AI diet recommendation system and method based on blood glucose measurement data and biometric information

Abstract

The present invention relates to an AI diet recommendation system based on blood glucose measurement data and bio-information, and more specifically, to an AI diet recommendation system based on blood glucose measurement data and bio-information that can precisely predict future blood glucose responses occurring upon the consumption of specific foods through artificial intelligence, based on multidimensional health data integrated and collected in real time.

Description

Technology Field

[0001] The present invention relates to an AI diet recommendation system based on blood glucose measurement data and bio-information, and more specifically, to an AI diet recommendation system based on blood glucose measurement data and bio-information that can precisely predict future blood glucose responses occurring upon the consumption of specific foods through artificial intelligence, based on multidimensional health data integrated and collected in real time. Background Technology

[0002] Modern society faces a serious public health issue characterized by a rapid increase in the prevalence of chronic diseases such as diabetes, obesity, and hypertension due to population aging, Westernized dietary habits, and the widespread adoption of sedentary lifestyles. In particular, diabetes is a metabolic disease characterized by persistently abnormally high blood glucose levels; if not properly managed, it can lead to fatal complications such as cardiovascular disease, renal failure, retinopathy, and neuropathy. Therefore, effective blood glucose management is recognized as an essential task for maintaining the quality of life for diabetic patients and reducing social healthcare costs.

[0003] Against this backdrop, the importance of self-care—where patients measure and manage their own health status in daily life—is being emphasized more than ever. Driven by technological advancements, various technical means have been developed to assist in collecting and managing personal health data, such as blood sugar levels, diet, and exercise. However, conventional technologies that have been commercialized or disclosed to date each contain clear technical limitations and fundamental problems, leaving users with significant difficulties in achieving practical and sustainable health management. This invention deeply recognizes the problems of these conventional technologies and aims to present a new paradigm of technical solution to overcome them.

[0004] The first category of conventional technology consists of management tools based on the traditional Self-Monitoring of Blood Glucose (SMBG) method. This involves the user directly pricking their fingertip with a lancet to collect blood and measuring blood glucose levels using test strips and a meter. While this method, which has been the standard for diabetes management for decades, offers the advantage of allowing users to directly check their blood glucose levels at specific points in time, it also presents serious underlying issues that hinder the user's motivation for continuous management. The most significant problems are the physical pain and psychological burden associated with the measurement process. The blood collection process, repeated several times a day, not only causes direct pain to the user but also induces significant mental stress and discomfort by constantly reminding them of their status as a "patient." These negative experiences act as a primary factor in drastically reducing user compliance with measurements, ultimately leading to irregular and insufficient data collection and making effective blood glucose management impossible.

[0005] Furthermore, self-monitoring blood glucose (SMBG) methods inherently possess fundamental limitations due to the discreteness and restrictiveness of their data. This approach provides blood glucose values ​​at only a few specific points during the day (e.g., fasting, 2 hours after a meal) in the form of points, failing to display continuous information—that is, a line—regarding how blood glucose levels change between those points. Consequently, users and medical professionals cannot identify critical clinical indicators such as asymptomatic hypoglycemia occurring at night, postprandial spikes following specific food intake, or overall glycemic variability throughout the day. This is akin to trying to guess the plot of an entire movie based on only a few static stills; it leads to a distorted understanding of blood glucose patterns and carries the risk of making incorrect management decisions. Users are compelled to react passively to fragmentary results from a single point in the past, which is a critical limitation that keeps management reactive rather than proactive and preventive.

[0006] The second category of conventional technology is Continuous Glucose Monitoring (CGM), which emerged to overcome the limitations of such intermittent measurements. CGM provides a continuous trend of blood glucose changes by measuring glucose concentration in the interstitial fluid at short intervals, such as five minutes, using micro-sensors attached subcutaneously. This is clearly a revolutionary advancement compared to the traditional SMBG method. However, current CGM-based systems also face several issues that prevent users from obtaining tangible value. The main problem is that they fail to provide actionable insights amidst a flood of data. While CGM systems pour out vast amounts of time-series data in graph form, they fail to offer clear answers as to 'why' blood glucose levels changed in that way and 'what' needs to be done' to better manage future blood glucose levels. Users feel anxious as they helplessly watch their blood glucose graphs fluctuate like a roller coaster, yet analyzing the causes of these complex patterns and deciding on the next course of action remains a significant cognitive burden for the user. In other words, while the volume of data has increased explosively, there is a lack of intelligent analytical capabilities to transform that data into meaningful information and actionable knowledge.

[0007] Furthermore, CGM data often remains isolated information (siloed data). Blood glucose levels are determined by the complex interaction of numerous lifestyle factors, such as diet, exercise, sleep, and stress. Therefore, for accurate analysis and prediction of blood glucose patterns, CGM data and these diverse lifelog data must be organically integrated and analyzed from multiple perspectives. However, most conventional CGM-linked applications require users to manually input meal and exercise records in a very cumbersome manner. Real-time synchronization and integrated analysis capabilities with activity levels, heart rate, and sleep data automatically collected via wearable devices are often inadequate or non-existent. When data remains isolated on their own islands in this way, users and medical professionals cannot gain a holistic view of the complex and dynamic impact of each factor on blood glucose, ultimately resulting in only superficial analysis.

[0008] The third category of conventional technology is diet and nutrition management applications. The healthcare app market is filled with numerous apps that record the food consumed by users and calculate calories and nutritional content. While these apps may contribute to raising health awareness, they have the following obvious limitations as practical tools for managing chronic diseases. First is the extreme inconvenience and inaccuracy of data entry. Users must manually search for and input the type and quantity of food they ate for every meal, which is a very tedious and time-consuming task. This excessive burden acts as the biggest barrier that ultimately causes users to abandon continuous use. Even if users persevere with the record-keeping, significant discrepancies are likely to occur between the recorded nutritional information and the actual intake due to inaccuracies in food databases, cooking methods, or minute differences in consumption amounts.

[0009] A more fundamental problem is that these diet management apps completely disregard an individual's unique physiological responses. Operating under the assumption that the same food will have the same effect on everyone, they provide only standardized nutritional information. However, numerous recent studies clearly demonstrate the existence of extreme inter-individual variability; even when consuming the same food, post-meal blood sugar responses vary widely depending on an individual's genetic factors, gut microbiome, and insulin sensitivity. For instance, while white rice causes a rapid spike in blood sugar for some people, bread may trigger a much larger spike for others. Therefore, any diet recommendation not based on an individual's actual blood sugar response data is only valid for the "average person" and may not be optimal for specific individuals, potentially even leading to incorrect choices. This implies that the information provided by conventional diet apps is fundamentally unsuitable to be labeled as "personalized."

[0010] Finally, attempts to combine the various technologies mentioned earlier—namely, rudimentary forms of personalized healthcare platforms—also exist as conventional technologies. While these platforms offer features such as displaying CGM data alongside manually entered dietary information, they still suffer from core unresolved issues. The most critical limitation is that these systems remain at the level of description rather than prediction. Although they can provide retrospective analysis capabilities that summarize and display a user's past data, they lack the proactive function of predicting, based on scientific models, what physiological responses will occur in the future when consuming specific foods or engaging in certain activities. It is difficult to find in conventional technologies a feature that supports optimal decision-making by simulating and displaying the consequences of a choice at the "moment of decision" when a user is deliberating on what to eat.

[0011] Furthermore, the recommendation functions provided by these systems largely rely on static rule-based systems. For instance, they are based on simple logic such as "recommend low-carbohydrate foods if blood sugar is high," and they lack a highly intelligent optimization process that dynamically derives the optimal solution by comprehensively considering the user's real-time condition, previous activity level, individual food preferences, and complex blood sugar response predictions for various food candidates. Consequently, the recommendations provided by conventional technologies are inevitably superficial and general, failing to offer truly hyper-personalized solutions that perfectly match the user's complex and ever-changing circumstances.

[0012] In conclusion, conventional blood glucose and diet management technologies share a common set of multi-layered and complex problems, including (1) the pain of measurement and data fragmentation, (2) data overload and lack of actionable insights, (3) the inconvenience of data input and the provision of uniform information that ignores individual differences, and (4) the lack of future prediction and intelligent optimization functions. As a result, users have been unable to effectively utilize their health data and have been forced to manage their health in an inefficient and passive manner, relying on trial and error. Therefore, there is a critical need for a new dimension of intelligent health management system that organically integrates multidimensional personal health data collected in real time, analyzes it using advanced artificial intelligence technologies such as deep learning to precisely predict future biological responses, actively recommends diet solutions optimized for individual users based on this, and continuously learns and evolves by receiving feedback on the results. The problem to be solved

[0013] The present invention has been devised to improve upon the aforementioned problems. The purpose of the present invention is to provide an AI diet recommendation system based on blood glucose measurement data and bio-information that precisely predicts future blood glucose responses to specific food intake using artificial intelligence based on multidimensional health data (blood glucose, activity level, sleep, etc.) collected in real-time, and actively provides a diet solution optimized for the user's individual health goals and physiological characteristics based on this prediction.

[0014] In addition, according to an embodiment of the present invention, the purpose is to provide a self-evolving health management system that becomes more precisely tailored to the individual user over time and self-improves prediction accuracy and recommendation satisfaction through a closed-loop structure that continuously collects the user's actual intake results and bio-response feedback and utilizes them for the retraining of a prediction model. means of solving the problem

[0015] To achieve the above objectives, one embodiment of the present invention relates to an intelligent system that provides personalized diet recommendations, wherein

[0016] A data integration collection module that receives multidimensional health data including blood glucose data and biometric information by communicating with a continuous glucose monitor (CGM) or self-monitoring glucose monitor (BGM) attached to or carried on a user's body to generate real-time blood glucose data, and a wearable device that generates biometric information including activity level, heart rate, and sleep data;

[0017] A database storing the above-mentioned multidimensional health data, user-specific profile information, and food-specific nutritional information; and

[0018] It includes the above-mentioned data integration collection module and an AI analysis server that performs central processing in conjunction with the above-mentioned database, and

[0019] The above AI analysis server is,

[0020] An AI blood glucose prediction model that receives past time-series health data, nutritional information of the food to be consumed, and biometric information before and after the time of the meal, and predicts future blood glucose change patterns in a time-series form through deep learning operations based on Long Short-Term Memory (LSTM) or Gate Recurrence Unit (GRU); and

[0021] It includes a nutrient optimization engine that sets a target nutrient range per meal based on profile information including the user's disease type, health goals, and food preferences stored in the database, and derives food combinations that satisfy the target nutrient range while minimizing the blood glucose response predicted by the AI ​​blood glucose prediction model through an optimization algorithm.

[0022] When a diet recommendation request is received from a user terminal, the nutrient optimization engine is driven to determine an optimal food combination as a diet, and for the determined diet, an expected blood glucose response graph is generated through the AI ​​blood glucose prediction model, and customized diet information including the diet and the expected blood glucose response graph is transmitted to the user terminal.

[0023] In an AI diet recommendation method based on blood glucose measurement data and biometric information executed by a computer,

[0024] (a) A step in which a data integration collection module receives real-time blood glucose data from a continuous glucose monitor (CGM) or a blood glucose meter (BGM) and receives biometric information including activity level, heart rate, and sleep data from a wearable device to integrate and collect multidimensional health data;

[0025] (b) A step in which the AI ​​analysis server receives a diet recommendation request from a user terminal;

[0026] (c) A step in which the AI ​​analysis server responds to the request to set a target nutrient range per meal based on user profile information stored in a database, and identifies candidate food combinations that satisfy the target nutrient range among a plurality of candidate foods stored in the database;

[0027] (d) For each identified candidate food combination, the AI ​​analysis server predicts the future blood glucose change pattern upon consumption of the combination in the form of a time series graph using an AI blood glucose prediction model based on Long Short-Term Memory (LSTM) or Gate Recycling Unit (GRU);

[0028] (e) A step in which the AI ​​analysis server selects, among the blood glucose change patterns predicted in step (d), one candidate food combination predicted to have the lowest blood glucose spike as the final customized diet; and

[0029] (f) The AI ​​analysis server transmits the selected final customized diet information and the corresponding predicted blood glucose change pattern graph to the user terminal and displays it. Effects of the invention

[0030] According to one embodiment of the present invention, by predicting post-meal blood glucose response in advance through virtual meal simulation and recommending an optimal diet that minimizes blood glucose spikes, users can move away from reactive health management relying on trial and error and perform proactive and scientific management through data-based prediction, thereby effectively preventing and managing blood glucose variability in daily life.

[0031] Furthermore, according to one embodiment of the present invention, the prediction accuracy and recommendation quality of the system are progressively enhanced through a closed-loop feedback structure that continuously retrains the AI ​​model by comparing prediction results with actual measurement data. This enables the provision of hyper-personalized solutions that perfectly match an individual's changing physiological characteristics, thereby maximizing user trust and health management compliance. Brief explanation of the drawing

[0032] FIG. 1 is a diagram schematically showing the overall configuration of an AI diet recommendation system based on blood glucose measurement data and bio-information according to one embodiment of the present invention. FIG. 2 is a diagram schematically showing the relationship between components implementing a health data collection and storage process according to one embodiment of the present invention. FIG. 3 is a diagram schematically showing the relationship between components implementing a virtual meal simulation function according to one embodiment of the present invention. FIG. 4 is a diagram schematically showing the relationship between components implementing an AI-based personalized diet recommendation function according to one embodiment of the present invention. FIG. 5 is a diagram schematically illustrating the relationship between components implementing a closed-loop feedback and model retraining process according to one embodiment of the present invention. FIG. 6 is a diagram schematically showing the relationship between components that implement a professional health report provision function according to one embodiment of the present invention. Specific details for implementing the invention

[0033] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. In describing the present invention, if it is determined that a detailed description of related known technologies may unnecessarily obscure the essence of the present invention, such detailed description will be omitted. Furthermore, terms used in this specification are defined considering their functions in the present invention, and these may vary depending on the intentions or conventions of the user or operator. Therefore, the definitions of such terms should be based on the content throughout this specification.

[0034] The present invention relates to a personalized diet recommendation system and a method of operating the same, which predicts future blood glucose changes based on an individual's real-time health data and recommends an optimized diet accordingly. More specifically, the present invention integrates and analyzes blood glucose data measured from a continuous glucose monitor (CGM) or a blood glucose meter (BGM) and biometric information such as activity levels, heart rate, and sleep data collected from a wearable device. Subsequently, the integrated data is applied to a deep learning-based artificial intelligence model to predict in advance the post-meal blood glucose response that may occur after consuming a specific food. Furthermore, the core technical feature of the present invention is to construct a diet that meets optimal nutrient intake standards set according to the user's disease state or personal health goals (e.g., weight loss, muscle mass gain, etc.), and to provide the user with a customized diet or recipe in real time by synthesizing the predicted blood glucose response and the results of nutritional optimization.

[0035] FIG. 1 is a block diagram schematically illustrating the overall configuration of a personalized diet recommendation system according to one embodiment of the present invention.

[0036] Referring to FIG. 1, a personalized diet recommendation system according to one embodiment of the present invention comprises a user (10), a Dr.-Meal app (100), a data integration collection module (200), an AI analysis server (300), a database (400), and a medical expert (500). Each of the above components can transmit and receive information to and from one another through a wired or wireless data communication network.

[0037] The above user (10) is a subject who receives the service of the system according to the present invention via a terminal or the like. The above user (10) may be a patient with type 1 diabetes, a patient with type 2 diabetes, a pre-diabetic risk group with impaired glucose tolerance, or a general person who is interested in health management. The above user (10) installs the Dr.-Meal app (100) on a terminal such as a personal smartphone or tablet PC, and provides their health data to the system by wearing or carrying a continuous glucose monitor (CGM) or a self-monitoring glucose monitor (BGM) (200-1) and a wearable device (200-3) on their body. The above user (10) monitors their health status through the Dr.-Meal app (100), receives personalized diet recommendations generated by an AI analysis server (300), and can consult with a medical professional (500) if necessary.

[0038] The above Dr.-Meal app (100) is an application installed on a user's (10) smart terminal to provide a user interface (UI) and user experience (UX). The above Dr.-Meal app (100) functions as the front-end of the system of the present invention and plays a key role in mediating the exchange of information between the user (10) and the remaining components of the system. The above Dr.-Meal app (100) includes a real-time dashboard (110) function. The real-time dashboard (110) visualizes key health indicators, such as current blood sugar levels, activity levels, heart rate, and recent meal records received from the data integration collection module (200), in the form of intuitive graphs or widgets and provides them to the user (10). Through this, the user (10) can grasp their health status at a glance.

[0039] In addition, the Dr.-Meal app (100) performs an AI diet recommendation function. The function presents optimal meal menu or recipe information received from the AI ​​analysis server (300) to the user (10). The recommendation information may include food images, an expected blood sugar response graph, content of major nutrients (carbohydrates, protein, fat, sodium, etc.), calorie information, etc.

[0040] Additionally, the above Dr.-Meal app (100) can provide a virtual meal simulation function. Through this function, the user (10) can input information about the food before consuming it and receive support in making a decision regarding food selection by checking in advance a graph of post-meal blood sugar changes predicted by an AI blood sugar prediction model (300-1).

[0041] Additionally, the above Dr.-Meal app (100) may include a nutritional component analysis function, a customized health report generation function, and a community function that allows sharing information and communicating with other users (10) who have the same disease or goal.

[0042] The above data integration collection module (200) is responsible for collecting and integrating various data required for the operation of the system of the present invention from various sources and processing them into a standardized format. The above data integration collection module (200) is largely composed of a part that collects data from CGM / BGM (200-1) and a wearable device (200-3).

[0043] The above CGM / BGM (200-1) refers to a Continuous Glucose Monitoring device or a Blood Glucose Meter. A Continuous Glucose Monitoring device measures the glucose concentration in the interstitial fluid at regular intervals, such as 5 minutes or 15 minutes, through a sensor attached subcutaneously, and provides real-time blood glucose trend data. A Blood Glucose Meter is a device that measures blood glucose by the user directly collecting blood from the fingertip, etc., and primarily provides blood glucose data at a specific point in time, such as fasting blood glucose or blood glucose 2 hours after a meal. The above data integration collection module (200) receives blood glucose data measured from the above CGM / BGM (200-1) device periodically or in real time using short-range wireless communication technology such as Bluetooth or NFC (Near Field Communication).

[0044] The above-mentioned wearable device (200-3) refers to an electronic device that a user (10) wears on their body, such as a smartwatch or a smart band. The above-mentioned wearable device (200-3) has built-in accelerometer sensors, gyroscope sensors, optical heart rate sensors (PPG), etc., so that it can measure various biometric information of the user (10). The above-mentioned data integration collection module (200) collects activity data (number of steps, distance traveled, activity calories, etc.), heart rate (resting heart rate, active heart rate, heart rate variability, etc.), and sleep data (total sleep time, deep sleep, light sleep, REM sleep stage time, etc.) from the above-mentioned wearable device (200-3).

[0045] The data integration collection module (200) sorts heterogeneous data collected from the CGM / BGM (200-1) and the wearable device (200-3) through a time synchronization process and performs a preprocessing process to correct missing values ​​or remove noise. Afterward, the preprocessed data is integrated into a dataset in a form that is easy for the AI ​​analysis server (300) to analyze and transmitted.

[0046] The AI ​​analysis server (300) is a central processing unit that serves as the core brain of the system of the present invention. The AI ​​analysis server (300) analyzes large-scale time-series health data received from the data integration collection module (200) to learn meaningful patterns, and performs blood sugar prediction and diet optimization based on this. The AI ​​analysis server (300) includes an AI blood sugar prediction model (300-1) and a nutrient optimization engine (300-2) as core components.

[0047] The above AI blood glucose prediction model (300-1) is a deep learning model that predicts in advance, in the form of a graph, the pattern of post-meal blood glucose changes that will occur over a certain period of time (e.g., 2 to 3 hours) when a user (10) consumes a specific food. The above model may be implemented based on an architecture such as a Recurrent Neural Network (RNN), which exhibits excellent performance in capturing temporal dependency and long-term dependency of time-series data, a Long Short-Term Memory (LSTM) network that improves upon it, or a Gated Recurrent Unit (GRU). The input features of the above AI blood glucose prediction model (300-1) include time-series data of blood glucose changes over a certain past period, expected nutritional information of the food to be consumed (carbohydrate, protein, fat content, etc.), and biological information such as activity levels or heart rate before and after the meal. The above model receives such multivariate time-series data as input and predicts a sequence of future blood glucose levels as output.

[0048] The above nutrient optimization engine (300-2) performs the role of deriving an optimal combination of nutrients that matches the individual characteristics and goals of the user (10) and generating food or a diet that satisfies the combination. The engine first refers to the user's (10) profile information stored in the database (400). The profile information includes disease types such as Type 1 or Type 2 diabetes, specific health goals such as weight loss or muscle mass gain, and food allergies or personal food preferences. Based on the profile information, the nutrient optimization engine (300-2) sets a range of daily or meal-specific target nutrients (total calories, appropriate ratio of carbohydrates, protein, and fat, upper limit of sodium, etc.) recommended for the user (10). Subsequently, the engine searches the vast food nutrition information library built in the database (400) to find a combination of foods that minimize the blood sugar spike predicted by the AI ​​blood sugar prediction model (300-1) while satisfying the set target nutrient range. Optimization techniques such as linear programming or genetic algorithms can be used in this process.

[0049] The above database (400) is a component that stores and manages all data necessary for the operation of the system of the present invention. The above database (400) systematically stores personal profile information of the user (10) (age, gender, disease information, health goals, etc.), raw health data and preprocessed data collected through the data integration collection module (200), meal records directly entered by the user (10) through the Dr.-Meal app (100), blood sugar prediction results and diet recommendation logs generated by the AI ​​analysis server (300), and standardized nutritional information for hundreds of thousands of foods. The data accumulated in the above database (400) is reused as training data to continuously improve the performance of the AI ​​blood sugar prediction model (300-1) and is utilized as base data to provide personalized services.

[0050] The medical professional (500) mentioned above refers to a terminal possessed by personnel with specialized knowledge in healthcare, such as doctors, nutritionists, and nurses. The medical professional (500) can monitor the health data reports of a specific user (10) through a separate web or app interface for administrators. Through this, the medical professional can review whether the recommendations from the AI ​​analysis server (300) are medically valid, or perform the role of providing more professional and individualized counseling or coaching to high-risk users (10). For example, the medical professional can provide additional advice on the diet recommended by the AI, or deliver professional feedback on medication use or lifestyle habits based on changes in the user's blood sugar patterns. The intervention of the medical professional (500) increases the reliability of the system and helps the user (10) manage their health more safely and effectively.

[0051] Hereinafter, the overall flow of a personalized diet recommendation method according to an embodiment of the present invention, which operates based on the above components, is described in detail step by step.

[0052] (S1) App launch and dashboard lookup request step:

[0053] First, the user (10) runs the Dr.-Mill app (100) on their smart device. When the app is launched, an action is performed to request a lookup of the real-time dashboard (110) located on the main screen.

[0054] (S2) Real-time Dashboard Function Activation and Data Request Steps:

[0055] In response to the request of the user (10), the Dr.-Meal app (100) activates the real-time dashboard (110) function. At the same time, in order to display the latest health data on the screen, the Dr.-Meal app (100) sends a signal to the data integration collection module (200) requesting the transmission of the latest health data.

[0056] (S3) Health Data Collection and Integration Step:

[0057] The data integration collection module (200), upon receiving a data request signal from the above Dr.-Meal app (100), initiates a procedure to collect data from connected devices. The data integration collection module (200) receives the latest blood glucose data from the CGM / BGM (200-1) attached to the user (10) via Bluetooth communication, etc. At the same time, the data integration collection module (200) collects biometric information, such as recent activity levels, current heart rate, and sleep information, from the wearable device (200-3) worn by the user (10). The collected heterogeneous data are sorted and integrated based on timestamps to form a single data packet, which is then transmitted to the Dr.-Meal app (100) and simultaneously transmitted to the AI ​​analysis server (300) and the database (400).

[0058] (S4) AI-based analysis and recommendation generation step:

[0059] The AI ​​analysis server (300) generates a customized diet recommendation by comprehensively analyzing health data updated in real-time from the data integration collection module (200) and historical data stored in the database (400). For example, when a user (10) wants to receive a recommendation for the next meal menu, the AI ​​analysis server (300) considers the user's (10) current blood sugar status, recent activity level, etc.

[0060] At this time, the nutrient optimization engine (300-2) operates first. The engine calculates the optimal amount and ratio of carbohydrates, protein, and fat to be consumed in this meal according to the user's (10) disease type and health goals.

[0061] Next, the AI ​​blood glucose prediction model (300-1) performs a virtual intake simulation for various food candidate groups that satisfy the above-mentioned optimized nutrient composition. That is, it predicts the pattern of post-meal blood glucose change that will occur when each food candidate is consumed.

[0062] Finally, the AI ​​analysis server (300) selects the food combination that shows the most stable and low spike in the predicted blood sugar response as the optimal recommended diet. In this process, the food preference data of the user (10) stored in the database (400) may be taken into account.

[0063] (S5) Personalized Diet Information Provision Step:

[0064] The customized diet recommendation information generated by the AI ​​analysis server (300) is transmitted to the Dr.-Meal app (100). The Dr.-Meal app (100) visualizes and presents the received information in a form that is easy for the user (10) to understand. For example, it outputs a screen that includes a photo of the recommended menu, an expected blood sugar response graph, and detailed nutritional information.

[0065] (S6) Feedback Collection and Model Retraining Step:

[0066] After the user (10) actually consumes the recommended diet, they record the meal in the Dr.-Meal app (100). Subsequently, the system collects the actual measured post-meal blood glucose response data through CGM / BGM (200-1). The AI ​​analysis server (300) compares and analyzes the blood glucose value predicted by the AI ​​blood glucose prediction model (300-1) with the actual measured blood glucose value. If a significant error occurs between the two, the error information is used as new training data to periodically update and retrain the AI ​​blood glucose prediction model (300-1). Through this closed-loop feedback structure, the system adapts more precisely to the user's (10) unique physiological response over time and continuously improves the accuracy of the prediction.

[0067] As described above, the system and method according to the present invention are clearly distinct from existing healthcare technologies in that they go beyond merely recording and displaying past data to ‘predict’ future biological responses based on real-time data and actively ‘recommend’ optimal solutions according to individual goals. The present invention can significantly contribute to supporting systematic self-management of diabetic patients and improving their quality of life, and furthermore, is expected to play an important role in reducing social medical costs caused by chronic diseases and shifting the medical paradigm from a ‘treatment’-centered approach to a ‘prevention and management’-centered approach.

[0068] Next, another embodiment of the present invention will be described in detail with reference to the attached drawings. The same reference numerals are assigned to components identical to those in the previously described embodiment, and redundant descriptions thereof will be omitted.

[0069] FIG. 2 is a flowchart illustrating a detailed process in which health data is collected and stored in a database (400), particularly centered around a data integration collection module (200), in a personalized diet recommendation system according to one embodiment of the present invention. The process of FIG. 2 is a core operation for securing data that forms the basis of the system and corresponds to a step of establishing a foundation for accurate analysis and recommendation by an AI analysis server (300).

[0070] In order for the system of the present invention to operate stably, it is very important to collect and systematically manage various health data generated from the user (10) in real time without omission. FIG. 2 specifically illustrates the collection and storage path of blood glucose data, in particular, among these data flows.

[0071] (1) In the step of requesting and transmitting real-time blood glucose data collection,

[0072] The data integration collection module (200) acts as a data gateway for the system and periodically communicates with connected external measurement devices to manage data collection. The data integration collection module (200) transmits a signal requesting data transmission to the CGM / BGM (200-1) attached to the user's (10) body or measured by the user, according to a pre-set schedule or the occurrence of a specific event. For example, in the case of a continuous glucose monitor (CGM), the data integration collection module (200) may request data collection at regular intervals, such as 5 minutes or 15 minutes. The request signal may be transmitted via a low-power wireless communication protocol such as Bluetooth Low Energy (BLE).

[0073] Upon receiving the data transmission request signal, the CGM / BGM (200-1) generates a data packet containing the latest blood glucose measurement value stored internally and transmits it in response to the data integration collection module (200). The transmitted blood glucose information may include not only the glucose level but also additional metadata such as timestamp information indicating the exact time at which the level was measured, a unique identification number of the measuring device, and battery status. This information is used to ensure the reliability of the data and to verify the consistency of the data.

[0074] In a similar manner, the data integration collection module (200) communicates with the wearable device (200-3) in parallel or sequentially with the CGM / BGM (200-1) to perform the process of collecting biometric information such as activity level, heart rate, and sleep data. That is, the blood glucose data collection process illustrated in FIG. 2 represents a standard procedure that can be commonly applied to various types of data collected in the present invention.

[0075] (2) To explain the storage step of the collected data,

[0076] The data integration collection module (200) receives various types of data from the CGM / BGM (200-1) or wearable device (200-3) and then proceeds with a procedure to immediately store the data in the database (400). The data integration collection module (200) may perform a series of preprocessing steps on the received raw data. For example, it may include a process of converting the data format into a standard format that fits the schema of the database (400), identifying outliers by checking the valid range of the data values, and attaching user identification information to clarify which user (10) the data was collected from.

[0077] The data from the above preprocessing process is transmitted to a database (400) for permanent storage and systematic management. The database (400) accurately records and indexes the received data according to the user's (10) account and chronological order. The data stored in this way is utilized for various important purposes. First, it is used as source data to visualize and display health data trends from the past to the present on the real-time dashboard (110) of the Dr.-Meal app (100). Second, it becomes part of a large-scale training dataset for the AI ​​blood glucose prediction model (300-1) within the AI ​​analysis server (300) to learn the unique physiological characteristics of the individual user (10) and continuously improve the performance of the model. Third, it is provided as basic data for medical professionals (500) to monitor and analyze changes in the user's (10) long-term health status.

[0078] (3) In the data storage completion response step,

[0079] After the database (400) successfully stores the data received from the data integration collection module (200), it transmits a storage completion response signal back to the data integration collection module (200) indicating that the storage operation has been successfully completed. This response signal is an important procedure for detecting errors that may occur during the data transmission and storage process and for ensuring the integrity of the data.

[0080] If the data integration collection module (200) fails to receive a storage completion response from the database (400) within a certain period of time, this may indicate a network error during the data transmission process or a temporary failure of the database (400). In such a case, the data integration collection module (200) may perform an error handling logic to prevent data loss by storing the data in a temporary buffer and attempting to retransmit it after a certain period of time. Through this verification and retry mechanism, the system of the present invention significantly improves the stability and reliability of data collection.

[0081] In conclusion, the data collection and storage process described in detail in FIG. 2 is the most basic prerequisite for the personalized diet recommendation system of the present invention to provide accurate and reliable services to the user (10). The high-quality personal health data accumulated in real time through the above process becomes the driving force that enables the AI ​​analysis server (300) to perform sophisticated prediction and optimization calculations, and constitutes a core foundation that determines the performance of the entire system.

[0082] Next, another embodiment of the present invention will be described in detail with reference to the attached drawings. The same reference numerals are used for components identical or similar to those described above, and redundant descriptions thereof may be omitted for the sake of brevity.

[0083] FIG. 3 is a flowchart illustrating the detailed operation process of a ‘virtual meal simulation’ function, which is one of the core functions provided in a personalized diet recommendation system according to an embodiment of the present invention, in steps. The virtual meal simulation function is a preventive health management tool that supports a user (10) in making healthier food choices by checking in advance, through artificial intelligence, the effect that the food will have on their blood sugar before actually consuming a specific food.

[0084] First, as a preliminary step for the simulation function to be performed, the system undergoes a process of identifying the user's (10) current state. Specifically, as described in FIG. 2 above, the data integration collection module (200) continuously collects and integrates real-time health data from the CGM / BGM (200-1) and the wearable device (200-3). The latest integrated data is transmitted to the Dr.-Meal app (100). Based on the received data, the Dr.-Meal app (100) visualizes the user's (10) key health status, such as current blood sugar levels, recent activity levels, and heart rate, as intuitive graphs or numerical values ​​on a real-time dashboard (110).

[0085] (S3) Step requesting prediction of blood sugar changes upon consuming specific food:

[0086] A user (10) who is aware of their current health status through the above dashboard may want to check in advance the effect that a specific food will have on blood sugar before eating. At this time, the user (10) inputs the type and amount of a specific food, such as 'one apple', through the interface of the Dr-Meal app (100), and sends a request to the system to predict the expected change in blood sugar when consuming the food.

[0087] (S4) Step to enable virtual meal simulation function:

[0088] When a prediction request is input from the user (10), the Dr.-Meal app (100) internally activates a virtual meal simulation (120) function and begins to perform the relevant procedures. The virtual meal simulation (120) function is a module that oversees a series of processes, including processing the user request, receiving the prediction result through communication with the AI ​​analysis server (300), and displaying it to the user (10).

[0089] Subsequently, the Dr.-Meal app (100) sends a prediction request message to the AI ​​analysis server (300) to perform AI-based prediction. The prediction request message includes various information to increase the accuracy of the prediction. Specifically, first, a unique identifier of the user (10) who is the subject of the prediction; second, the most recent health status data received from the data integration collection module (200) (e.g., current blood sugar, activity intensity over the past hour, etc.); and third, nutritional information of the food ('apple') entered by the user (10). The nutritional information of the food is obtained by the Dr.-Meal app (100) by querying the standard nutritional components (carbohydrate, sugar, dietary fiber, protein, fat content, etc.) of the food internally or through linkage with the database (400).

[0090] The AI ​​analysis server (300) that receives the above prediction request calls an AI blood glucose prediction model (300-1) to perform actual prediction calculations based on the data included in the request. As previously described, the AI ​​blood glucose prediction model (300-1) is implemented with a deep learning architecture specialized in learning patterns of time-series data, such as a recurrent neural network (RNN) or a long short-term memory (LSTM). The model uses the current state of the input user (10) and the nutritional components of the food to be consumed as input features to perform calculations that predict the trend of changes in blood glucose levels over a certain period of time (e.g., after 15 minutes, 30 minutes, 60 minutes, 120 minutes) in a time-series form. Since the model has parameters optimized for the individual based on the user's (10) past meal records and actual blood glucose response data accumulated in the database (400), it can derive highly personalized prediction results rather than general predictions.

[0091] When the prediction operation is completed, the AI ​​blood glucose prediction model (300-1) returns the raw data (e.g., a sequence of predicted blood glucose values ​​by time period) necessary to generate a graph of predicted blood glucose changes to the AI ​​analysis server (300). The AI ​​analysis server (300) then finally transmits the returned prediction result data back to the Dr.-Mill app (100).

[0092] Finally, the Dr.-Meal app (100) that receives the prediction results processes them into a visual form that the user (10) can easily understand and displays them on the screen. For example, it outputs a line graph that allows the user to see the expected blood glucose changes over time at a glance. The graph may also provide specific numerical information such as the expected peak glucose, the time to reach peak glucose, and the expected blood glucose after 2 hours.

[0093] Through this virtual meal simulation function, the user (10) can recognize in advance whether their blood sugar will rise rapidly or remain stable when eating an 'apple'. If the predicted blood sugar spike is at a concerning level, the user (10) can take active and preemptive health management actions, such as adjusting the intake amount or choosing other foods with a milder blood sugar response. This has significant technical significance in that it helps to make optimal decisions based on data-driven scientific evidence, moving away from the conventional method of learning diets through trial and error.

[0094] FIG. 4 is a flowchart illustrating in detail the overall operation flow of the 'AI-based personalized diet recommendation' function, which is the most core value provided by the system of the present invention. This function goes beyond simply calculating calories or nutrients and listing foods; it is a highly personalized service that dynamically reflects the user's real-time health status and individual long-term goals, and predicts future blood sugar responses that may occur upon food intake to actively suggest optimal food combinations.

[0095] (S5) Lunch Menu Recommendation Request Step:

[0096] A user (10) using the system of the present invention may request a diet recommendation by running the Dr.-Meal app (100) to obtain professional guidance on what to eat, for example, when it is time for lunch. The request is made through explicit user interface operations, such as pressing the 'Get Diet Recommendation' button within the app.

[0097] (S6) AI Diet Recommendation Feature Activation Steps:

[0098] When a request for a diet recommendation is received from the user (10), the Dr.-Meal app (100) internally activates the AI ​​diet recommendation (130) function and communicates with the AI ​​analysis server (300) to initiate a procedure for generating an optimal diet.

[0099] Subsequently, the Dr.-Meal app (100) transmits a data packet requesting an optimal diet recommendation to the AI ​​analysis server (300). The data packet contains very specific and multi-layered information. First, a unique identifier of the user (10) who is the subject of the recommendation. Second, real-time health status information most recently updated from the data integration collection module (200), such as current blood sugar levels, fasting time before meals, morning activity levels, etc. Third, personal profile information of the user (10) stored in the database (400), including medical conditions such as Type 1 or Type 2 diabetes, specific health goals such as weight loss or muscle gain, and allergy information or data on preferences or dislikes regarding specific foods.

[0100] The AI ​​analysis server (300) that receives the above recommendation request sequentially performs two stages of sophisticated artificial intelligence analysis and optimization.

[0101] The first step is the blood glucose response prediction step. The AI ​​analysis server (300) first selects a group of candidate foods suitable for the current meal (lunch) from a vast food library stored in the database (400). Subsequently, for each of the selected candidate foods, the AI ​​analysis server (300) instructs the AI ​​blood glucose prediction model (300-1) to predict the post-meal blood glucose response that will occur if the user (10) consumes the food. The AI ​​blood glucose prediction model (300-1) performs a virtual consumption simulation for each candidate food based on the user's (10) current state and transmits the prediction results (e.g., expected peak blood glucose level, shape of the blood glucose response curve, etc.) back to the AI ​​analysis server (300).

[0102] The second step is the nutrient and blood sugar response-based optimization step. The AI ​​analysis server (300) now has two key pieces of information. One is the nutritional goal for this meal specified in the user's (10) personal profile (e.g., 40-50g carbohydrates, 20g or more protein, 600mg or less sodium, etc.), and the other is the blood sugar prediction result for each candidate food obtained in the first step. The AI ​​analysis server (300) transmits both of these pieces of information to the nutrient optimization engine (300-2).

[0103] The above-mentioned nutrient optimization engine (300-2) solves complex optimization problems based on the received information. That is, it performs calculations to find the best combination of foods that are expected to have the most stable predicted blood sugar response and low blood sugar spikes while accurately satisfying the set nutritional target range. The engine does not simply return a list of foods, but calculates the appropriate intake amount for each food to derive a result in the form of a completed 'optimized diet' or a specific 'recipe', and returns this to the AI ​​analysis server (300).

[0104] Finally, the AI ​​analysis server (300) transmits the final customized meal recommendation received from the nutrient optimization engine (300-2) to the Dr.-Meal app (100). The Dr.-Meal app (100) visualizes the received recommendation and displays it on the screen so that the user (10) can understand it easily and find it attractive. The final meal display screen provides the name and photo of the recommended menu, a detailed recipe, total calories and key nutrient information, and, most importantly, a graph showing the expected post-meal blood sugar change when consuming this meal. Additionally, during this process, user preference information accumulated in the database (400), such as prioritizing ingredients that the user (10) has previously given positive feedback on or excluding non-preferred ingredients, can be reflected in the composition and display of the final recommendation.

[0105] Through this series of processes, the present invention can provide the user (10) with an intelligent diet solution that is most suitable for their current physical condition, contributes to the achievement of their personal long-term health goals, and further takes taste and satisfaction into account.

[0106] Next, specific embodiments of the present invention will be described in detail with reference to the attached drawings. The same reference numerals are used for components identical to or equivalent to those described in the previously explained embodiments, and redundant descriptions thereof are omitted or simplified for clarity.

[0107] FIG. 5 is a flowchart conceptually illustrating the 'closed-loop feedback and model retraining' process, in which the system according to the present invention does not stop at providing simple one-time information but learns on its own and enhances performance through continuous interaction with the user. The above process corresponds to a key mechanism that enables the system to evolve into an adaptive artificial intelligence system that becomes more precisely tailored to each individual user (10) over time.

[0108] (S7) Feedback input step after meal intake:

[0109] After the user (10) consumes the diet recommended by the AI ​​analysis server (300) according to the process of FIG. 4, after a certain amount of time has elapsed, the user inputs feedback on the diet through the Dr.-Meal app (100). The feedback may include two main types of data.

[0110] The first is subjective satisfaction data regarding taste, satiety, cooking convenience, etc. Users (10) can convey their subjective experiences to the system in the form of star ratings, selecting 'like' or 'dislike', or simple text comments.

[0111] The second is objective actual response data based on dietary intake. The most important part of the above data is the actual measured value of blood sugar change after a meal. The above actual blood sugar data is automatically collected from the user's (10) CGM / BGM (200-1) through the data integration collection module (200). That is, the system obtains both the blood sugar change value predicted in advance by the AI ​​blood sugar prediction model (300-1) and the blood sugar change value actually measured.

[0112] The above Dr.-Meal app (100) integrates the subjective satisfaction data and objective actual response data into a single feedback record and transmits and stores it in a database (400). In the database (400), information regarding which user (10) consumed which recommended diet and when, what the predicted result was, and what the actual blood sugar response and subjective satisfaction were is systematically accumulated as a complete pair.

[0113] (S8) Model retraining and performance enhancement step:

[0114] When a significant amount of feedback data is accumulated in the database (400), the AI ​​analysis server (300) performs a model retraining procedure periodically or when specific conditions are met. The retraining process aims to continuously enhance the performance of the AI ​​models included in the system.

[0115] Specifically, in the retraining process of the AI ​​blood glucose prediction model (300-1), the AI ​​analysis server (300) retrieves data pairs of 'predicted blood glucose values' and 'actually measured blood glucose values' from the database (400). The model calculates the error between the predicted value and the actual value and proceeds with training through an algorithm such as backpropagation, which fine-tunes the weights and parameters within the model in a direction that minimizes the error. By repeating this process, the AI ​​blood glucose prediction model (300-1) more accurately identifies the unique physiological characteristics of the user (10) and sensitivity to specific foods, which leads to a result of gradually improving the accuracy of future blood glucose predictions.

[0116] Additionally, the performance of the nutrient optimization engine (300-2) can also be enhanced through the feedback data. For example, if a user (10) consistently expresses low satisfaction with a recommended diet containing a specific food ingredient, the engine can learn to lower the preference weight for that food ingredient. Through this, when recommending future diets, food combinations that are low in preference to the user (10), even if they are excellent in terms of nutrition or blood sugar management, can be adjusted to a lower priority, while food combinations that are high in preference are given priority, thereby increasing the practical acceptance of the recommendation.

[0117] In conclusion, the feedback and relearning loop illustrated in FIG. 5 is a key device that makes the system of the present invention like a living organism. The more the user uses the system and provides feedback, the smarter the system becomes and the more it is optimized for the individual. This virtuous cycle structure induces long-term use by the user (10), maximizes service satisfaction, and ultimately serves as the foundation for realizing successful, data-driven personalized health management.

[0118] FIG. 6 is a flowchart illustrating the operation process of a ‘health report for experts’ function that goes beyond supporting the health management of an individual patient (10), who is the end user of the present invention, to enabling a medical professional (500) managing the patient to remotely monitor the patient’s condition and provide professional advice based on data. This function plays an important role in enhancing the reliability of the system and maximizing treatment effects through linkage with the actual clinical environment.

[0119] (S9) Patient Health Report Inquiry Request Step:

[0120] A medical professional (500) who has obtained prior consent for information provision from a user (10) may want to check the health status of a specific patient through a professional web or app interface installed on their terminal (e.g., PC, tablet). At this time, the medical professional (500) selects a specific user (10) from a list of patients under management and requests to view the user's weekly or monthly health report. In this specification, the professional interface is described as a function included in the Doctor-Meal app (100) for convenience of explanation.

[0121] (S10) Step to activate the customized health report feature:

[0122] The Doctor-Mill app (100), upon receiving a request to view a report from the medical professional (500), activates a customized health report (140) function designed to systematically provide clinically significant information to the professional.

[0123] When the above function is activated, the Dr.-Meal app (100) sends a query to the database (400) requesting the retrieval and analysis of user data for a specific period to generate a report. The query includes a unique identifier of the user (10) who is the subject of the report and information on the period of the report (e.g., last 7 days, last 1 month). The request is characterized by including an 'analysis' request to analyze the data for the period and derive statistical indicators and patterns, rather than simply retrieving the raw data as is.

[0124] The database (400) that receives the above analysis request, or a separate analysis module linked to the database (400), extracts data from a designated user (10) for a designated period and performs a comprehensive analysis. The above analysis may include the following items. First, an analysis of blood glucose data, such as average blood glucose, blood glucose variability index (Standard Deviation or Coefficient of Variation), time in range (TIR), frequency of occurrence of hyperglycemia or hypoglycemia, and pattern analysis by time period. Second, an analysis of diet data, such as the average calorie intake, average intake ratio of carbohydrates, protein, and fat, and sodium intake for the corresponding period. Third, an analysis of activity data, such as the daily average number of steps or activity time. Fourth, an analysis of correlations, such as a pattern in which hyperglycemia frequently occurs after consuming a specific food or an analysis of the correlation between exercise volume and blood glucose stability.

[0125] When the above analysis is completed, the database (400) generates analyzed report data including derived statistics, patterns, and data necessary for graph generation, and returns it to the Dr.-Meal app (100).

[0126] Finally, the Doctor-Meal app (100) receives the returned report data and generates and provides a visually organized weekly or monthly report so that a medical professional (500) can grasp the patient's condition at a glance. The report may include various visual elements such as a blood sugar trend graph, a pie chart showing the ratio of nutrient intake, and a summary table of key statistical indicators.

[0127] Through this, medical professionals (500) can accurately evaluate the patient's lifestyle management status based on objective data accumulated over a certain period, rather than fragmentary information. In addition, they can quantitatively determine whether the diet recommended by the AI ​​or the patient's actual diet has a positive effect on blood sugar management, and if necessary, provide professional feedback to the user (10) regarding drug dosage adjustment or more specific lifestyle correction. As such, the professional report function of the present invention facilitates communication between patients and medical staff and provides an important foundation for realizing data-based precision medicine. Explanation of the symbols

[0128] 10 : Users 100 : Dr.-Mill App 110: Real-time Dashboard 120: Virtual Meal Simulation 130: AI Diet Recommendation 140: Personalized Health Report 200 : Data Integration Collection Module 200-1 : CGM / BGM 200-3 : Wearable Device 300 : AI Analysis Server 300-1: AI Blood Sugar Prediction Model 300-2: Nutrient Optimization Engine 400 : Database 500 : Medical professionals

Claims

Claim 1 In an intelligent system for providing personalized diet recommendations, the system comprises: a data integration collection module that receives blood glucose data packets from a continuous glucose monitor (CGM) or a self-monitoring glucose monitor (BGM) attached to the body of a user, including a patient with type 1 or type 2 diabetes or a general user, which measures glucose concentration in interstitial fluid at set intervals of 5 or 15 minutes, or from which the user measures through blood sampling; and receives biometric information including activity level, heart rate, and sleep stages (deep sleep, light sleep, REM sleep) from a wearable device embedded with an accelerometer, a gyroscope, and an optical heart rate sensor (PPG); and generates a standardized multidimensional health dataset by performing time synchronization based on timestamps on the received heterogeneous data and performing a preprocessing process to correct missing values ​​or remove outliers; personal profile information including the user's age, gender, disease information, health goals (weight loss, muscle mass gain), allergy information, and food preferences; preprocessed health data received from the data integration collection module; and the user A database that stores meal records entered from a terminal and standardized nutritional information including carbohydrate, protein, fat, and sodium content for hundreds of thousands of foods, and manages data by indexing it according to user identification information; and an AI analysis server that performs central processing in conjunction with the data integration collection module and the database;The AI ​​analysis server includes time series data of blood glucose changes over a past period, expected nutritional information of the food to be consumed, and biometric information of activity level and heart rate before and after the meal as input features, and an AI blood glucose prediction model that predicts a sequence of future blood glucose levels to appear in a time series form for the next 2 to 3 hours after consuming the food through a recurrent neural network (RNN) architecture based on Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU) capable of learning the temporal dependency and long-term dependency of the time series data; A nutrient optimization engine that sets a target nutrient range per meal, including the ratios of calories, carbohydrates, protein, and fat recommended for the user, by referring to the user's profile information stored in the database; primarily selects multiple candidate food combinations that satisfy the set target nutrient range by searching the food nutrition information library of the database; derives a predicted blood sugar spike value by performing a virtual intake simulation by inputting each selected candidate food combination into the AI ​​blood sugar prediction model; and derives the optimal food combination and intake amount that minimize blood sugar elevation while satisfying nutritional goals.The AI ​​analysis server is equipped with the following: upon receiving a diet recommendation request from a user terminal, it determines the optimal food combination derived by operating the nutrient optimization engine as the recommended diet; transmits visualized customized diet information, including an expected blood glucose response graph generated by the AI ​​blood glucose prediction model and nutritional component information for the diet, to the user terminal; the AI ​​analysis server calculates a time-series error between the actual post-meal blood glucose data collected through the data integration collection module after the user consumes the recommended diet and the blood glucose value previously predicted by the AI ​​blood glucose prediction model; and periodically retrains the AI ​​blood glucose prediction model by performing a backpropagation algorithm to fine-tune the weights and parameters within the AI ​​blood glucose prediction model in a direction that minimizes the calculated error; additionally, by receiving satisfaction feedback regarding subjective taste and satiety of the diet entered by the user through the user terminal, it lowers the preference weight of the nutrient optimization engine for ingredients with low satisfaction and raises the preference weight for ingredients with high satisfaction, thereby simultaneously improving the accuracy of the prediction model and personalizing the recommendation engine. An AI diet recommendation system based on blood glucose measurement data and bio-information characterized by constructing a dual-structured closed-loop feedback system. Claim 2 delete Claim 3 delete Claim 4 delete Claim 5 In an AI diet recommendation method using an AI diet recommendation system based on blood glucose measurement data and biometric information according to claim 1, the method comprises: (a) a step in which a data integration collection module receives real-time blood glucose data from a continuous glucose monitor (CGM) or a self-monitoring blood glucose monitor (BGM) and receives biometric information including activity level, heart rate, and sleep data from a wearable device to integrate and collect multidimensional health data; (b) a step in which an AI analysis server receives a diet recommendation request from a user terminal; (c) a step in which the AI ​​analysis server responds to the request to set a target nutrient range per meal based on user profile information stored in a database, and identifies candidate food combinations that satisfy the target nutrient range among a plurality of candidate foods stored in the database; (d) a step in which the AI ​​analysis server predicts, for each identified candidate food combination, a future blood glucose change pattern upon consumption in the form of a time series graph using an AI blood glucose prediction model based on Long Short-Term Memory (LSTM) or a Gated Recycling Unit (GRU); and (e) a step in which the AI ​​analysis server finally selects one candidate food combination predicted to have the lowest blood glucose spike among the blood glucose change patterns predicted in step (d). The method includes a step of selecting a diet; and (f) a step in which the AI ​​analysis server transmits the selected final customized diet information and the corresponding predicted blood glucose change pattern graph to the user terminal and displays them, and after step (f), a step of collecting actual post-meal blood glucose data through the data integration collection module after the user consumes the final customized diet; and a step in which the AI ​​analysis server calculates the error between the collected actual post-meal blood glucose data and the blood glucose data predicted in step (d).The method for recommending an AI diet using an AI diet recommendation system based on blood glucose measurement data and biometric information is further characterized by including the step of the AI ​​analysis server retraining the AI ​​blood glucose prediction model based on the calculated error to update the model parameters, and the step of performing the retraining further includes the step of additionally receiving subjective satisfaction feedback regarding the diet entered by the user through the user terminal and updating the nutrient optimization engine to reflect the subjective satisfaction as a weight when selecting the final customized diet in step (e); the step of receiving a request to view a health report for a specific user from a medical professional's terminal; the step of the AI ​​analysis server extracting and analyzing the user's health data for a specified period from the database to generate a report including the time-in-retention ratio (TIR), blood glucose variability, and patterns between the diet and blood glucose; and the step of transmitting the generated report to the medical professional's terminal. Claim 6 delete Claim 7 delete Claim 8 delete Claim 9 delete Claim 10 delete

Images