present invention, the food nutrition library can only record the nutritional content of various foods, as shown in Table 2.1. After identifying the type and weight of the food in the food image, the nutritional content of each nutrient can be calculated by reading the nutritional content of the food in the food nutrition library. The food nutrition library can also record the nutritional content of each nutrient contained in a specified weight of food, as shown in Table 2.2. After identifying the type and weight of the food in the food image, the nutritional content of each food in the food image can be obtained by directly reading the food nutrition library.

[0158] Table 2.1 Food Nutrition Database

[0159] In an embodiment of the present invention, 500g of potatoes are identified in a food image, and the food nutrient database records the potato's carbohydrate content as α, fat content as β, and protein content as γ. Therefore, the carbohydrates provided by the potatoes in the food image are 500g*α, 500g*β, and 500g*γ. If 250g of beef is also identified in the same food image, and the food nutrient database records the beef's carbohydrate content as α', fat content as β', and protein content as γ', then the carbohydrates provided by the beef in the food image are 250g*α', 250g*β', and 250g*γ'. The total carbohydrates ultimately identified in the food image are 500g*α+250g*α', the total fat is 500g*β+250g*β', and the total protein is 500g*γ+250g*γ'. Based on these identified total nutrient contents, a simulated blood glucose curve for the patient is predicted, the postprandial insulin infusion amount is calculated, and the corresponding insulin infusion is completed. The above data are only for illustrative purposes. When calculating the amount of insulin infusion, algorithms such as PID, MPC, and neural networks can be used to predict the patient's simulated blood glucose curve.

[0160] Table 2.2 Food Nutrition Database

[0161] In an embodiment of the present invention, identify 400g of potatoes and 200g of beef in the food image, record potato carbon water content in the food nutrition library as a (mg), fat content as b (mg), protein content as c (mg), record beef carbon water content as d (mg), fat content as e (mg), protein content as f (mg), and the total carbon water content finally identified in the food image is a+d (mg), fat content as b+e (mg), and protein content as c+f (mg). According to these total nutrient content predictions identified, the patient simulated blood glucose curve is obtained, and the insulin infusion amount after the meal is calculated, and corresponding insulin infusion is completed. The above data are only described as an example. When calculating the insulin infusion amount, insulin algorithms such as PID, MPC, neural networks are used to predict the patient simulated blood glucose curve.

[0162] In an embodiment of the present invention, the food nutrition library can be stored in a smart device, a closed-loop artificial pancreas, or a cloud server 104, without limitation. Due to the limited storage space of the readable storage device, the food nutrition library will not record the nutritional content of all foods of a specified weight and all types. The system can display the closest food column and its corresponding nutritional content for the patient to confirm. When the patient or his / her medical guardian is able to provide more accurate nutritional content of the food, it can be input into the system and stored through the interactive interface of the smart device or PDM to update the original data recorded in the food nutrition library, or to add unrecorded raw data to the food nutrition library, where the unrecorded raw data includes food type, weight, and nutrient content.

[0163] In an embodiment of the present invention, the food nutrition library can record not only the nutritional content of carbohydrates, fats and proteins, but also the content of nutrients such as cholesterol, vitamins, trace elements, such as vitamin B, vitamin B, vitamin C, vitamin D, calcium, magnesium, iron, zinc, sodium, potassium, etc.

[0164] In the embodiment of the present invention, after identifying the nutritional content in the food image, the simulated blood glucose curve corresponding to each nutrient is calculated according to the total carbohydrate, total fat and total protein, that is, the carbohydrate blood glucose curve l α , fat and blood sugar curve β and protein blood glucose curve γ , Carbohydrate Blood Sugar Curve α , fat and blood sugar curve β and protein blood glucose curve γ It can reflect the effects of post-meal carbohydrates, fats and proteins on the patient's blood sugar. When calculating the simulated blood sugar curve, historical data and IOB can be combined. α , fat and blood sugar curve β and protein blood glucose curve γ After that, the three curves are fitted to obtain the simulated blood glucose curve l1. The simulated blood glucose curve l1 can be obtained by the carbohydrate blood glucose curve l α , fat and blood sugar curve β and protein blood glucose curve γ Obtained by linear fitting:

[0165] l1=g*l α +h*l β +i*l γ (1)

[0166] Among them, g, h, and i are the fitting coefficients of the blood glucose curve of each nutrient.

[0167] The simulated blood glucose curve l1 reflects the theoretical blood glucose changes after the nutrient content in the food image is identified by the food image recognition model, the insulin infusion amount is calculated and then infused into the patient.

[0168] In the embodiment of the present invention, the fitting of the simulated blood glucose curve l1 lasts for a period of time, for example, 2 to 5 hours. During this period, the detection module 100 obtains the patient's actual blood glucose curve l2 in real time. By comparing the simulated blood glucose curve l1 data with the actual blood glucose curve l2 data, the simulated blood glucose curve l1 can be optimized to improve the carbohydrate blood glucose curve l2. α , fat and blood sugar curve β and protein blood glucose curve γ This allows the food image recognition model to learn further and adjust the parameters of the food image recognition algorithm and the insulin algorithm to improve the accuracy of food image recognition and the accuracy of insulin infusion calculation. Specifically, after the fitting duration of the simulated blood glucose curve l1 ends, by comparing the difference between the simulated blood glucose curve l1 and the actual blood glucose curve l2 during this period, the carbohydrate blood glucose curve l2 is adjusted and optimized. α , fat and blood sugar curve β and protein blood glucose curve γ This means that the total carbohydrate, total fat and total protein contents identified by the food image recognition model before the meal are adjusted and optimized, and then the insulin algorithm parameters are corrected to improve the accuracy and applicability of the insulin algorithm parameters. Similarly, based on the aforementioned comparison differences, the food image recognition algorithm parameters in the food image recognition model can also be corrected so that the next food image recognition result will be more accurately corrected. The correction procedure for the food image recognition model algorithm parameters and the insulin algorithm parameters is repeatable, and such a correction procedure can be repeated to continuously iterate the food image recognition model algorithm and the insulin algorithm. This continuous iterative process is the deep learning process of the food image recognition model and the insulin algorithm.

[0169] In an embodiment of the present invention, the iterative correction of the food image recognition model algorithm and the insulin algorithm is completed based on the patient's actual blood glucose curve l2. When infusing insulin, since the physiological characteristics and living habits of different patients may not be the same, in the initial state, even if the food image recognition model recognizes exactly the same nutrient content and completes the corresponding insulin infusion, the actual blood glucose curve l2 finally detected by the detection module 100 will not be exactly the same. This will make the correction results of the food image recognition model parameters and the insulin algorithm parameters different. Moreover, with the repeated iterations of the correction program, the parameters of the food image recognition model will become more accurate, and the insulin algorithm parameters will gradually become the patient's unique parameters and will no longer be applicable to other patients. After repeated iterations of the parameters, the insulin algorithm also changes from the broad insulin algorithm at the factory to the patient's narrow insulin algorithm.

[0170] In this embodiment of the present invention, the patient can choose whether to upload the food image recognition model and insulin algorithm after repeated parameter iterations to cloud server 104. Alternatively, the patient can consent to the system automatically uploading the iterated food image recognition model and insulin algorithm to cloud server 104. Alternatively, the food image recognition model and insulin algorithm are running on cloud server 104, and the patient can choose to consent to their disclosure to the server backend. After obtaining a sufficient number of narrow insulin algorithms, they can be aggregated to form a narrow insulin algorithm cluster. Within a cluster, patients can be categorized into different groups based on identifiable features such as their physiological characteristics and lifestyle habits, which are the source of the model. These groups share common identifiable features. For example, patients can be divided into multiple groups based on age or age ranges, such as 0-10, 10-15, 15-18, and 18-20. Male and female patients can also be divided into two groups based on gender, with female patients further divided based on whether they are pregnant. Patients can also be divided into multiple groups based on their habitual daily exercise duration, such as 0-10 minutes, 10-30 minutes, or 30-60 minutes. Alternatively, groups can be formed based on a combination of multiple common identifiable features, such as "patients aged 10-15 with a daily exercise duration of 10-30 minutes," "male patients aged 15-18," and "pregnant patients aged 20-22." Identifiable features can also include individual insulin resistance, weight, genetic history, medical history, nationality, and occupational status. Since the patients from which the models in each group come have one or more common identifiable features, any algorithm in the group can be applied to other patients with the identifiable features to a certain extent. This provides convenience for patients with the identifiable features. Before eating, these patients can choose whether to use the food image recognition model and narrow insulin algorithm of other patients with the same identifiable features. This has higher accuracy and applicability than directly using the broad insulin algorithm, and does not require multiple corrections of the insulin algorithm parameters and image recognition model. It has higher accuracy in recognizing food images and improves the therapeutic effect of the closed-loop artificial pancreas system.

[0171] In an embodiment of the present invention, after using the group's food image recognition model and narrow insulin algorithm, a certain group of patients iterates the food image recognition model and narrow insulin algorithm after one or several meals. The iterated food image recognition model and narrow insulin algorithm parameters can also be fed back into the patient's identifiable feature group. For example, a patient with the identifiable feature of "16-year-old male patient" uses the food image recognition model to identify food images before a meal, and then uses the narrow insulin algorithm to complete insulin infusion. Five hours after the meal, the food image recognition model and narrow insulin algorithm he used complete the iteration. The patient chooses to disclose the iterated food image recognition model and narrow insulin algorithm parameters to the backend. Then, these parameters can be fed back to the group of "16-year-old patients", the group of "male patients", and the group of "16-year-old male patients".

[0172] In an embodiment of the present invention, identifiable feature groups exist in independent parallel forms. For example, "16-year-old male patient" and "16-year-old patient" are two independent groups, and applicable food image recognition models and narrow insulin algorithm parameters are stored in both groups. In these independent groups, the more identifiable features there are, the more accurate the food image recognition model and narrow insulin algorithm in the group, and the more applicable its parameters are to the patient. For example, a person whose identifiable feature is "16-year-old male patient" can find a food image recognition model and narrow insulin algorithm that are applicable to him in either the "16-year-old patient" or "16-year-old male patient" groups. Obviously, the food image recognition model and narrow insulin algorithm he found in the "16-year-old male patient" group are more applicable to him because the food image recognition model and narrow insulin algorithm in the "16-year-old patient" group have also been learned from patient-sourced data of "female patients" that do not match the physiological state of this patient, which will obviously affect the calculation of the insulin infusion amount for the "male patient".

[0173] In other embodiments of the present invention, the identifiable feature group is stored in the system or cloud server 104 in the form of a subset. For example, "16-year-old male patients" is a subordinate subset of "16-year-old patients". When searching for the "16-year-old male patients" group, it is necessary to first retrieve the "16-year-old patients" group, and then search for the "male patients" group in the "16-year-old patients" group. Obviously, among the identifiable features, age and gender are two independent features, and both can be used as subordinate subsets. When searching for the "16-year-old male patients" group, it is also possible to first retrieve the "male patients" group, and then search for the "16-year-old patients" group in the "male patients" group. Both of the above methods can ultimately point to the "16-year-old male patients" group, that is, "16-year-old patients" and "male patients" can be subsets of each other.

[0174] In an embodiment of the present invention, the identifiable feature subset is oriented to individual patients or patient groups, and its directory is editable. It is impossible for the manufacturer to pre-store all identifiable features in the system or cloud server 104 in an exhaustive manner for patients to select. Therefore, patients may not be able to retrieve the identifiable features that apply to them. In this case, the patient can edit the identifiable feature subset to add, modify, or delete the identifiable feature subset, or apply to the server to add, modify, or delete the identifiable feature subset.

[0175] In this embodiment of the present invention, due to the objective determinability of the nutritional content in food images, the parameters of the food image recognition model will tend to be consistent after multiple iterations. The food image recognition model parameters provided by each patient can be used to improve the food image recognition model and food nutrient library. By leveraging big data and deep learning to improve the food image recognition model, the accuracy of food image recognition can be improved, which is beneficial for the development of closed-loop artificial pancreas systems.

[0176] Referring to Figures 11a and 11b, in an embodiment of the present invention, in step 2001, when a patient uses a closed-loop artificial pancreas system with food image recognition, they first need to use the imaging module 103 of the smart device to capture an image of the food they are currently preparing to eat. Considering that the food may be scattered across multiple tableware when the patient is eating, the patient can capture one or more food images for each piece of food in the tableware. When capturing multiple food images, different angles can be selected to capture them, so that portions of food that may be obscured can be identified. The food image recognition model can identify foods that the patient has repeatedly captured by judging factors such as the color, shadow, shape, and size of the food in the food image, and eliminate redundant, repeatedly captured foods when calculating the nutritional content of the food.

[0177] In step 2002, after the patient takes a food image, if the food image recognition model is set on the smart device, the food image can be directly transmitted to the food image recognition application (hereinafter referred to as the application) where the food image recognition model is located. The application can be downloaded to the smart device by the patient or his / her guardian, or the application is a subroutine of the closed-loop artificial pancreas program, and the recognition of the food image can be completed in the closed-loop artificial pancreas program.

[0178] In step 2003, after the application completes the recognition of the food image, it can obtain the nutrient content (mainly including carbohydrate, fat and protein content) in the food image and display it to the patient through an interactive interface. The display content includes the recognized food type, weight, corresponding nutrient content and total nutrient content. The interactive interface can also prompt the patient whether to confirm the display content. If the patient confirms the food type, weight, corresponding nutrient content and total nutrient content obtained by the food image recognition model, it can be used to calculate the insulin infusion amount. If the patient believes that the food type, weight, corresponding nutrient content and total nutrient content obtained by the food image recognition model have obvious deviations, the patient can choose to re-recognize the food image, or re-photograph the food and recognize it until the recognition result of the food image recognition model is confirmed by the patient.

[0179] In step 2004, the application transmits the total nutrient content to the closed-loop artificial pancreas. Based on the total nutrient content, the insulin algorithm is invoked to calculate and administer the insulin infusion amount. The postprandial insulin infusion amount includes a basal dose and a bolus dose. Simultaneously, the system simulates the patient's blood glucose curve l1 for a period of time after eating based on the calculated insulin infusion amount. In this step, it is assumed that the insulin algorithm is stored in the closed-loop artificial pancreas. The insulin algorithm can also be stored in the application. After the application recognizes the food image, it can directly calculate the postprandial insulin infusion amount locally. The postprandial insulin infusion amount data is then transmitted to the closed-loop artificial pancreas, which then administers the infusion. In this step, the insulin algorithm invoked can be an insulin algorithm stored in the application or the closed-loop artificial pancreas, or an algorithm within a narrow insulin algorithm cluster. Invoking an algorithm within the narrow insulin algorithm cluster requires the patient to first identify their own identifiable features and then retrieve an appropriate insulin algorithm based on these features. The more identifiable features the patient has, the more appropriate the retrieved insulin algorithm is for the patient, and the more accurate the calculated postprandial insulin infusion amount.

[0180] In step 2005, when the postprandial insulin infusion starts, the detection module 100 detects and records the patient's actual postprandial blood glucose data to form the patient's actual blood glucose curve l2, which lasts for a period of time, such as 2 to 5 hours, which coincides with the duration of the simulated blood glucose curve l1.

[0181] In step 2006, after the detection module 100 has recorded blood glucose data for a preset period of time, the system compares the data of the simulated blood glucose curve l1 with the data of the actual blood glucose curve l2. Based on the comparison results, the insulin algorithm parameters are modified so that the simulated blood glucose curve l1 gradually converges with the actual blood glucose curve l2. The narrow insulin algorithm with the modified parameters is more suitable for the patient, and therefore the narrow insulin algorithm with the modified parameters can be referred to as the narrow insulin algorithm. At the same time, on the one hand, the system returns the comparison results to the application, which can modify the parameters of the food image recognition model to optimize the food image recognition model. The parameter-modified insulin algorithm and food image recognition model will be used for food image recognition and insulin calculation at the next meal. On the other hand, with the patient's consent, the parameter-modified insulin algorithm can be aggregated into the narrow insulin algorithm cluster and classified into corresponding groups according to the patient's identifiable characteristics, so that the patient or other patients with the same identifiable characteristics can call it before a meal.

[0182] In some embodiments of the present invention, due to the limited computing power and storage capacity of smart devices or closed-loop artificial pancreas, they may not be able to support complex image recognition and insulin infusion amount calculations, may not be able to store enough narrow insulin algorithms, and may not be able to store a sufficiently rich food nutrient library. Considering this practical problem, the cloud server 104 can be connected to the closed-loop artificial pancreas system. The cloud server 104 can establish communication with the smart device or closed-loop artificial pancreas via wired or wireless means to achieve data exchange. After connecting to the cloud server 104, the cloud server 104 can provide powerful computing power and storage capacity support for the closed-loop artificial pancreas and smart device. The image recognition model, narrow insulin algorithm cluster or food nutrient library can be stored in the cloud server 104. The cloud server 104 can also complete the recognition of food images or the calculation of insulin infusion amount, or complete the comparison of simulated blood glucose data with actual blood glucose data, and even complete the correction of insulin algorithm or image recognition model parameters. In this way, food recognition and insulin infusion amount calculation during meals can be completed without using a high-performance closed-loop artificial pancreas and smart device, saving the cost of use for patients.

[0183] After accessing the cloud server 104, the method of using the closed-loop artificial pancreas system will change. Referring to FIG. 11 c , the specific steps are as follows:

[0184] In step 3001, the patient uses a smart device to take a food image, and the food image recognition model recognizes the food image.

[0185] In step 3002, a food image recognition model is set in the cloud server 104. After taking a food image, the patient uploads the food image to the cloud server 104 via a smart device and completes food image recognition.

[0186] In step 3003, the cloud server 104 transmits the identified food type, weight, corresponding nutrient content and total nutrient content data back to the smart device, and displays it to the patient through an interactive interface. If the patient confirms the food type, weight, corresponding nutrient content and total nutrient content obtained by the food image recognition model, it can be used to calculate the post-meal insulin infusion amount. If the patient believes that there is an obvious deviation in the food type, weight, corresponding nutrient content and total nutrient content obtained by the food image recognition model, the patient can choose to re-identify the food image, or re-photograph the food and identify it until the recognition result of the food image recognition model is confirmed by the patient.

[0187] In step 3004, the patient calls the insulin algorithm, or retrieves an insulin algorithm with the same identifiable features as the patient from the narrow insulin algorithm cluster. Based on the determined insulin algorithm and nutrient content, the cloud server 104 calculates the patient's postprandial insulin infusion amount, simulates the patient's postprandial blood sugar changes, obtains simulated blood sugar data, and forms a simulated blood sugar curve l1. The cloud server 104 sends the postprandial insulin infusion amount data to the closed-loop artificial pancreas and completes the infusion. The postprandial insulin infusion amount includes a basal amount and a large dose.

[0188] In step 3005, when insulin infusion begins, the detection module 100 detects and records the patient's actual blood glucose data to form an actual blood glucose curve l2 for a period of time, such as 2 to 5 hours, which coincides with the duration of the simulated blood glucose curve l1, and transmits the actual blood glucose data during this period to the cloud server 104.

[0189] In step 3006, cloud server 104 compares the simulated blood glucose curve l1 data with the actual blood glucose curve l2 data. Based on the comparison results, it modifies the insulin algorithm parameters so that the simulated blood glucose curve l1 gradually converges with the actual blood glucose curve l2. The modified insulin algorithm is more suitable for the patient, and thus the modified insulin algorithm can be referred to as the narrow insulin algorithm. Simultaneously, cloud server 104 uses the comparison results and the modified insulin algorithm parameters to adjust the parameters of the food image recognition model to optimize the food image recognition model. The narrow insulin algorithm and food image recognition model will be used for food image recognition and insulin calculation at the next meal. With the patient's consent, the narrow insulin algorithm can be aggregated into a narrow insulin algorithm cluster and classified into corresponding groups based on the patient's identifiable features. This algorithm will replace the pre-modified insulin algorithm and be used by the patient or other patients with the same identifiable features before meals.

[0190] In the embodiment of the present invention, the calculation of the insulin infusion amount can be completed in the closed-loop artificial pancreas, or the smart device, or the cloud server 104, and the specific implementation process will not be repeated here.

[0191] In the embodiment of the present invention, the recognition of food images can be completed in a closed-loop artificial pancreas, or an intelligent device, or a cloud server 104, and the specific implementation process will not be repeated here.

[0192] In the embodiment of the present invention, the simulation calculation of the simulated blood glucose curve l1 can be completed in a closed-loop artificial pancreas, or an intelligent device, or a cloud server 104, and the specific implementation process will not be repeated here.

[0193] Refer to Figure 12, which is a schematic diagram of another closed-loop infusion method based on food image recognition according to an embodiment of the present invention. The insulin algorithm can be integrated into the food image recognition model. Once the food image recognition model identifies the type and weight of the food, the insulin algorithm can directly determine the postprandial insulin infusion amount without having to identify the nutritional content of the food, as described in detail below. The model resulting from the normalization of the food image recognition model and the insulin algorithm is referred to as the "large model" in the present invention.

[0194] In the embodiment of the present invention, the insulin algorithm can be divided into two modes: logical operation and lookup table, which are described in detail below.

[0195] In an embodiment of the present invention, the large model can be stored in a smart device, a closed-loop artificial pancreas system or a cloud server 104. After receiving the food image input, the postprandial insulin infusion amount can be directly output, and the closed-loop artificial pancreas can complete the postprandial insulin infusion.

[0196] Taking the large model stored in a smart device as an example, in an embodiment of the present invention, in step 4001, the large model still needs to use the imaging module 103 of the smart device to obtain food images. The steps for obtaining food images are consistent with the previous text and will not be repeated here.

[0197] In step 4002, the food image is transmitted to a food image recognition application, and a large model in the application recognizes the type and weight of food in the food image.

[0198] In step 4003, after the large model identifies the food type and weight in the food image, it displays the information to the patient through the interactive interface of the smart device and prompts the patient to confirm. If the patient confirms the food type and weight displayed on the interactive interface, the pre-set insulin algorithm will output the post-meal insulin infusion amount. If the patient believes that the identified food type and weight are significantly different, the patient can choose to re-identify the food image or take a photo of the food and identify it again until the large model's recognition result is confirmed by the patient.

[0199] In step 4004, after the patient confirms the type and weight of the food, the large model calculates the patient's postprandial insulin infusion amount based on the type and weight of the food using a preset insulin algorithm. The preset insulin algorithm can be one or more logical operations such as a neural network, PID, and MPC, which can calculate the postprandial insulin infusion amount based on the type and weight of the food. After recognizing the food image, the large model calculates the postprandial insulin infusion amount for each food separately, and then adds up the postprandial insulin infusion amounts of all foods to obtain the total postprandial insulin infusion amount. For example, the large model identifies 500g of potatoes, 250g of pasta, and 250g of beef in the food image provided by the patient. Using the preset insulin algorithm, it is calculated that the postprandial insulin infusion amount corresponding to 500g of potatoes is I1, the postprandial insulin infusion amount corresponding to 250g of pasta is I2, and the postprandial insulin infusion amount corresponding to 250g of beef is I3. Then, the patient's total postprandial insulin infusion amount is I1+I2+I3. The above is only a schematic description.

[0200] The preset insulin algorithm can also be a query table. Based on the query table, the postprandial insulin infusion amount is retrieved according to the type and weight of the food. The postprandial insulin infusion amount of each food and its corresponding weight is stored in the large model in the form of a query table, as shown in Table 3.1. For example, the large model identifies 300g of potatoes, 200g of pasta and 200g of beef for the food image provided by the patient. By searching the query table, it is found that the postprandial insulin infusion amount corresponding to 300g of potatoes is I4, the postprandial insulin infusion amount corresponding to 200g of pasta is I5, and the postprandial insulin infusion amount corresponding to 200g of beef is I6. Then the total postprandial insulin infusion amount of the patient is I4+I5+I6. The above is only a schematic description.

[0201] Table 3.1 Food-Insulin Lookup Table

[0202] Table 3.1 is a representation of a query table, which includes at least columns for "Food Type," "Food Weight," and "Insulin Infusion Amount." More detailed query tables may also include additional food information, such as "Food Origin," "Cooking Method," and "Food Storage Time." After capturing a food image, the patient can select this additional food-related information on the smart device's interactive interface, which helps the food image recognition model better identify the food image. This additional food information is critical for influencing the nutrient content of the food. This information may be difficult for the food image recognition model to identify, which can affect the quantitative analysis of nutrients in the food. Therefore, before the food image is recognized, the patient must input or select information related to the food as additional information.

[0203] In the large model, whether logical operations or lookup tables are used to determine the amount of post-meal insulin infusion, commonly used empirical parameters are initially used. The same food may have different effects on the blood sugar of different patients due to factors such as meal time, patient digestion level, cooking method, etc. The insulin algorithm cannot take all influencing factors into account in an exhaustive way, so the insulin algorithm can be modified.

[0204] After obtaining the patient's postprandial insulin infusion amount, the large model simulates the patient's postprandial blood glucose data based on the postprandial insulin infusion amount to form a simulated blood glucose curve l3. The simulation process lasts for a period of time, such as 2 to 5 hours.

[0205] In step 4005, the application transmits the patient's postprandial insulin infusion data to the closed-loop artificial pancreas, and the system starts postprandial insulin infusion. At the same time, the system records the actual blood glucose data detected by the detection module 100 to form an actual blood glucose curve l4. The duration of the actual blood glucose curve l4 coincides with the duration of the simulated blood glucose curve l3.

[0206] In step 4006, the system transmits the actual blood glucose data to the smart device. The large model compares the difference between the simulated blood glucose curve l3 data and the actual blood glucose curve l4 data, and corrects the simulated blood glucose curve until it is consistent with the actual blood glucose curve. The comparison result is used to correct the preset insulin logic operation parameters or query table data in the large model, and can even be used to correct the image recognition model parameters.

[0207] In an embodiment of the present invention, the insulin logic operation parameters or query table data will gradually adapt to the patient's eating habits after each correction. Therefore, the correction process of the insulin algorithm parameters in this solution is a learning process. The insulin algorithm becomes more and more suitable for the patient's physiological characteristics and living habits during the learning process. The post-meal insulin infusion amount calculated in this way will be more accurate, which will help the patient's diabetes treatment.

[0208] In this embodiment of the present invention, when modifying a lookup table, the data in the "Insulin Infusion Amount" column can be modified, and data in the "Type" and "Weight" columns can also be added, modified, or deleted. The modified lookup table can still be aggregated and grouped according to the patient's identifiable characteristics for easy access by other patients. As previously mentioned, the modified lookup table can be referred to as a narrow-sense lookup table to distinguish it from the pre-modified lookup table.

[0209] In some embodiments of the present invention, the food image recognition model and the insulin algorithm can be separated and stored in different modules. For example, the food image recognition model is stored in an intelligent device, and the insulin algorithm is stored in one of the detection module 100, program module 101, or infusion module 102 of the closed-loop artificial pancreas system. After the intelligent device completes food image recognition, the food type and weight data are transmitted to the closed-loop artificial pancreas, and the system calculates or queries the amount of insulin and completes insulin infusion. For another example, the food image recognition model is stored in a cloud server, and the insulin algorithm is stored in the intelligent device. For another example, the food image recognition model is stored in a cloud server, and the insulin algorithm is stored in one of the detection module 100, program module 101, or infusion module 102 of the closed-loop artificial pancreas system.

[0210] In an embodiment of the present invention, the order of some of the above steps can be interchanged. For example, the step of calculating the simulated blood glucose data can be set before the step of obtaining the actual blood glucose data, or it can be set after the step of obtaining the actual blood glucose data. This does not affect the calculation and infusion of postprandial insulin for the patient, nor does it affect the comparison of the simulated blood glucose data with the actual blood glucose data to correct the narrow insulin algorithm parameters.

[0211] It will be understood by those skilled in the art that any step exchange that does not affect the implementation of this solution, and both the solution before the step exchange and the solution after the step exchange should be included in the protection scope of the present invention.

[0212] It will be understood by those skilled in the art that no matter whether the food image recognition model and the insulin algorithm are normalized or separated, or how the storage location is changed, it should be included in the scope of protection of the present invention.

[0213] In summary, the present invention discloses a closed-loop infusion system and method based on food image recognition, wherein an imaging module acquires food images, and a food image recognition model analyzes the food images to acquire food data; an insulin algorithm determines the patient's postprandial insulin infusion amount based on the food data and simulates the patient's postprandial blood sugar; an artificial pancreas detects real-time blood sugar data while performing insulin infusion, and compares the real-time blood sugar data with the simulated blood sugar data; the comparison results are used to correct the parameters of the insulin algorithm, making the insulin algorithm more suitable for the patient, thereby improving the accuracy and adaptability of the insulin algorithm and benefiting the patient's diabetes treatment.

[0214] Although some specific embodiments of the present invention have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It should be understood by those skilled in the art that modifications may be made to the above embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A closed-loop artificial pancreas system, comprising: An imaging module for acquiring food images before eating; a food image recognition model, used for recognizing the food image to obtain food data, wherein the food data at least includes the type and weight of the food; an insulin algorithm, which determines a postprandial insulin infusion amount for the patient based on the food data; and Closed-loop artificial pancreas, used to complete post-meal insulin infusion and real-time blood glucose testing to obtain actual blood glucose data; After determining the postprandial insulin infusion amount, the patient's postprandial blood sugar data is simulated based on the postprandial insulin infusion amount to obtain simulated blood sugar data, and the simulated blood sugar data is compared with the actual blood sugar data. The parameters of the insulin algorithm are corrected according to the comparison result.

2. The closed-loop artificial pancreas system according to claim 1, characterized in that: The insulin algorithm after parameter correction is a narrow insulin algorithm, and the narrow insulin algorithms of different patients are aggregated to form a narrow insulin algorithm cluster.

3. The closed-loop artificial pancreas system according to claim 2, characterized in that: In the cluster of narrow insulin algorithms, the narrow insulin algorithms are classified into groups according to at least one identifiable characteristic of a patient, and the groups have the identifiable characteristic in common.

4. The closed-loop artificial pancreas system according to claim 3, characterized in that: The groups are independent of each other.

5. The closed-loop artificial pancreas system according to claim 3, characterized in that: Two of the groups having at least one common identifiable feature are subsets of each other.

6. The closed-loop artificial pancreas system according to claim 3, characterized in that: The identifiable feature is at least associated with the patient's physiological features or living habits.

7. The closed-loop artificial pancreas system according to claim 1, characterized in that: The parameters of the image recognition model are modified according to the comparison results.

8. The closed-loop artificial pancreas system according to claim 1, characterized in that: The insulin algorithm includes a logical operation or a lookup table to determine the post-meal insulin infusion amount based on the type and weight of the food.

9. The closed-loop artificial pancreas system according to claim 8, characterized in that: The insulin algorithm is integrated into the image recognition model.

10. The closed-loop artificial pancreas system according to claim 1, characterized in that: It also includes a food nutrition library, and the food data also includes the content of nutrients in the food. The food nutrition library is searched based on the type and weight of the food to determine the content of the nutrients in the food.

11. The closed-loop artificial pancreas system according to claim 10, characterized in that: Based on the content of the nutrient, the insulin algorithm determines the post-meal insulin infusion amount.

12. The closed-loop artificial pancreas system according to claim 1 or 10, characterized in that: A cloud server is also included, and the cloud server establishes communication with the imaging module or the closed-loop artificial pancreas.

13. The closed-loop artificial pancreas system according to claim 12, characterized in that: The insulin algorithm is located in the imaging module, the closed-loop artificial pancreas or the cloud server.

14. The closed-loop artificial pancreas system according to claim 12, characterized in that: The image recognition model is located in the imaging module, the closed-loop artificial pancreas or the cloud server.

15. The closed-loop artificial pancreas system according to claim 12, characterized in that: The food nutrition library is located in the imaging module, the closed-loop artificial pancreas or the cloud server.

16. The closed-loop artificial pancreas system according to claim 1, characterized in that: The imaging module is an image acquisition device independent of the closed-loop artificial pancreas.

17. The closed-loop artificial pancreas system according to claim 1, characterized in that: The imaging module is a submodule of the closed-loop artificial pancreas.

18. A closed-loop insulin infusion method, comprising providing: an imaging module, a closed-loop artificial pancreas, a food image recognition model and an insulin algorithm, characterized in that: Includes steps: Ⅰ. Before eating, use the imaging module to obtain food images; II. The food image recognition model recognizes the food image to obtain food data related to the food; III. Based on the food data, the insulin algorithm determines the patient's postprandial insulin infusion amount, and simulates the postprandial blood glucose data to obtain simulated blood glucose data; IV. The closed-loop artificial pancreas completes the postprandial insulin infusion and detects and records the actual blood sugar data; V. Compare the simulated blood glucose data with the actual blood glucose data, and modify the parameters of the insulin algorithm based on the comparison result.

19. The closed-loop insulin infusion method according to claim 18, characterized in that: In step II, there is also a step of patient confirmation of the food data. If the food data is confirmed by the patient, step III is performed. Otherwise, the process returns to step I, or the food image recognition model is used to re-recognize the food image.

20. The closed-loop insulin infusion method according to claim 18, characterized in that: In step II, the food data obtained includes at least the type and weight of the food.

21. The closed-loop insulin infusion method according to claim 20, characterized in that: The method also includes providing a food nutrition library, and determining the nutrition content of the food based on the type and weight of the food.

22. The closed-loop insulin infusion method according to claim 21, characterized in that: In step III, the post-meal insulin infusion amount is determined by the insulin algorithm based on the nutritional content of the food.

23. The closed-loop insulin infusion method according to claim 20, characterized in that: In step III, the post-meal insulin infusion amount is determined by the insulin algorithm based on the type and weight of food.

24. The closed-loop insulin infusion method according to claim 18 or 21, characterized in that: It also includes providing a cloud server, in which the image recognition model, the insulin algorithm or the food nutrition library is stored.

25. The closed-loop insulin infusion method according to claim 24, characterized in that: In step II, the food image is also uploaded to the cloud server, and food image recognition is completed in the cloud server.

26. The closed-loop insulin infusion method according to claim 24, characterized in that: In step III, it also includes uploading the food data to the cloud server and determining the post-meal insulin infusion amount in the cloud server.

27. The closed-loop insulin infusion method according to claim 26, characterized in that: The method also includes simulating the patient's postprandial blood sugar in the cloud server based on the postprandial insulin infusion amount to obtain simulated blood sugar data.

28. The closed-loop insulin infusion method according to claim 26, characterized in that: In step IV, the cloud server also transmits the postprandial insulin infusion amount data to the closed-loop artificial pancreas.

29. The closed-loop insulin infusion method according to claim 24, characterized in that: In step V, it also includes transmitting the actual blood sugar data to the cloud server, and completing the comparison between the simulated blood sugar data and the actual blood sugar data on the cloud server.

30. The closed-loop insulin infusion method according to claim 18, wherein: In step IV, the actual blood sugar data and the simulated blood sugar data are overlapped in time.

31. The closed-loop insulin infusion method according to claim 18, characterized in that: In step V, the comparison result is also used to correct the image recognition model parameters.

32. The closed-loop insulin infusion method according to claim 18, wherein: In step V, the insulin algorithm with corrected parameters is used to calculate the patient's next post-meal insulin infusion amount.

33. The closed-loop insulin infusion method according to claim 18, characterized in that: In step V, the insulin algorithm after parameter correction is a narrow insulin algorithm, and the narrow insulin algorithms of different patients are aggregated into a narrow insulin algorithm cluster.

34. The closed-loop insulin infusion method according to claim 33, characterized in that: In the cluster of narrow insulin algorithms, the narrow insulin algorithms are classified into groups according to at least one identifiable characteristic of a patient, with the identifiable characteristic being common within the group.

35. The closed-loop insulin infusion method according to claim 18, characterized in that: The postprandial insulin infusion amount includes a basal amount and a large dose.

Images