KR102338964B1 - Apparatus and method for symtome and disease management based on learning - Google Patents

Abstract

The present invention is configured to derive appropriate action information that can improve musculoskeletal symptoms by analyzing the user's information and status using a reinforcement learning algorithm, and then provide it to the user. It relates to devices and methods for symptom and disease management. A learning-based symptom and disease management apparatus related to an example of the present invention includes an input unit for receiving a first input composed of a plurality of state values related to a user's state, a database in which a plurality of action information is stored, and the inputted first input. Accordingly, a first set of action information is selected from the plurality of action information, and a plurality of evaluation values corresponding to each of the first set of action information are calculated using a predetermined model for the first input, and calculation a processing unit that selects first recommended action information that is at least one of the first set of action information based on the plurality of evaluation values and an output unit configured to provide the selected first recommended action information to the user .

Description

Apparatus and method for symptom and disease management based on learning {APPARATUS AND METHOD FOR SYMTOME AND DISEASE MANAGEMENT BASED ON LEARNING}

The present invention relates to a learning-based symptom and disease management apparatus and method, and more particularly, by using a reinforcement learning algorithm to analyze user information and status to derive appropriate action information to improve musculoskeletal symptoms, It is configured to provide to a user, and relates to a learning-based symptom and disease management apparatus and method in which such a series of processes are repeated.

The environment in which a person's repetitive daily life actions and actions take place has a high degree of involvement in disease occurrence and symptom manifestation. Since it changes from time to time, if you measure it, time series information is continuously generated over time. If information that can be related to diseases and symptoms can be extracted from among these information, processed and analyzed and visualized as summarized information, can be used to evaluate musculoskeletal health and provide recommended actions to improve musculoskeletal health , it can have the effect of improving musculoskeletal health by making the user newly aware of his or her health condition and paying attention to specific actions, and the doctor does not need to directly observe the patient's posture and movement and the surrounding environment Since the patient's information can be checked in a short time, it is possible to efficiently perform the cause identification, diagnosis, removal of cause, treatment and education of the problem reflecting the patient's daily life behavior and surrounding environment. In addition, users can check and receive feedback on bad posture and motion before symptoms worsen or disease develops, so that the effect of preventing the onset or exacerbation of symptoms and diseases can be expected. When providing a recommended action for improving musculoskeletal health, along with a rule that determines the recommended action that is previously set, the rule itself is learned according to the user's status information input through reinforcement learning and is continuously updated to learn the user's individual status It has the advantage of being able to provide an effective recommended action for

Republic of Korea Patent Publication No. 10-1774752

The present invention has been devised to solve the conventional problems as described above, and derives appropriate action information that can improve musculoskeletal symptoms based on information about the user's health condition and improvement by using a reinforcement learning algorithm. An object of the present invention is to provide a learning-based symptom and disease management device and method configured to provide

Specifically, reinforcement learning is used based on user information for the management of musculoskeletal symptoms or diseases, and for symptom and disease management, which utilizes the user's health status, the user's body posture and motion information, and information about the surrounding environment. Therefore, there has been no report on the management method and technology of the system, which is operated as a recommendation system that is continuously learned and updated.

Until now, life-related information acquisition in the medical field has been achieved in a way that, when a patient visits a hospital, a doctor asks a question about life-related information related to a disease, obtains oral information of the patient, and records it in an electronic medical record. Due to the limitations of this method, it is difficult to collect information about the patient's lifestyle and habits, the surrounding environment for living, etc. It is difficult to obtain a sufficient amount of information because it must be collected within. Therefore, doctors focus on the diagnosis and treatment of the current symptoms, and are unable to acquire enough information on the patient's lifestyle and habits, the surrounding environment, etc. Preventive measures, such as blocking the progression and exacerbation of symptoms or disease in advance, may be delayed, and even if you visit a hospital without taking preemptive measures, the underlying cause may not be resolved and the disease may continue to progress.

In the case of recommendation of management and control methods based on existing medical and medical knowledge related to symptoms or diseases, the recommendation model itself is fixed by recommending management and control methods for symptoms or diseases to users based on existing literature or knowledge There is no room for learning new knowledge or models as users' information is accumulated because knowledge and models are not updated with respect to the reaction or changes in health status, but in the case of this method and system, the It has the advantage of updating and refining the recommendation algorithm for the user's state through the reinforcement learning-based model update by collecting behaviors and results and reflecting them again in time series.

In the case of the method and system for the collection, processing, analysis, and health level evaluation of the user's health and biometric information proposed in the present invention, the user who is not aware of the disease in relation to the information about the user and the external environment is newly It has the effect of making people aware of or paying attention to their health status. In addition, since these users, including existing patients, are non-patients, but periodically or aperiodically aware of health abnormalities, and those who want to manage their health, the effect of preventing symptoms through health care before worsening or onset of diseases can bring

The collected information is structured so that the information is acquired, and the user's health information is structured and stored without requiring additional processing such as natural language processing for clinical notes described in the electronic medical record of the clinical site. It has the advantage of being able to accurately and conveniently deliver information to clinical medical staff by calling it or reconfiguring it through visualization.

Existing online-based disease diagnosis systems provide users with information on suspected diseases, disease information, and pharmacies/clinics/hospitals to visit after temporarily collecting health or biometric information. However, such a system can be effective in blocking the progression of a single disease, but there is a limit in that it is difficult to have an effect on the continuous management of living factors and living factors that are the root cause of musculoskeletal diseases. On the other hand, the existing method and system using three-dimensional biomechanical data does not deal with the fact that information about the user's existing health condition and the body's posture and motion can be determined in relation to surrounding objects. It uses only the information obtained when a specific measuring tool is installed, and includes only measured values with a time axis, so it does not include various types of measurement without a time axis and information on daily life activities and surrounding environment. It does not include information about the torso such as the neck and waist, which is important for posture and movement, including only the upper and lower body. In addition, there is a disadvantage that the user's judgment is applied collectively rather than customized because the rules are fixed.

On the other hand, the technical problems to be achieved in the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned are clearly to those of ordinary skill in the art to which the present invention belongs from the description below. can be understood

A learning-based symptom and disease management apparatus related to an example of the present invention for realizing the above-described problem includes: an input unit configured to receive a first input composed of a plurality of state values related to a user's state; a database in which a plurality of action information is stored; A first set of action information is selected from the plurality of action information according to the input first input, and a plurality of pieces of action information corresponding to each of the first set of action information are selected by using a predetermined model for the first input. a processing unit that calculates an evaluation value and selects first recommended action information that is at least one of the first set of action information based on the plurality of evaluation values calculated; and an output unit for providing the selected first recommended action information to the user.

In addition, the first input received by the input unit and the first recommended action information selected for the first input by the processing unit may be configured to be stored in the database.

Also, the processing unit may update the model based on the update information.

Also, the update information may include a state value of the first input stored in the database and information on the first recommended action selected with respect to the first input.

In addition, the update information may further include a state value of the second input received in the input unit.

In addition, the processing unit may be configured to calculate a plurality of evaluation values for the second input by using the model updated by the learning module.

Also, the plurality of evaluation values for the second input may be used to select second recommended action information for the second input.

Also, the first input may be received at a first time, and the second input may be received at a second time after the first time.

In addition, the update of the model may adjust a method in which the plurality of evaluation values are calculated with respect to the second input in order to induce the user's condition to be improved.

In addition, the processing unit calculates a plurality of rewards representing the effect of applying each of the first set of action information on the state of the user, and the calculated plurality of rewards are used to calculate the plurality of evaluation values. can be reflected in

Also, the processing unit may assign a higher weight to the action information in which a higher reward is expected among the first set of action information.

In addition, the processing unit may identify an evaluation value greater than a threshold value among the plurality of calculated evaluation values, and select action information corresponding to the identified evaluation value as the first recommended action information.

In addition, the plurality of rewards may be limited at least in part by clinical information input through the input unit or clinical information stored in the database.

Also, when it is determined that at least some of the plurality of rewards do not match the clinical information, the processing unit may calculate the reward based on the clinical information.

In addition, the selection of the first set of action information may be limited at least in part by clinical information input through the input unit or clinical information stored in the database.

In addition, when it is determined that at least a portion of the selected first set of action information does not match the clinical information, the first set of action information may be reselected based on the clinical information.

In addition, the processing unit may analyze the musculoskeletal symptoms of the user using the first input, and the analyzed musculoskeletal symptoms of the user may be used to calculate the plurality of evaluation values for the first input. .

In addition, analyzing the musculoskeletal symptoms of the user may include searching for factors exacerbating the musculoskeletal symptoms or factors improving the musculoskeletal symptoms from the plurality of state values of the first input.

In addition, analyzing the musculoskeletal symptoms of the user may include predicting the diagnosis, progression state or prognosis of the musculoskeletal symptoms based on the searched factors that aggravate the musculoskeletal symptoms or improve the musculoskeletal symptoms. have.

In addition, the output unit may be configured to provide a factor that aggravates the musculoskeletal system symptom or a factor that improves the musculoskeletal system symptom to the user.

In addition, the plurality of state values may include information about the user's daily life behavior, information about the user's posture, information about the user's operation, information about the environment around the user, biometric information of the user, and the It may include at least one of the user's musculoskeletal health information.

In addition, the information on the user's daily living behavior includes the position, length, and angle of the user's body part in the daily living behavior, and changes according to the time, and the type, time, intensity, method, and type of the user's activity. including at least one of these patterns, and the information on the user's posture includes at least one of a position, a length, an angle, and a change of the user's body part in a static state with time, the The information on the user's motion includes at least one of a position, length, and angle of the user's body part in a dynamic state, and changes over time thereof, and the information on the environment around the user includes the It includes at least one of the location, size, height, distance, intensity, and patterns thereof of any object within a preset radius from the user, and the user's biometric information includes the user's age, gender, family history, past history, and present including at least one of treatment, past treatment, skeletal muscle mass, weight, height, and genetic information, wherein the user's musculoskeletal health information includes symptoms of pain, location of pain, frequency of pain, duration of pain, It may include at least one of an aspect of pain and a pattern thereof.

Also, the first input may be generated by the user's input, by an input other than the user, or by sensing of a sensor.

On the other hand, the learning-based symptom and disease management method related to an example of the present invention for realizing the above-described task, the method comprising: receiving a first input composed of a plurality of state values related to a user's state; selecting a first set of action information from a plurality of action information stored in a database according to the first input; calculating a plurality of evaluation values corresponding to each of the first set of action information by using a predetermined model with respect to the first input; selecting first recommended action information that is at least one of the first set of action information based on the plurality of calculated evaluation values; and providing the selected first recommended action information to the user.

In addition, the received first input and the first recommended action information selected for the first input may be configured to be stored in the database.

Also, in the calculating, the model may be updated based on the update information.

Also, the update information may include a state value of the first input stored in the database and information on the first recommended action selected with respect to the first input.

In addition, the update information may further include a state value of the second input received in the input unit.

In addition, it may be configured to calculate a plurality of evaluation values for the second input by using the model updated by the step of applying the reinforcement learning algorithm.

Also, the plurality of evaluation values for the second input may be used to select second recommended action information for the second input.

Also, the first input may be received at a first time, and the second input may be received at a second time after the first time.

In addition, the update of the model may adjust a method in which the plurality of evaluation values are calculated with respect to the second input in order to induce the user's condition to be improved.

In addition, the calculating includes calculating a plurality of rewards representing an effect of applying each of the first set of action information on the state of the user, wherein the plurality of rewards calculated in the calculating step Compensation may be reflected in calculating the plurality of evaluation values.

Also, in the calculating, a higher weight may be given to action information in which a higher reward is expected among the first set of action information.

In addition, the calculating includes identifying an evaluation value greater than a threshold value among the plurality of calculated evaluation values, and the selecting includes action information corresponding to the evaluation value identified in the identifying step. It may include selecting the first recommended action information.

In addition, the plurality of rewards may be limited at least in part by clinical information input through the input unit or clinical information stored in the database.

In addition, when it is determined in the calculating step that at least some of the plurality of rewards do not match the clinical information, the reward may be calculated based on the clinical information.

In addition, in the selecting step, the selection of the first set of action information may be limited at least in part by clinical information input through the input unit or clinical information stored in the database.

In addition, when it is determined that at least a part of the first set of action information selected in the selecting does not match the clinical information, the first set of action information may be reselected based on the clinical information.

In addition, the calculating includes analyzing the musculoskeletal symptoms of the user using the first input, and the musculoskeletal symptoms of the user analyzed in the analyzing step include the plurality of musculoskeletal symptoms for the first input. An evaluation value can be calculated.

In addition, the analyzing of the musculoskeletal symptoms of the user may include searching for factors exacerbating the musculoskeletal symptoms or factors improving the musculoskeletal symptoms from the plurality of state values of the first input.

In addition, the analyzing of the musculoskeletal symptoms of the user is based on the factors exacerbating the musculoskeletal symptoms or the factors that improve the musculoskeletal symptoms found in the searching step. Diagnosis, progress or prognosis of the musculoskeletal symptoms It may further include the step of predicting.

In addition, the step of outputting may include providing the user with a factor that aggravates the musculoskeletal symptoms or a factor that improves the musculoskeletal symptoms.

In addition, the plurality of state values may include information about the user's daily life behavior, information about the user's posture, information about the user's operation, information about the environment around the user, biometric information of the user, and the It may include at least one of the user's musculoskeletal health information.

In addition, the information on the user's daily living behavior includes the position, length, and angle of the user's body part in the daily living behavior, and changes according to the time, and the type, time, intensity, method, and type of the user's activity. including at least one of these patterns, and the information on the user's posture includes at least one of a position, a length, an angle, and a change of the user's body part in a static state with time, the The information on the user's motion includes at least one of a position, length, and angle of the user's body part in a dynamic state, and changes over time thereof, and the information on the environment around the user includes the It includes at least one of the location, size, height, distance, intensity, and patterns thereof of any object within a preset radius from the user, and the user's biometric information includes the user's age, gender, family history, past history, and present including at least one of treatment, past treatment, skeletal muscle mass, weight, height, and genetic information, wherein the user's musculoskeletal health information includes symptoms of pain, location of pain, frequency of pain, duration of pain, It may include at least one of an aspect of pain and a pattern thereof.

Also, the first input may be generated by the user's input, by an input other than the user, or by sensing of a sensor.

On the other hand, a non-transitory computer-readable storage medium related to an example of the present invention for realizing the above-described problem is a non-transitory computer-readable storage medium including instructions, the instructions, when executed, cause one or more processors to : to receive a first input composed of a plurality of state values related to a user's state, to select a first set of action information from a plurality of action information stored in a database according to the inputted first input, and to select the first set of action information; to calculate a plurality of evaluation values corresponding to each of the first set of action information by using a predetermined model for an input, and based on the calculated plurality of evaluation values, at least one of the first set of action information to select the first recommended action information, and to provide the selected first recommended action information to the user.

The present invention uses a reinforcement learning algorithm to analyze and learn actions that can improve the user's musculoskeletal health status, symptoms, and diseases, and is continuously updated. may provide a user with an apparatus and method for managing symptoms and diseases.

Specifically, musculoskeletal disease-related medicine (orthopedics, rehabilitation medicine, etc.), occupational environmental medicine and medical expertise based on underlying health conditions, body posture and motion information in repetitive daily life activities, and information about surrounding objects By using this, the user can check bad posture and motion before symptoms worsen or a disease develops, and thus, the effect of preventing disease through health management can be expected. Classify and identify factors that can aggravate or improve musculoskeletal disorders by grade, predict the occurrence and diagnosis of musculoskeletal disorders, disease course and prognosis according to related factors, and improve musculoskeletal disorders by controlling related factors You can receive recommendations on how to manage your health, and you can use relevant information such as the user and the surrounding environment that change over time to check whether the user's life is improving in the direction of preventing musculoskeletal diseases. function can be

In addition, in the case of the existing recommendation system related to symptoms or diseases, the recommendation model itself, including the classification and recommendation rules, is not updated but fixed by recommending adjustment and management methods to users based on established medical knowledge. However, in the case of the method and system proposed by the present invention, the existing medical knowledge is set as the initial value of the model and utilized, but as users accumulate, the health level, condition, and recommendation results of users are affected. It is meaningful in that it seeks to improve more effectively than the existing method by updating the management and control method suitable for the user's situation by reflecting the user's various responses and increasing the accuracy.

In addition, a health management method and system that can recommend a health management method that can improve musculoskeletal disorders by controlling symptoms and disease-related factors, which can be linked with actual treatment in a clinical setting by providing relevant factors to medical institutions is comprised of Because the doctor can check the patient's related information in a short time, using it, the patient's life is reflected and the cause of the problem can be identified, diagnosed, and the cause can be removed and treated efficiently.

In addition, in many cases of musculoskeletal disorders, micro-damages in the musculoskeletal system accumulate over time and become chronic. By preventing chronic musculoskeletal diseases, it is possible to preemptively maintain and improve personal health, and it has a great effect on reducing individual and national medical expenses.

On the other hand, the effects obtainable in the present invention are not limited to the above-mentioned effects, and other effects not mentioned will be clearly understood by those of ordinary skill in the art to which the present invention belongs from the following description. will be able

The following drawings attached to this specification illustrate a preferred embodiment of the present invention, and serve to further understand the technical spirit of the present invention together with the detailed description of the present invention, so the present invention is limited to the matters described in such drawings It should not be construed as being limited.
1 shows an example of a block diagram of a learning-based symptom and disease management apparatus that can be applied to the present invention.
2 is a flowchart illustrating an embodiment of a learning-based symptom and disease management method that can be implemented according to the present invention.

Hereinafter, a preferred embodiment of the present invention will be described with reference to the drawings. In addition, one embodiment described below does not unreasonably limit the content of the present invention described in the claims, and it cannot be said that the entire configuration described in the present embodiment is essential as a solution for the present invention.

In addition, the same reference numerals are used throughout the drawings for parts having similar functions and functions. Throughout the specification, when a part is said to be 'connected' to another part, this includes not only 'directly connected' but also 'indirectly connected' with another element interposed therebetween. do. In addition, 'including' a certain component does not exclude other components unless otherwise stated, but means that other components may be further included.

The present invention operates to prevent musculoskeletal symptoms or diseases, such as pain, or to identify a cause that causes or aggravates a disease, and to effectively remove or control it. According to the present invention, the user can check what factors related to his or her musculoskeletal system disease and be provided with an explanation thereof, and can receive explanations on how to remove or control the factors that caused the musculoskeletal disease, alleviation or exacerbation. You can receive information on factors related to yourself and the health of the musculoskeletal system in a comprehensive, systematic, and easy to understand format. If the individual consents and the severity is severe, information is provided to medical institutions so that they can receive health-related consultations or lead to actual treatment. The present invention allows individuals to regularly input health-related information and health result information, continuously collects relevant data, forms a database, and continuously updates the user's status and recommendation plan using reinforcement learning. It can be used alone or in combination with an existing treatment plan by suggesting a management and control method suitable for the user's condition. Using health-related data, it is possible to research, establish prevention and health promotion strategies and policies, and provide customized information and recommendation services according to conditions.

Hereinafter, the learning-based symptom and disease management apparatus 100 to be proposed by the present invention will be described in detail with reference to the drawings.

1 shows an example of a block diagram of a learning-based symptom and disease management apparatus that can be applied to the present invention.

Referring to FIG. 1 , the learning-based symptom and disease management apparatus 100 according to the present invention includes an input unit 10 , a database 20 , a processing unit 30 , an output unit 40 , a power supply unit 50 and a communication unit. (60) and the like. However, since the components shown in the drawings are not essential, the learning-based symptom and disease management apparatus 100 having more or fewer components may be implemented.

Hereinafter, the components will be described in turn.

The input unit 10 receives an input from a user. Here, the input may be composed of a plurality of state values related to the user's state.

The plurality of state values may include information pertaining to the user, such as information on daily living activities and information on posture and motion. In addition, the plurality of state values may include information on the environment around the user, biometric information of the user, musculoskeletal health information, and the like. The type of information includes direct input information and measurement information of a user in an input method.

Living behavior includes the daily life activities of modern people, including sedentary life in which people sit for a long time on the floor or chair, which are the main lifestyles of modern people, and a life in which they are absorbed in the display while in front of a desk or on the go. , specific work styles such as housework, and lifestyle habits such as diet and exercise. Life-related information is quantitative and qualitative data, and data that can be obtained through measurement using rulers, protractors, and camera images such as length and angle, etc., types of user activity that can be obtained through surveys, etc. It includes data on time, intensity, method, etc., and data measured in real time using sensors during life, and includes their patterns.

The information on the user's posture includes the location of each part of the user's body or their patterns in a static state when the user performs daily life, and the information on the motion is in a dynamic state when the user is living It includes the location of each part of the user's body or a pattern thereof. Here, the information on posture and motion includes quantitative data measured using a sensor for position, length, angle, and their change over time, and information that can be obtained through surveys, drawings, and the like. Here, the body parts include the upper body including the head, neck, chest, shoulders, arms, hands, etc., the lower body including the pelvis, thighs, knees, calves, feet, etc., and the entire body, such as the torso such as the back and waist.

The information about the environment around the user includes information about an arbitrary object within a specific range from the user when the user lives. Here, the information of the surrounding environment includes quantitative or qualitative data in which the location, size, height, distance, intensity, and changes over time are measured, and information that can be obtained through surveys, drawings, etc. and their patterns includes

The user's biometric information includes the user's age, gender, family history, past history, present and past treatment, skeletal muscle mass, weight, height, genetic information, etc., and includes information that can be used for treatment in clinical settings.

The user's musculoskeletal health information is information containing the user's current health state according to the user's input. For example, symptoms, location, frequency, duration, pattern, and patterns thereof of pain may be considered as an example of the user's health information.

Postures and movements in life are closely related to surrounding objects. Since living postures and movements can be greatly affected by various factors such as the size, height, and strength of surrounding objects, information about the environment around the user should be used as important information for life together with body information. Accordingly, the present invention utilizes both the user's own information and the interaction information between the user and the environment by utilizing both body information and information on the surrounding environment related to daily life.

The input may be generated by a user's input, by an input other than the user, or by sensing of a sensor.

In addition, the database 20 may store a program for processing and control of the processing unit 30 , and may perform a function for storing input/output data. In the database 20, not only a plurality of action information, but also a plurality of evaluation values for an input received in the input unit 10 or an input calculated by the processing unit 30, recommended action information selected by the processing unit 30, etc. may be stored. have.

The action information may be defined as a series of actions capable of alleviating symptoms or diseases of the user's musculoskeletal system. A plurality of action information is stored in the database 20, and at least some of the action information determined to be suitable for the user among the plurality of action information may be recommended to the user.

The database 20 is configured to be able to store changes in information over time. That is, the database 20 may store the first input received at the first time, a plurality of evaluation values for the first input, and the first action information selected for the first input, and the first input after the first time. The second input received at two times, a plurality of evaluation values for the second input, and second action information selected for the second input may be stored.

In addition, the processing unit 30 calculates a plurality of evaluation values for the input received by the input unit 10 . The plurality of evaluation values correspond to each of a plurality of action information pre-stored in the database 20 . That is, when action information is applied to an input, an expected user's state is calculated as an evaluation value, and a result of applying each of the plurality of action information to one input corresponds to each of the plurality of evaluation values.

The processing unit 30 may analyze the user's condition, particularly, the user's musculoskeletal symptoms by using a plurality of state values included in the input. The user's musculoskeletal symptoms analyzed in this way may be used to calculate a plurality of evaluation values. Furthermore, the processing unit 30 may search for factors that aggravate musculoskeletal symptoms or factors that improve musculoskeletal symptoms from the plurality of state values.

The processing unit 30 utilizes a methodology in the field of machine learning and artificial intelligence to find factors related to health status among user and environmental information and classify levels. Rule-based classification, case-based reasoning, decision tree, support vector machine, logistic and linear regression, k-nearest neighbor, artificial neural network, boosting, etc. Based on this, characteristics such as the patient's current condition and risk factors are classified and the future condition is predicted. This classification is based on data-based learning such as supervised learning using existing labeling and unsupervised learning that classifies by setting its own standards, and guidelines using textbook knowledge of clinical care. and protocol-based, answer-based learning, and preferably uses data-based learning and correct-answer-based learning together.

User input information can be mapped in multiple dimensions through dimensional transformation algorithms such as auto-encoder and PCA, and the mapped vector is used as a feature vector. It is possible to evaluate the health status and level from the obtained user's information, and find out what factors are related to the symptoms or diseases of the user's own musculoskeletal system by searching for factors related to symptoms or diseases in a customized way according to the matched feature vector.

Also, in order to determine the action information to be provided to the user, the processing unit 30 may select at least one action information from among a plurality of action information based on a plurality of calculated evaluation values. For example, the action information may be selected in such a way that a higher weight is given to the action information in which a higher reward is expected among the first set of action information. Also, for example, the processing unit 30 identifies an evaluation value greater than a threshold value or an evaluation value greater than or equal to a threshold value among a plurality of calculated evaluation values, and determines as action information for providing action information corresponding to the identified evaluation value to the user. may be Also, as another example, action information may be probabilistically provided by varying or comparing the size of the evaluation value itself.

The processing unit 30 may include a learning module 32 that applies a reinforcement learning algorithm. For example, the reinforcement learning algorithm of the learning module 32 may include a plurality of state values of the first input stored in the database 20 and a first value stored in the database 20 in the process of calculating a plurality of evaluation values for the second input. 1 Allow multiple evaluation values for input to be utilized. Also, in the process of calculating the plurality of evaluation values for the second input, a plurality of state values of the second input may be additionally utilized. By application of the reinforcement learning algorithm, a method in which a plurality of evaluation values are calculated with respect to the second input in a direction in which improvement of the user's state is induced may be adjusted.

The learning module 32 optimizes the recommendation of a management and adjustment method according to the user's state value through reinforcement learning. Using the user's feature vector as the state, the options that the user can select in each state are defined as actions (action information), and the health level (evaluation value) that the user can have according to the state and behavior is Q(s) , a) is defined as When changing from a specific state s1 to another specific state s2 through a specific action a1, the value obtained by the user from the health level is defined as a reward, and Q(s,a) is calculated according to the reinforcement learning method. It finds a set of maximizing actions and presents it to the user. The action here belongs to a method for controlling and managing musculoskeletal symptoms or diseases.

Methods for managing or controlling symptoms or diseases may vary depending on the patient's condition, and initial values for Q(s,a) and behavior, condition, and reward are set in advance so that existing medical knowledge can be reflected and reinforced It progresses in a way that is updated through learning. For example, such an update reflects the difference between the predicted level and the actual level of the user's state at the time of the next input or measurement according to the user's state and behavior, and the more it deviates from the predicted state, the more meaningful information is considered and the prediction error An update is made to Q(s,a) in such a way as to reduce . In addition, Q(s,a) can be directly adjusted by the judgment of a medical institution or medical staff, in addition to automatic update.

In addition, the output unit 40 is for generating an output related to visual, auditory or tactile sense, and this may include a display unit, a sound output module, an alarm unit, a haptic module, a projector module, and the like. The output unit 40 performs a function for providing the action information selected by the processing unit 30 to the user. Through this, the output unit 40 may sequentially recommend a customized health management method to the user.

The output unit 40 may perform a function to provide a factor for exacerbating musculoskeletal symptoms or a factor for improving musculoskeletal symptoms searched for by the processing unit 30 to the user. Through the output unit 40, the user can check what his/her musculoskeletal symptoms or disease-related factors are and receive an explanation thereof, and a method of removing or controlling the factors that have caused or aggravated the musculoskeletal symptoms or diseases can be explained, and information on related factors and musculoskeletal health status can be provided. Daily life information is stored in the database 20, and is provided to the user in the form of a visualized report by analyzing the data by the method previously. The contents of the result report consist of a list of risk factors presented in the order of suspicion, each suspected musculoskeletal disease, and the descriptions, descriptions of how these risk factors can be eliminated or controlled, short-term and long-term goals, long-term and short-term plans to achieve those goals, and actions that users can implement in their daily lives as they relate to their daily lives.

In addition, the power supply unit 50 receives external power and internal power under the control of the processing unit 30 to supply power necessary for the operation of each component.

In addition, the communication unit 60 is a network in which the learning-based symptom and disease management apparatus 100 and the wireless communication system or the learning-based symptom and disease management apparatus 100 and the learning-based symptom and disease management apparatus 100 are located. It may include one or more modules that enable wireless communication between them. For example, the communication unit 60 may include a broadcast reception module, a mobile communication module, a wireless Internet module, a short-range communication module, and a location information module.

The mobile communication module transmits/receives a radio signal to and from at least one of a base station, an external terminal, and a server on a mobile communication network. The wireless signal may include various types of data according to transmission/reception of a voice call signal, a video call signal, or a text/multimedia message.

The wireless Internet module refers to a module for wireless Internet access, and may be built-in or external to the learning-based symptom and disease management apparatus 100 . The wireless communication is WLAN (Wireless LAN) (Wi-Fi), Wibro (Wireless broadband), Wimax (World Interoperability for Microwave Access), HSDPA (High Speed Downlink Packet Access), LTE (Long Term Evolution), LTE-A ( Advanced) and the like may be used, and may be built based on 5G communication systems as well as existing 2G, 3G, and 4G communication systems.

The user's information can be shared through the communication unit 60 to medical staff who have agreed to disclose the information to the user. When a communication channel exists and the medical staff determines that it is necessary, the user's symptoms and disease-related factors can be changed through communication, and the user can be recommended to visit the hospital and receive medical treatment. If the suspected disease does not match the existing list of suspected diseases after medical treatment, the medical staff can change the list of suspected diseases to be input into the user's terminal. A description of the suspected disease and treatment, current health status and level, relevant factors such as demographic/disease/lifestyle calculated to be highly correlated with the current disease or condition, short-term and long-term goals, and long-term goals to achieve and short-term plans, actions related to daily life, and the like that the user can practice in daily life, etc. may be changed. Such data may be used as learning data for modeling a user or disease.

By using the user's health information, location information, etc., a nearby hospital may be automatically disclosed, and a clinic/hospital advertisement may be suggested.

Using the user's health information and location information, products and services may be recommended, related advertisements may be suggested, and the data may be used for business strategy, sales strategy, production strategy, and the like.

Meanwhile, FIG. 2 is a flowchart illustrating an embodiment of a learning-based symptom and disease management method that can be implemented according to the present invention. Although the flowchart of FIG. 2 is configured for the first input, it is also possible that the same method as in FIG. 2 is applied to an arbitrary second input input after the first input.

Referring to FIG. 2 , a first input composed of a plurality of state values related to a user's state is received ( S10 ). Each input consists of a plurality of state values, and the plurality of state values include information about the user's daily life behavior, information about the user's posture, information about the user's operation, information about the environment around the user, information about the user's It may include biometric information and musculoskeletal health information. Such an input may be generated by a user's input, by an input other than the user, or by sensing a sensor.

Specifically, the state value refers to a state in which all information obtainable from the user or the environment is mapped in multiple dimensions. For example, if a user or environment has a certain element, it is expressed as a binary representation of 1 or 0 otherwise, so that several n elements can be expressed in n dimensions. In addition, when there are many such n, dimension transformation or reduction can be performed through autoencoder, principle component analysis, etc. In addition, as another example, if there is a continuous variable (eg, systolic blood pressure), the state value can be expressed discretely by taking a range of continuous values and converting it into a multi-class. Also, as another example, it is possible to define a dimension on the state value for an element by scaling the continuous value (eg, 10 times, 0.5 times, etc.) and then taking an integer value. Also, as another example, it may have a value of multiclass. Since the state value contains all of these elements, it can be multidimensional.

It is also possible to set when to receive information (health information, biometric information, etc.) for deriving the next state from the user for each action a. A user who has been recommended a specific action a based on the current time point t may receive a request for input next time by using the time period information recorded in the specific action a. Also, even before an input request, the user may directly input information about his or her state.

Next, a first set of action information is selected from a plurality of action information stored in the database 20 according to the first input (S20).

For a specific state, it is defined as a set A of actions that can be recommended, which can be expressed as A(s). For example, for a state of _{s 1 , if actions a 1} , a ₂ , a ₃ are possible, A(s ₁ ) = {a ₁ , a ₂ , a ₃ }. Depending on the state, the number or type of action elements that A(s) has may vary. It is also possible to limit or adjust the number or type of elements of a set of actions with respect to a specific state s according to clinical knowledge.

These actions may include, in addition to recommending a specific action to the user, progress observation or follow-up.

Next, a plurality of evaluation values corresponding to each of the first set of action information selected in step S20 are calculated using a predetermined model with respect to the first input (S30).

When a state value is defined, a reward and a value function can be defined. The reward means a score that can be obtained internally in the model when the user reaches a specific state. A positive value indicates a direction that is beneficial to the user, and a negative value indicates a detrimental direction for the user. For example, when a state is defined in three dimensions as (weight, blood pressure, blood sugar level), it may be defined as compensation (normal weight, norm blood pressure, norm blood sugar level) = +100.

를 정의할 수 있다. 그리고 이 경우 상태의 가치를 표현할 수 있으며, 이는 가치 함수 V(s) = E(G_t|S_t=s) 로 표현된다. 이는 상태값 s에 대한 가치 함수 V(s)를 표현하는 식이며, E는 expectation(기대값)을 나타낸다. 기대값이 붙은 이유는, 특정 시점 t 이후에 생길 수 있는 모든 시나리오에 대하여 확률적으로 반영해야 하기 때문이다. 이 보상은 임상 의사의 지식에 따라 다르게 부여되도록 조정하는 것도 가능하다.However, although the present state is not a completely normal state, there is a possibility that it will go to a normal state in the future. In this case, a discounting factor γ is introduced. Here, γ has a value of 0 or more and 1 or less. At this time, the return, which is the total reward that can be received at a specific point in time t

can be defined. And in this case, the value of the state can be expressed, which is expressed by the value function V(s) = E(G _t |S _t =s). This is an expression expressing the value function V(s) for the state value s, and E represents the expectation. The reason that the expected value is attached is that it should be reflected probabilistically for all scenarios that may occur after a specific time point t. It is also possible to adjust this reward to be given differently according to the knowledge of the clinician.

In this case, the reward when action a is performed in the state s may be subdivided and defined. If this _{is defined as R s} ^a , it is a reward for recommending action a in state s. If the reward in state s is uniformly defined regardless of action a, then R _s ^a = R _s (invariant to action).

Furthermore, in order to calculate the evaluation value, a policy should be defined. The policy has information about the probability of taking action a in a particular state. That is, it means which action a will be performed with what probability in a specific state s.

The above expression represents the probability of performing action a in state s. In the present invention, it can be understood as the probability of recommending the action a to the user in the state s. The goal of reinforcement learning is to find a policy π that maximizes return G for all states.

For example, suppose A(s ₂ ) = {a ₁ , a ₂ , a ₃ , a _{4 } for state s 2} , and if π ₁ (a ₁ |s ₂ ) = 0, π ₁ (a ₂ If there is a _{policy π 1} with |s ₂ ) = 1, π ₁ (a ₃ |s ₂ ) = 0, π ₁ (a ₄ |s ₂ ) = 0, then only a ₂ will be _{unconditionally selected in state s 2 .} However, there is another policy π _{2 ,} where π ₂ (a ₁ |s ₂ ) = 0.3, π ₂ (a ₂ |s ₂ ) = 0, π ₂ (a ₃ |s ₂ ) = 0, π ₂ (a ₄ If |s ₂ ) = 0.7, _{under this policy π 2} , a ₁ will be selected with a 30% _{probability and a 4} will be selected with a 70% probability in the state s _{2 .}

Based on this policy, a value function, an action-value function (Q) can be defined. The action changes according to the policy selected in the state, and the policy π is reflected because the state is changed again in the future. That is, V _π (s) = E _π (G _t |S _t =s). Also, for an action, Q _π (s, a) = E _π (G _t |S _t =s, A _t =a) is given, which when taking action a in state s, given policy π, It represents the return, which is the sum of the rewards obtained later.

The model can continuously update policies, value functions, and action-value functions. The update cycle is when the user inputs information required by the algorithm, and includes his/her health information, biometric information, and environmental information. That is, when the user inputs information at a certain point in time t, S _t is derived and A _t is recommended. Next time, when the user inputs information again, this time S _t+1 is derived.

In practice, the action-value function and the value function have the following relationship:

For a given state s, action a, we can consider a situation that maximizes V and Q. This is expressed as follows.

Based on the above Q _opt , the optimal policy can be obtained as follows.

or

Here, ε has a value of 0 or more and 1 or less. If ε is 0, it means that an optimal action is recommended based on what has been learned so far without additional search for action recommendations.

The step S30 may include applying a reinforcement learning algorithm. In the step of applying the reinforcement learning algorithm, in order to alleviate the user's condition or musculoskeletal symptoms, the method of calculating the evaluation value may be adjusted to suit the user's condition. For example, according to the reinforcement learning algorithm, when a plurality of evaluation values for the second input are calculated, a plurality of state values of the first input stored in the database 20 and first recommended action information for the first input can be configured to be the basis.

The step S30 may include analyzing the user's musculoskeletal symptoms using the input. The analyzing may include searching for factors that aggravate musculoskeletal symptoms or factors that improve musculoskeletal symptoms in a plurality of state values.

Next, the first recommended action information, which is at least one of the first set of action information, is selected based on the plurality of evaluation values calculated in step S30 ( S40 ). For example, the action information may be selected in such a way that a higher weight is given to the action information in which a higher reward is expected among the first set of action information. Also, for example, when an evaluation value greater than a threshold value is identified among a plurality of evaluation values or an evaluation value equal to or greater than a threshold value is identified among a plurality of evaluation values, the step S40 may be configured to select action information corresponding to the thus identified evaluation value. can Also, as another example, action information may be probabilistically provided by varying or comparing the size of the evaluation value itself.

In order to recommend an action to a user, various methods are possible. With respect to the action set A(s) assigned to a specific state, the weight of each element action may be the same, or an action extraction probability based on the policy may be given based on the previously presented policy. In addition, there is a method of deriving an optimal recommended action policy based on Q and policy learned so far, but adjusting ε.

For model update, we aim to derive a policy by updating Q(s,a). For this method, Q-learning, Markov Decision Process, Dynamic Programming, Temporal-Difference Learning, Deep Q-Learning, etc. may be used. If the learning rate is α (it has a value between 0 and 1), then the value of Q can be updated based on the equation below.

Meanwhile, in the case of a general Q function, it is based on updating Q and policy by reflecting input from all users (the Q and policy used here are referred to _{as Q population} and π _population). In this case, input information of all users is reflected in one Q function.

Alternatively, a user-specific Q and policy may be used. These are called Q _users and π _users . For example, for user C, there are Q _C and π _C . In _{the case of Q C} , since the model learning itself of reinforcement learning in the first place updates the model according to the change value of the state different from the action, even if the user C receives an action recommendation through _{Q population} and π _{population instead of Q C} and π _C , the model It is possible to update However, in _{the case of the update of Q C} , only the input corresponding to a specific user is taken, and in _{the case of Q C} , the input of users A or B is not involved in the update.

When a certain amount of input for a specific user is accumulated, the user or clinician can receive an action recommendation through Q _{user instead of Q population . In this case, π user} is used instead of π _population. (Of course, the opposite is also possible.) A numerical value corresponding to a certain amount can be set as a threshold value. When Q _user is used, the user uses a specific model for his/her state, actions, and changes in his/her state according to actions, instead of a model derived from the entire user group. In this platform, when a new user is added, the Q _population is used, and the Q _user for the new user is first set to be the same as the _{Q population, and then updated as well.}

Next, the action information selected in step S40 is provided to the user (S50). In addition, the user may provide information on relevant factors to the medical institution to which the user has consented to disclose the information, which makes it possible to link with actual treatment in the clinical setting.

Meanwhile, the present invention can also be implemented as computer-readable codes on a computer-readable recording medium. The computer-readable recording medium includes all types of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device. In addition, the computer-readable recording medium is distributed in a computer system connected through a network, so that the computer-readable code can be stored and executed in a distributed manner. In addition, functional programs, codes, and code segments for implementing the present invention can be easily inferred by programmers in the technical field to which the present invention pertains.

In addition, in the method and apparatus using the same as described above, the configuration and method of the above-described embodiments are not limitedly applicable, but all or part of each embodiment is selective so that various modifications can be made to the embodiments. It may be configured in combination with .

Claims (45)

an input unit for receiving a first input composed of a plurality of state values related to a user's state;
a database in which a plurality of action information is stored;
A first set of action information is selected from the plurality of action information according to the input first input, and a plurality of pieces of action information corresponding to each of the first set of action information are selected by using a predetermined model for the first input. a processing unit that calculates an evaluation value and selects first recommended action information that is at least one of the first set of action information based on the plurality of evaluation values calculated; and
and an output unit for providing the selected first recommended action information to the user,
Selecting the first set of action information is limited at least in part by clinical information input through the input unit or clinical information stored in the database,
When it is determined that at least a portion of the selected first set of action information does not match the clinical information, the first set of action information is reselected based on the clinical information, a learning-based symptom and disease management apparatus. an input unit for receiving a first input composed of a plurality of state values related to a user's state;
a database in which a plurality of action information is stored;
A first set of action information is selected from the plurality of action information according to the input first input, and a plurality of pieces of action information corresponding to each of the first set of action information are selected by using a predetermined model for the first input. a processing unit that calculates an evaluation value and selects first recommended action information that is at least one of the first set of action information based on the plurality of evaluation values calculated; and
and an output unit for providing the selected first recommended action information to the user,
The processing unit calculates a plurality of rewards indicating an effect of applying each of the first set of action information on the state of the user, and the calculated plurality of rewards are reflected in calculating the plurality of evaluation values becomes,
The plurality of rewards are at least partially limited by clinical information input through the input unit or clinical information stored in the database, a learning-based symptom and disease management apparatus. 3. The method according to claim 1 or 2,
and the first input received in the input unit and the first recommended action information selected for the first input by the processing unit are configured to be stored in the database. 4. The method of claim 3,
The processing unit includes a learning module that applies a reinforcement learning algorithm to update the model based on the update information,
The update information includes a state value of the first input stored in the database and information on the first recommended action selected with respect to the first input, a learning-based symptom and disease management apparatus. 5. The method of claim 4,
The update information further includes a state value of the second input received in the input unit, a learning-based symptom and disease management apparatus. 5. The method of claim 4,
The learning-based symptom and disease management apparatus, wherein the processing unit is configured to calculate a plurality of evaluation values for a second input by using the model updated by the learning module. 7. The method of claim 6,
The plurality of evaluation values for the second input are used to select second recommended action information for the second input, a learning-based symptom and disease management apparatus. 7. The method of claim 6,
wherein the first input is received at a first time, and the second input is received at a second time after the first time. 7. The method of claim 6,
The update of the model controls a method in which the plurality of evaluation values are calculated with respect to the second input in order to induce the user's condition to be improved, a learning-based symptom and disease management apparatus. 3. The method of claim 2,
The processing unit, a learning-based symptom and disease management apparatus for assigning a higher weight to the action information that is expected to receive a higher reward among the first set of action information. 3. The method of claim 2,
The processing unit identifies an evaluation value greater than a threshold value among the plurality of calculated evaluation values, and selects action information corresponding to the identified evaluation value as the first recommended action information, a learning-based symptom and disease management apparatus. 3. The method of claim 2,
The learning-based symptom and disease management apparatus, wherein the processing unit calculates the reward based on the clinical information when it is determined that at least some of the plurality of rewards do not match the clinical information. delete 3. The method according to claim 1 or 2,
The processing unit analyzes the musculoskeletal symptoms of the user using the first input, and the analyzed musculoskeletal symptoms of the user are used to calculate the plurality of evaluation values for the first input, a learning-based Symptom and disease management devices. 15. The method of claim 14,
Analyzing the musculoskeletal symptoms of the user includes searching for factors exacerbating the musculoskeletal symptoms or factors improving the musculoskeletal symptoms from the plurality of state values of the first input, learning-based symptoms and diseases management device. 16. The method of claim 15,
Analyzing the musculoskeletal symptoms of the user is based on the searched factors for exacerbating the musculoskeletal symptoms or improving the musculoskeletal symptoms. Diagnosis, progression or prognosis of the musculoskeletal symptoms, including predicting the prognosis, learning based of symptom and disease management devices. 16. The method of claim 15,
The output unit is configured to provide a user with a factor that aggravates the musculoskeletal symptoms or a factor that improves the musculoskeletal symptoms, a learning-based symptom and disease management device. 3. The method according to claim 1 or 2,
The plurality of state values may include information about the user's daily life behavior, information about the user's posture, information about the user's operation, information about the environment around the user, biometric information of the user, and information about the user's posture. A learning-based symptom and disease management device comprising at least one of musculoskeletal health information. 19. The method of claim 18,
The information on the user's daily life behavior includes the position, length, and angle of the user's body part in the daily life behavior, and their change over time, and the type, time, intensity, method, and their activity type of the user. comprising at least one of the patterns;
The information about the user's posture includes at least one of a position, a length, an angle of the user's body part in a static state, and their change with time,
The information on the user's motion includes at least one of a position, a length, an angle, and a change over time of the body part of the user in a dynamic state,
The information about the environment around the user includes at least one of the location, size, height, distance, intensity, and patterns of any object within a preset radius from the user,
The user's biometric information includes at least one of age, sex, family history, past history, current treatment, past treatment, skeletal muscle mass, weight, height, and genetic information of the user,
The user's musculoskeletal health information includes at least one of symptoms of pain, a location of pain, a frequency of pain, a duration of pain, an aspect of pain, and a pattern thereof, a learning-based symptom and disease management apparatus. 19. The method of claim 18,
The first input is a learning-based symptom and disease management apparatus that is generated by the user's input, by an input other than the user, or by sensing a sensor. receiving, by the input unit, a first input composed of a plurality of state values related to a user's state;
selecting, in the processing unit, a first set of action information from a plurality of action information stored in a database according to the first input;
calculating, in the processing unit, a plurality of evaluation values corresponding to each of the first set of action information by using a predetermined model with respect to the first input;
selecting, by the processing unit, first recommended action information that is at least one of the first set of action information based on the plurality of calculated evaluation values; and
and providing, by an output unit, the selected first recommended action information to the user,
In the selecting step, selecting the first set of action information is limited at least in part by clinical information input through the input unit or clinical information stored in the database,
When it is determined that at least a portion of the first set of action information selected in the selecting step does not match the clinical information, the first set of action information is reselected based on the clinical information, learning-based symptom and methods of disease management. receiving, by the input unit, a first input composed of a plurality of state values related to a user's state;
selecting, in the processing unit, a first set of action information from a plurality of action information stored in a database according to the first input;
calculating, in the processing unit, a plurality of evaluation values corresponding to each of the first set of action information by using a predetermined model with respect to the first input;
selecting, by the processing unit, first recommended action information that is at least one of the first set of action information based on the plurality of calculated evaluation values; and
and providing, by an output unit, the selected first recommended action information to the user,
The calculating includes calculating, in the processing unit, a plurality of rewards indicating an effect of applying each of the first set of action information on the state of the user, and The plurality of rewards are reflected in calculating the plurality of evaluation values,
The plurality of rewards are at least partially limited by clinical information input through the input unit or clinical information stored in the database, a learning-based symptom and disease management method. 23. The method of claim 21 or 22,
The received first input and the first recommended action information selected for the first input are configured to be stored in the database, a learning-based symptom and disease management method. 24. The method of claim 23,
The calculating includes applying, by a learning module, a reinforcement learning algorithm, such that the model is updated based on the update information,
The update information includes a state value of the first input stored in the database and information on the first recommended action selected with respect to the first input, a learning-based symptom and disease management method. 25. The method of claim 24,
The update information further includes a state value of the second input received in the input unit, a learning-based symptom and disease management method. 25. The method of claim 24,
A learning-based symptom and disease management method, configured to calculate a plurality of evaluation values for a second input using the model updated by the step of applying the reinforcement learning algorithm. 27. The method of claim 26,
The plurality of evaluation values for the second input are used to select second recommended action information for the second input, a learning-based symptom and disease management method. 27. The method of claim 26,
wherein the first input is received at a first time and the second input is received at a second time after the first time. 27. The method of claim 26,
The update of the model controls a method in which the plurality of evaluation values are calculated with respect to the second input in order to induce improvement of the user's condition, a learning-based symptom and disease management method. 23. The method of claim 22,
In the calculating step, a higher weight is given to the action information for which a higher reward is expected among the first set of action information, a learning-based symptom and disease management method. 23. The method of claim 22,
The calculating includes, in the processing unit, identifying an evaluation value greater than a threshold value among the plurality of calculated evaluation values,
The selecting may include selecting, in the processing unit, action information corresponding to the evaluation value identified in the identifying as the first recommended action information, a learning-based symptom and disease management method. 23. The method of claim 22,
In the calculating step, when it is determined that at least some of the plurality of rewards do not match the clinical information, the learning-based symptom and disease management method is configured to calculate the reward based on the clinical information. delete 23. The method of claim 21 or 22,
The calculating includes, in the processing unit, analyzing the user's musculoskeletal symptoms using the first input,
The musculoskeletal symptoms of the user analyzed in the analyzing step are used to calculate the plurality of evaluation values for the first input, a learning-based symptom and disease management method. 35. The method of claim 34,
The step of analyzing the musculoskeletal symptoms of the user includes, in the processing unit, searching for factors that aggravate the musculoskeletal symptoms or factors that improve the musculoskeletal symptoms from the plurality of state values of the first input. Learning-based management of symptoms and conditions. 36. The method of claim 35,
The step of analyzing the musculoskeletal symptoms of the user may include, in the processing unit, the diagnosis of the musculoskeletal symptoms based on the factors exacerbating the musculoskeletal symptoms or the factors that improve the musculoskeletal symptoms found in the searching step. A learning-based symptom and disease management method, further comprising predicting a prognosis. 36. The method of claim 35,
The step of providing, by the output unit, a factor that aggravates the musculoskeletal system symptom or a factor that improves the musculoskeletal system symptom to the user, a learning-based symptom and disease management method. 23. The method of claim 21 or 22,
The plurality of state values may include information about the user's daily life behavior, information about the user's posture, information about the user's operation, information about the environment around the user, biometric information of the user, and information about the user's posture. A learning-based symptom and disease management method comprising at least one of musculoskeletal health information. 39. The method of claim 38,
The information on the user's daily life behavior includes the position, length, and angle of the user's body part in the daily life behavior, and their change over time, and the type, time, intensity, method, and their activity type of the user. comprising at least one of the patterns;
The information about the user's posture includes at least one of a position, a length, an angle of the user's body part in a static state, and their change with time,
The information on the user's motion includes at least one of a position, a length, an angle, and a change over time of the body part of the user in a dynamic state,
The information about the environment around the user includes at least one of the location, size, height, distance, intensity, and patterns of any object within a preset radius from the user,
The user's biometric information includes at least one of age, sex, family history, past history, current treatment, past treatment, skeletal muscle mass, weight, height, and genetic information of the user,
The user's musculoskeletal health information includes at least one of symptoms of pain, a location of pain, a frequency of pain, a duration of pain, an aspect of pain, and a pattern thereof, a learning-based symptom and disease management method. 39. The method of claim 38,
Wherein the first input is generated by the user's input, by an input other than the user, or by sensing a sensor, a learning-based symptom and disease management method. A non-transitory computer-readable storage medium comprising instructions that, when executed, cause one or more processors to:
receive a first input comprising a plurality of state values related to the user's state;
select a first set of action information from a plurality of action information stored in a database according to the inputted first input;
calculate a plurality of evaluation values corresponding to each of the first set of action information by using a predetermined model with respect to the first input;
select first recommended action information that is at least one of the first set of action information based on the calculated plurality of evaluation values;
to provide the selected first recommended action information to the user, and
Selecting the first set of action information is limited at least in part by clinical information input through an input unit or clinical information stored in the database,
When it is determined that at least a portion of the selected first set of action information does not match the clinical information, the first set of action information is reselected based on the clinical information. A non-transitory computer-readable storage medium comprising instructions that, when executed, cause one or more processors to:
receive a first input comprising a plurality of state values related to the user's state;
select a first set of action information from a plurality of action information stored in a database according to the inputted first input;
calculate a plurality of evaluation values corresponding to each of the first set of action information by using a predetermined model with respect to the first input;
select first recommended action information that is at least one of the first set of action information based on the calculated plurality of evaluation values; and
to provide the selected first recommended action information to the user,
wherein the calculating comprises calculating a plurality of rewards representing an effect of applying each of the first set of action information on the state of the user, wherein the calculated plurality of rewards comprises the plurality of evaluations. is reflected in calculating the value,
wherein the plurality of rewards are limited at least in part by clinical information input through an input unit or clinical information stored in the database. delete delete delete

Images