KR102509534B1 - Method of learning a deep learning model to diagnose health and apparatuses performing the same - Google Patents

Abstract

The present disclosure relates to a method for learning a deep learning model, and the method for learning a deep learning model according to an embodiment of the present disclosure includes acquiring training data of the deep learning model, and an object corresponding to the training data. Generating correction data in which the learning data is filtered by filtering the learning data based on at least one of first information about the electronic device and second information about the movement of the object; filtering the correction data based on third information for , generating re-correction data in which the correction data is filtered; and simultaneously inputting the training data and the re-correction data into the deep learning model to generate the deep learning model. It includes the step of learning.

Description

A method for learning a deep learning model for diagnosing health and an apparatus for performing the same

The present disclosure relates to a method for learning a deep learning model for diagnosing health and an apparatus for performing the same.

According to the World Health Organization, cardiovascular disease is the leading cause of death today. More than 17.7 million people died from cardiovascular disease, accounting for about 31% of all deaths, more than 75% of which occurred in low- and middle-income countries.

Arrhythmia, a representative type of cardiovascular disease, is a disease that shows irregular changes in the normal heart rhythm. Such arrhythmias include atrial fibrillation, premature contraction, ventricular fibrillation, and tachycardia. This is included. Although a single arrhythmia may not have a life-threatening effect, continuous arrhythmia can lead to a fatal situation, so it is important to monitor the heart rate continuously as a management and prevention measure.

Many tests are used to diagnose arrhythmia and coronary artery disease (cardiac artery disease), but among them, an electrocardiogram (ECG) test has many advantages and is the most used test in clinical practice.

ECG analysis is a well-established method for studying a patient's cardiac function and identifying disorders of the heart. Doctors have been using ECG systems to monitor heart activity in Chinese characters for decades. An electrocardiogram system is a useful device that can conveniently diagnose a patient's heart condition, and the system monitors the patient's heart's electrical activity.

Currently, there are several systems that monitor the electrical activity of a patient's heart and analyze the ECG signal to identify the type of heart disease the patient is suffering from. However, these systems are generally not suitable for portable use because they must be fixed.

The present disclosure has been derived in response to the above background art, and one of the various objects of the present disclosure is to provide a method for learning a deep learning model for diagnosing health and an apparatus for performing the same.

However, the technical challenges are not limited to the above-described technical challenges, and other technical challenges may exist.

A method for learning a deep learning model according to an embodiment of the present disclosure for achieving the above technical problem includes acquiring learning data of the deep learning model, and providing an electronic device of an object corresponding to the learning data. filtering the learning data based on at least one of information 1 and second information about the motion of the object to generate correction data obtained by filtering the learning data; filtering the correction data based on the filtering method to generate recalibration data in which the correction data is filtered, and learning the deep learning model by simultaneously inputting the training data and the recalibration data to the deep learning model. do.

In addition, a program for implementing the method as described above may be recorded in a computer-readable recording medium.

A method for learning a deep learning model according to an embodiment of the present disclosure may provide a technique of generating a general-purpose deep learning model and analyzing various ECG data measured through various ECG induction methods.

Effects obtainable in the present disclosure are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below. .

1 is a diagram for explaining a massage device according to an embodiment of the present disclosure.
2 is a diagram for explaining a main frame according to an embodiment of the present disclosure.
3 is a diagram for explaining components of a massage device according to an embodiment of the present disclosure.
4 is a diagram for explaining an external electronic device capable of communicating with a massage device according to an embodiment of the present disclosure.
5A shows an example for explaining an electrocardiogram induction method according to an embodiment of the present disclosure.
5B shows another example for explaining an electrocardiogram induction method according to an embodiment of the present disclosure.
5C shows another example for explaining an electrocardiogram induction method according to an embodiment of the present disclosure.
6 illustrates an example for describing electrocardiogram data according to an embodiment of the present disclosure.
7 shows an example for explaining a health diagnosis system according to an embodiment of the present disclosure.
8A illustrates an example for describing a data measurement device according to an embodiment of the present disclosure.
8B illustrates another example for describing a data measurement device according to an embodiment of the present disclosure.
9 illustrates an example for explaining a health diagnosis device according to an embodiment of the present disclosure.
10 is a flowchart for explaining a deep learning model learning operation of a processor according to an embodiment of the present disclosure.
11 illustrates an example for describing an electrical signal included in electrocardiogram data according to an embodiment of the present disclosure.
12 illustrates another example for explaining a deep learning model learning operation of a processor according to an embodiment of the present disclosure.
13A illustrates an example for explaining an operation of determining learning data by a processor according to an embodiment of the present disclosure.
13B illustrates another example for explaining an operation of determining learning data by a processor according to an embodiment of the present disclosure.
13C shows another example for explaining an operation of determining learning data by a processor according to an embodiment of the present disclosure.
14 shows an example for explaining the learning data determination operation shown in FIG. 13B.
FIG. 15 shows an example for explaining the operation of determining candidate learning data and the operation of determining learning data shown in FIG. 13C.
FIG. 16 shows an example for explaining a filtering operation and a learning operation of the processor shown in FIG. 12 .
17A illustrates an example for describing original electrocardiogram data according to an embodiment of the present disclosure.
17B illustrates an example for describing correction data according to an embodiment of the present disclosure.
18A shows another example for explaining correction data according to an embodiment of the present disclosure.
18B shows another example for explaining correction data according to an embodiment of the present disclosure.
19 illustrates an example for describing a health state diagnosis operation of a processor according to an embodiment of the present disclosure.

Objects, features and advantages of the present disclosure described above will become more apparent through the following examples in conjunction with the accompanying drawings. The following specific structural or functional descriptions are merely illustrated for the purpose of explaining embodiments according to the concept of the present disclosure, and embodiments according to the concept of the present disclosure may be implemented in various forms and are described in this specification or application. It should not be construed as being limited to the examples.

Embodiments according to the concept of the present disclosure can be applied with various changes and can have various forms, so specific embodiments are illustrated in the drawings and described in detail in this specification or application. However, this is not intended to limit the embodiments according to the concept of the present disclosure to a specific disclosure form, and should be understood to include all changes, equivalents, or substitutes included in the spirit and technical scope of the disclosure.

Terms such as first and/or second may be used to describe various components, but the components are not limited to the above terms. The above terms are used only for the purpose of distinguishing one component from other components, for example, without departing from the scope of rights according to the concept of the present disclosure, a first component may be named a second component, and similar Happily, the second component may also be referred to as the first component.

It is understood that when a component is referred to as being “connected” or “connected” to another component, it may be directly connected or connected to the other component, but other components may exist in the middle. It should be. On the other hand, when a component is referred to as “directly connected” or “directly connected” to another component, it should be understood that no other component exists in the middle. Other expressions used to describe the relationship between components, such as "between" and "directly between" or "adjacent to" and "directly adjacent to", should be interpreted similarly.

Terms used in this specification are only used to describe specific embodiments, and are not intended to limit the present disclosure. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as "comprise" or "have" are intended to indicate that the described feature, number, step, operation, component, part, or combination thereof exists, but one or more other features or numbers, It should be understood that the presence or addition of steps, operations, components, parts, or combinations thereof is not precluded.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in this specification, it should not be interpreted in an ideal or excessively formal meaning. don't

In this specification, an actuator means a component capable of providing a driving force. For example, the actuator may include a motor, a linear motor, an electric motor, a DC motor, an AC motor, a linear actuator, an electric actuator, and the like, but is not limited thereto.

In this specification, binaural bits may mean a specific type of audio information capable of controlling brain waves.

In the present specification, a massage device may refer to a massage device including a body massage unit and a leg massage unit. In addition, in the present specification, the body massage unit and the leg massage unit may exist as separate devices (eg, a body massage device and a leg massage device), and the massage device may refer to a body massage device or a leg massage device. .

In the present specification, a target object may be a diagnosis target for diagnosing heart disease. In the present specification, a plurality of electrocardiogram data for a plurality of objects may be electrocardiogram data measured from a plurality of objects through a plurality of electrocardiogram induction methods.

In the present specification, a plurality of electrocardiogram induction methods may be a method of measuring electrocardiogram data of an object using electrodes.

In this specification, learning data may be electrocardiogram data for learning a deep learning model. In this specification, candidate learning data may be electrocardiogram data that may be determined as learning data.

In this specification, a sub data unit is a reference unit for extracting learning data and may be a data unit corresponding to a single training data. For example, 12 electrocardiogram data (Lead I electrocardiogram data, Lead II electrocardiogram data, Lead III electrocardiogram data) measured at the same time through the standard limb lead method, amplified limb lead method, and chest lead method (or 12 lead method) , aVR ECG data, aVL ECG data, aVF ECG data, V1 ECG data, V2 ECG data, V3 ECG data, V4 ECG data, V5 ECG data, and V6 ECG data) are all determined as training data of the deep learning model, sub The data unit is a criterion for extracting each of the 12 ECG data and may be the same data unit as the data unit of each ECG data.

In this specification, sub-learning data may be learning data in units of sub-data. For example, the learning data of the sub-data unit may be any one of a plurality of learning data and may be a single learning data.

In this specification, the learning data of the sub-data unit is electrocardiogram data measured from any one object among a plurality of objects, and may be original electrocardiogram data including various noises. In the present specification, the correction data may be ECG data obtained by filtering a frequency corresponding to noise among a plurality of frequencies included in the original ECG data. In the present specification, the recalibration data may be electrocardiogram data obtained by filtering a specific frequency from among a plurality of frequencies included in the correction data.

In the present specification, the representative ECG induction method may be an ECG induction method representing each object among ECG induction methods corresponding to each object. In the present specification, representative electrocardiogram data may be electrocardiogram data corresponding to a representative electrocardiogram induction method representing each object among electrocardiogram data corresponding to each object.

In this specification, the total number of ECG data may be the number of a plurality of ECG data for a plurality of objects. In the present specification, the number of ECG data may be the number of ECG data to be selected as learning data from among a plurality of ECG data. In this specification, the total number of candidate training data may be the number of candidate training data. In this specification, the number of candidate training data may be the number of candidate training data to be selected as training data from among candidate training data.

In this specification, the first electromagnetic signal may be an electromagnetic signal generated by an electronic device of an object while electrocardiogram data is measured. In this specification, the second electromagnetic signal may be an electromagnetic signal generated by muscle movement of an object while electrocardiogram data is measured. In the present specification, the third electromagnetic signal may be an electromagnetic signal generated by respiration of the object while electrocardiogram data is measured.

Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

1 is a diagram for explaining a massage device according to an embodiment of the present disclosure.

The massage device 100 forms an area for accommodating at least a portion of the user's body, and includes a body massage unit 1000 that massages the user's upper and lower body parts, and an arm that accommodates the user's arm and massages the user's arm. It may include a massage unit 2000 and a leg massage unit 3000 that accommodates the user's legs and massages the user's legs.

The body massage unit 1000 may form any type of space for accommodating a user. The body massage unit 1000 may have a space corresponding to the shape of the user's body. For example, as shown in FIG. 1 , the body massage unit 1000 may be implemented as a seating type capable of accommodating the user's entire body or a part of the body.

The part in contact with the ground in the body massage unit 1000 may include any material for increasing frictional force or any member for increasing frictional force (eg, non-slip pad, etc.), and the massage device 100 It may include wheels to enhance the mobility of.

At least a portion of the body massage unit 1000 may slide. For example, when the body massaging unit 1000 starts massaging, at least a portion of the body massaging unit 1000 may slide forward. Also, the body massage unit 1000 may be inclined backward. As a result, the body massage unit 1000 may provide a massage in a backward tilted state.

The body massage unit 1000 includes a head contact unit that can contact the user's head and support the head, a back contact unit (or backrest) that can contact and support the user's back, and a user's back. The main frame 1100 constituting the skeleton (or internal skeleton) of the buttock contact part (or seat part), the body massage part 1000, which is in contact with the hip part of the body and can support the hip part, and any form of An audio unit (or audio output module) 1600 for providing audio output, a massage module 1700 for providing massage to the user's upper and lower body parts, and a user input unit for receiving any type of input from the user ( 1800) may be included. Components included in the aforementioned body massaging unit 1000 are merely exemplary embodiments, and the body massaging unit 1000 may include various components other than the aforementioned components.

The massage module 1700 may be provided inside the body massage unit 1000 to provide any type of mechanical stimulation to the user accommodated in the body massage unit 1000 . As shown in FIG. 1 , the massage module 1700 may move up and down along the main frame 1100 provided inside the body massage unit 1000 .

For example, a rack gear may be provided in the main frame 1100 of the body massage unit 1000, and the massage module 1700 moves along the rack gear while moving the user's body (eg, Mechanical stimulation may be provided to various parts of the user's upper body and lower body. The massage module 1700 may include a ball massage unit or a roller massage unit, but is not limited thereto.

The main frame 1100 constitutes the skeleton of the internal structure of the body massage unit 1000 (or the internal frame of the body massage unit 100), and may be implemented with a metal material or a plastic material. For example, the main frame 1100 may be implemented with iron, alloy, steel, etc., but is not limited thereto and may be implemented with various hard materials.

The audio unit 1600 may be provided in various locations. For example, the audio unit 1600 may be disposed above the body massage unit 1000 .

The audio unit 1600 includes an upper audio output unit disposed on the top of the seat portion that contacts the user, a front audio output unit attached to the front end of the arm massage unit on the left and right sides of the seat unit, and/or a rear audio output unit attached to the rear end of the arm massage unit. It may include a plurality of output units, such as, but is not limited thereto. In this case, the audio unit 1600 may provide stereo sound such as 5.1 channels, but is not limited thereto.

The audio unit 1600 may provide an arbitrary type of audio output to the user. For example, the audio unit 1600 provides brain stimulation to the user by outputting to the user a comment (or voice), sound source, and/or binaural beat optimized for the massage pattern provided by the massage device 100. can

The audio unit 1600 may output a sound signal received through a network (not shown) or stored in an internal/external storage medium (not shown). For example, the audio unit 1600 may output a sound source under control of the massage apparatus control device 200 through a network connection (eg, Bluetooth connection) with the massage apparatus control device 200 . In addition, the audio unit 1600 may output any type of sound signal generated in relation to the operation of the massage device 100 .

The user input unit 1800 may receive commands related to operation control of the massage device 100 from a user, and the user input unit 1800 may be implemented in various forms. For example, the user input unit 1800 may be provided in the body massaging unit 1000 and may be provided in the leg massaging unit 3000, but is not limited thereto.

The massage device 100 may acquire various commands from the user through the user input unit 1800 . For example, the massage device 100 includes selection of a massage module, selection of a massage type (or massage mode), selection of a massage intensity, selection of a massage time, selection of a massage part, position and operation of the body massage unit 1000. It is possible to receive arbitrary commands for selection, selection for On-Off of the power of the massage device 100, selection for whether to operate the cooling and heating function, selection related to sound source reproduction, etc., but is not limited thereto. .

The user input unit 1800 may include buttons in the form of hot keys and/or selection buttons for selecting directions, canceling, and executing inputs according to preset user-set functions or self-preset functions. there is. For example, the user input unit 1800 may be implemented as a key pad, a dome switch, a touch pad (static pressure/capacity), a jog wheel, a jog switch, and the like, but is not limited thereto. In addition, the user input unit 1800 may obtain a command through a user's speech based on voice recognition technology.

The user input unit 1800 may include a display for displaying an operating state of the massage device 100 or a user's current state. In this case, the display includes a liquid crystal display (LCD), a thin film transistor-liquid crystal display (TFT LCD), an organic light-emitting diode (OLED), a flexible display ), and at least one of a 3D display, but is not limited thereto.

A user may control the massage device 100 using the massage device control device 200 . The massage device control device 200 is an electronic device (or electronic device) used by a user and may be connected to the massage device 100 through wired communication and/or wireless communication. The massage device control device 200 may include a remote controller, a cellular phone, a personal digital assistant (PDA), and the like, but is not limited thereto, and through wired or wireless communication with the massage device 100. A variety of connectable electronic devices may be included.

The arm massage unit 2000 is disposed on both sides of the body massage unit 1000 and may provide a massage to the user's arms through air cells formed inside the arm massage unit 2000 .

The leg massage unit 3000 is disposed below the body massage unit 1000 and may provide a massage to the user's legs through an air cell formed inside the leg massage unit 3000 . For example, the leg massage unit 3000 may include a calf massage unit that massages the user's calves and/or a foot massage unit that massages the user's feet.

The length of the leg massage unit 3000 may be adjustable according to the user's physical characteristics. For example, when a tall user uses the massage device 100, the length of the leg massage unit 3000 needs to be increased because the length of the calf is long. In addition, when a short user uses the massage device 100, the length of the calf is short, so the leg massage unit 3000 needs to be shortened. Accordingly, the leg massage unit 3000 may provide a leg massage customized to the height of the user.

The massage device 100 may further include at least one air cell (not shown) in a region other than the arm massage unit 2000 and the leg massage unit 3000 . At least one air cell may be located on the user's shoulder or pelvis, but is not limited thereto and may be disposed on various parts of the massage device 100.

The massage device 100 may include an air supply unit, and the air supply unit may inflate (or inflate) the air cell by supplying air to the air cell. The air supply unit may be located inside the body massage unit 1000 and may be located in the leg massage unit 3000, but is not limited thereto. For example, the air supply unit may be located outside the massage device 100.

The shape and structure of the massage device 100 shown in FIG. 1 are merely exemplary, and various types of massage device 100 are also of the present disclosure, unless departing from the scope of rights defined by the claims of the present disclosure. may be included within the scope.

2 is a diagram for explaining a main frame according to an embodiment of the present disclosure.

The main frame 1100 may include an upper frame 1150 having a massage module 1700 and a base frame 1110 supporting the upper frame 1150 .

The upper frame 1150 may include an upper frame 1152 and a lower frame 1153 . The upper frame 1152 may be a frame supporting the user's upper body (or back). The lower frame 1153 may be a frame supporting the user's lower body (or buttocks, hips).

A rack gear 1151 may be provided on at least a part of the upper frame 1150 . The rack gear 1151 is a member for guiding the vertical movement of the massage module 1700, and may include a plurality of valleys and a plurality of crests.

The rack gear 1151 may be provided on both sides of the upper frame 1150 in a form facing each other, and the massage module 1700 may move along the rack gear 1151 .

For example, the massage module 1700 may include a gear engaged with the rack gear 1151, and the gear rotates by an actuator provided in the massage module 1700, so that the massage module 1700 moves upward or can move downwards.

The rack gear 1151 may be implemented with a metal material or a plastic material. For example, the rack gear 1151 may be implemented with iron, steel, alloy, reinforced plastic, melamine resin, phenol resin, etc., but is not limited thereto.

The upper frame 1150 may be implemented in various shapes. For example, the upper frame 1150 may be classified into an S frame, an L frame, an S&L frame, and a double S&L frame depending on the shape, but is not limited thereto.

The S frame refers to a frame in which at least a part of the upper frame 1150 includes a curved shape like “S”. The L frame means a frame in which at least a part of the upper frame 1150 includes a curved shape like “L”, and the S&L frame means a frame including both a curved shape like “S” and a curved shape like “L” , and the double S&L frame means a frame that includes a curved shape like an L” and a curved shape like an “S” of two parts.

The base frame 1110 refers to a portion of the main frame 1100 that supports the upper frame 1150 and contacts the ground. The base frame 1110 may include a base upper frame 1111 and a base lower frame 1112 .

The base upper frame 1111 may support the upper frame 1150, and the base lower frame 1112 may contact the ground. In addition, the base upper frame 1111 may be located in contact with the base lower frame 1112 .

The base upper frame 1111 may move along the base lower frame 1112 . For example, the base upper frame 1111 may slide forward or backward along the base lower frame 1112 . In this case, the upper frame 1150 is connected to the base upper frame 1111 and can move according to the movement of the base upper frame 1111 .

For example, when the base upper frame 1111 moves forward, the upper frame 1150 may move forward together, and when the base upper frame 1111 moves rearward, the upper frame 1150 also moves backward. can move Due to this, the sliding movement of the body massaging unit 1000 may be permitted.

Specifically, in order to allow movement of the base upper frame 1111, a moving wheel may be provided below the base upper frame 1111. In addition, a guide member capable of guiding the moving wheel may be provided above the base lower frame 1112 . The moving wheel provided on the base upper frame 1111 moves along the guide member provided on the base lower frame 1112, so that the base upper frame 1111 can be moved forward or backward.

The massage device 100 may not provide a sliding function, and in this case, the base frame 1110 may not be separated into upper and lower frames.

3 is a diagram for explaining components of a massage device according to an embodiment of the present disclosure.

The massage device 100 may further include a control unit 1200, a sensor unit 1300, a network connection unit 1400, and a storage unit 1500.

The controller 1200 may be a processor that controls components of the massage device 100. For example, the controller 1200 may control various components of the massage device 100, such as the body massage unit 1000, the arm massage unit 2000, and the leg massage unit 3000.

The control unit 1200 may be implemented with one processor or may be implemented with a plurality of processors. When the controller 1200 is implemented as a plurality of processors, at least some of the plurality of processors may be physically located at a distance. In addition, the controller 1200 is not limited thereto and may be implemented in various ways.

The controller 1200 may control the operation of the massage device 100. For example, the massage device 100 may include a plurality of actuators, and the controller 1200 may control the operation of the massage device 100 by controlling the operation of the plurality of actuators. For example, the massage device 100 includes at least one of a vertical movement actuator, at least one actuator included in the massage module 1700, a back angle actuator, a leg angle actuator, a foot massage actuator, a leg length adjustment actuator, and a sliding actuator. It may include, and the controller 1200 may control the operation of the massage device 100 by controlling them.

The vertical movement actuator is an actuator that enables the vertical movement of the massage module 1700, and the massage module 1700 may move along the leg gear by the operation of the vertical movement actuator. Although the vertical movement actuator is included in the massage device 100, it is not limited thereto. For example, a vertical movement actuator may be included in the massage module 1700.

The isometric actuator is an actuator that adjusts an angle of a part of the massage device 100 in contact with a user's back, and the isometric angle of the massage device 100 can be adjusted by the operation of the isometric actuator.

The leg angle actuator is an actuator that adjusts the angle of the leg massage unit 300 of the massage device 1000, and the angle between the leg massage unit 3000 and the body massage unit 1000 can be adjusted by the operation of the leg angle actuator. there is.

The foot massage actuator refers to an actuator that operates a foot massage module included in the leg massage unit 3000 . Using the foot massage actuator, the massage device 100 may provide a foot massage to the user.

The massage module 1700 may include at least one actuator, and the controller 1200 may provide various massage operations by operating the at least one actuator. For example, the at least one actuator may be an acupressure actuator that allows the massage module 1700 to provide acupressure massage to the user, a kneading actuator that allows the massage module 1700 to provide a kneading massage to the user, and a massage module 1700 that provides a kneading massage to the user It may include a tapping actuator that allows a tapping massage to be provided. The controller 1200 may provide acupressure massage, kneading massage, and tapping massage by operating at least one actuator, but is not limited thereto and may provide various massage operations. The acupressure actuator may be an actuator that provides acupressure massage to the user by moving the massage module 1700 in the front and rear directions of the massage device 100 . The kneading actuator may be an actuator that rotates the massage module 1700 to provide a kneading massage to the user. The tapping actuator may be an actuator that provides a tapping massage to the user by rapidly moving (or vibrating) the massage module 1700 in the forward and backward directions.

The leg length adjustment actuator refers to an actuator that adjusts the length of the leg massage unit 3000 . For example, the controller 1200 may adjust the length of the leg massage unit 3000 according to the user by using the leg length adjusting actuator, and as a result, the user may receive a massage suitable for the body shape.

The sliding actuator enables the sliding operation of the massage device 100. For example, the horizontal base upper frame 1114a may move forward or rearward by the operation of the sliding actuator, and as a result, the upper frame connected to the horizontal base upper frame 1114a may also move forward or rearward.

The sensor unit 1300 may acquire various types of information using at least one sensor. For example, the sensor may include various sensors such as a biometric information acquisition sensor, a sound sensor, a pressure sensor, an infrared sensor, and an LED sensor, but is not limited thereto. At this time, the biometric information acquisition sensor may acquire the user's fingerprint information, face information, voice information, iris information, weight information, electrocardiogram information, body composition information, etc., but is not limited thereto, and includes various biometric information of the user. can do.

The massage device 100 may detect a contact area and/or a contact position between the user and the massage device 100 through the sensor unit 1300 . Also, the massage device 100 may acquire user's shoulder position information through the sensor unit 1300 . In addition, the massage device 100 may provide customized massage suitable for the user's body size based on the acquired information. For example, when the massage device 100 provides a shoulder massage, the massage device 100 recognizes the user's shoulder position based on information acquired through the sensor unit 1300, and places the shoulder in the recognized shoulder position. Massage can be provided.

The massage device 1000 may include a network connection unit 1400. The network connection unit 1400 may communicate with a module inside the massage device 1000, an external electronic device, and/or the massage device control device 200 through any type of network. The network connection unit 1400 may include a wired/wireless connection module for network access. As the wireless access technology, for example, Wireless LAN (WLAN) (Wi-Fi), Wibro (Wireless broadband), Wimax (World Interoperability for Microwave Access), HSDPA (High Speed Downlink Packet Access), etc. may be used. . As a wired access technology, for example, Digital Subscriber Line (XDSL), Fibers to the home (FTTH), Power Line Communication (PLC), and the like may be used. In addition, the network connection unit may include a short-distance communication module to transmit/receive data with any device/terminal located in a short distance. For example, Bluetooth, Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra Wideband (UWB), ZigBee, and the like may be used as short range communication technologies. Not limited.

The storage unit 1500 may store various information related to the massage device 100. For example, the storage unit 1500 may include massage control information and personal authentication information, but is not limited thereto.

The storage unit 1500 may be implemented through a non-volatile storage medium capable of continuously storing arbitrary data. For example, the storage unit 1500 may include, but is not limited to, disks, optical disks, and magneto-optical storage devices, as well as flash memory and/or battery-backed memory-based storage devices. It doesn't work.

Also, the storage unit 1500 may include a memory. Memory is the main storage device that the processor directly accesses, such as random access memory (RAM) such as dynamic random access memory (DRAM) and static random access memory (SRAM). It may mean a volatile (volatile) storage device erased with, but is not limited thereto. Such memory may be operated by the controller 1200 .

4 is a diagram for explaining an external electronic device capable of communicating with a massage device according to an embodiment of the present disclosure.

The massage device 100 communicates with the external electronic device 400 by wire or wirelessly, and can transmit/receive various data. The external electronic device 400 may be an electronic device used by various objects such as a user and/or a manager managing the user.

For example, the external electronic device 400 may include various devices such as a personal computer (PC) used by a user using the massage device 100, a data server, a display, and a portable electronic device 410. Not limited to this. The portable electronic device 410 is dedicated to the massage device 100 or a general-purpose portable electronic device such as an AI speaker, a laptop computer, a mobile phone, a smart phone, a tablet PC, and a mobile Internet device. Internet device (MID)), PDA (personal digital assistant), EDA (enterprise digital assistant), digital still camera, digital video camera, PMP (portable multimedia player), PND (personal navigation) device or portable navigation device), a handheld game console, an e-book, and a smart device. In this case, the smart device may include a wearable device 420 such as a smart watch or a smart band.

In addition, the external electronic device 400 is a massage device 430 other than the massage device 100 currently used by the user, a hospital server 440, a personal health record (PHR) server, and/or a cloud server 450. ) may be included. The external electronic device may include, but is not limited to, a medical device capable of measuring biometric information, such as an electronic weight scale, blood glucose meter, or blood pressure meter. For example, a medical device may be included in the massage device 100 .

Figure 5a shows an example for explaining the electrocardiogram induction method according to an embodiment of the present disclosure, Figure 5b shows another example for explaining the electrocardiogram induction method according to an embodiment of the present disclosure, Figure 5c shows this Another example for explaining an electrocardiogram induction method according to an embodiment of the disclosure is shown.

5A to 5C, electrocardiogram data (ECG data or electrocardiogram; ECG) is obtained through a plurality of electrocardiogram induction methods (or a plurality of electrocardiogram measurement methods) using electrodes in contact with an object. Can be measured (or derived). In this case, the plurality of electrocardiogram induction methods may be methods of measuring (or inducing) electrocardiogram data of an object using electrodes. The object may refer to a life form having a heart, such as a human and/or animal. However, in the following description, it is assumed that the object is a person for convenience of description.

A plurality of electrocardiogram induction methods may be classified according to electrodes used to measure electrocardiogram data. The plurality of electrocardiogram induction methods may differ from each other in at least one of a body contact position of an actual electrode used when measuring electrocardiogram data and the number of electrodes.

For example, the plurality of ECG induction methods may include at least one ECG induction method among a first ECG induction method and a second ECG induction method.

The first electrocardiogram induction method may be a method of measuring electrocardiogram data using actual electrodes. The actual electrode may be contacted (or attached) to an object. For example, actual electrodes may be attached to various body parts of the object, such as arms, legs, and/or chest (or chest) of the object, as shown in FIG. 5A. A real electrode may be activated to measure electrocardiogram data, and the activated real electrode may be referred to as an active electrode.

The first ECG induction method may be a standard limb induction method as shown in FIGS. 5A and 5B , but is not limited thereto. Standard limb induction modalities may include Lead I modality, Lead II modality and Lead III modality. The Lead I method may be a method of measuring electrocardiogram data by inducing a potential difference between the right and left arms of the object through electrodes respectively attached to the right and left arms of the object. The Lead II method may be a method of measuring electrocardiogram data by inducing a potential difference between the right arm and left leg through electrodes attached to the right arm and left leg of the object, respectively. The Lead III method may be a method of measuring electrocardiogram data by inducing a potential difference between the left plate and the left leg through electrodes respectively attached to the left arm and left leg of the object. At this time, the electrode attached to the left leg may not actually be used as a ground electrode.

The second electrocardiogram induction method may be a method of measuring electrocardiogram data using real electrodes and virtual electrodes (or inductive grounding). In this case, the virtual electrode may be a virtual electrode generated (or determined) by a real electrode. As shown in FIG. 5C , the virtual electrode may be located at the center point of an Eindhoven's triangle formed through actual electrodes attached to the arms and legs of the object.

For example, the second electrocardiogram induction method may be at least one of an amplification limb induction method and a chest induction method as shown in FIGS. 5A and 5C , but is not limited thereto.

The amplification limb induction scheme may include an aVR scheme, an aVL scheme, and/or an aVF scheme. The aVR method may be a method of measuring electrocardiogram data by inducing a potential difference between the right arm and the center point of the Eindhoven triangle through electrodes respectively attached to the right arm, left arm, and left leg of the object. The aVL method may be a method of measuring electrocardiogram data by inducing a potential difference between the left arm and the center point of the Eindhoven triangle through electrodes respectively attached to the right arm, left arm, and left leg of the object. The aVF method may be a method of measuring electrocardiogram data by inducing a potential difference between the left leg and the center point of the Eindhoven triangle through electrodes respectively attached to the right arm, left arm, and left leg of the object.

The chest induction method may include V1 method, V2 method, V3 method, V4 method, V5 method, and V6 method. The V1 method obtains electrocardiogram data by inducing a potential difference between the 4th intercostal space of the right sternal border of the chest and the center point of the Eidhoven triangle through electrodes attached to the 4th intercostal space, right arm, left arm and left leg of the right sternum border of the chest of the subject, respectively. It could be a way to measure it. The V2 method induces a potential difference between the 4th intercostal space of the left sternal border of the chest and the center point of the Eidhoven triangle through electrodes attached to the 4th intercostal space of the left sternal border of the chest, right arm, left arm and left leg of the object, respectively, to obtain electrocardiogram data. It could be a way to measure it. The V3 method is the body part between the 4th intercostal space on the left sternum border of the chest and the 5th intercostal space on the midline of the left clavicle of the chest, and the 4th intercostal space on the left sternum border of the chest and the chest through electrodes attached to the right arm, left arm, and left leg, respectively. It may be a method of measuring electrocardiogram data by inducing a potential difference between a body part between the 5th intercostal space on the left clavicle midline and the center point of the Eidhoven triangle. The V4 method induces a potential difference between the 5th intercostal space on the midline of the left clavicle of the chest of the subject, the 5th intercostal space on the midline of the left clavicle of the chest and the center point of the Eidhoven triangle through electrodes attached to the right arm, left arm and left leg, respectively, to obtain electrocardiogram data. It could be a way to measure it. The V5 method derives electrocardiogram data by inducing a potential difference between the 5th intercostal space on the left anterior fossa line of the chest and the center point of the Eidhoven triangle through electrodes attached to the 5th intercostal space, right arm, left arm and left leg respectively on the left anterior fossa line of the chest of the object. It could be a way to measure it. The V6 method obtains electrocardiogram data by inducing a potential difference between the 5th intercostal space on the left axillary midline of the chest and the center point of the Eidhoven triangle through electrodes attached to the 5th intercostal space on the left axillary midline of the chest, right arm, left arm and left leg, respectively. It could be a way to measure it.

As described above, the plurality of electrocardiogram induction methods may include at least one of the standard limb induction method, the amplified limb induction method, and the chest induction method, but are not limited thereto. For example, the direction (or axis) of the electrode toward the heart may be different according to the position of the body of the object in contact with the object. A plurality of electrocardiogram induction methods include an electrocardiogram induction method using electrodes (or electrodes attached differently) in a direction facing the heart different from the electrodes used in the standard limb induction method, amplified limb induction method, and/or chest induction method, and/or An electrocardiogram induction method for measuring electrocardiogram data through various electrodes, such as an electrocardiogram induction method using a patch-type electrode, which is an actual electrode configured in a patch shape, may be included.

As described above, the standard limb induction scheme, the amplified limb induction scheme, and the thoracic induction scheme may be referred to as the 12 induction scheme.

6 illustrates an example for describing electrocardiogram data according to an embodiment of the present disclosure.

Electrocardiogram data is a signal recording electrical activity of the heart, and may be data indicating (or recording) a change in potential between electrodes in contact with an object. In this case, since the potential change is generated by the heart movement, it may represent the heart movement of the object. For example, a potential between electrodes may be changed based on an electrical signal generated by heart motion.

Electrocardiogram data may be measured on the body surface of the object through electrodes attached to the body surface. For example, as shown in FIG. 6 , electrocardiogram data may be measured on the body surface of the object through each of a plurality of electrocardiogram induction methods. In this case, as shown in FIG. 6, the electrocardiogram data may be configured by editing time (about 10 seconds) on the x-axis and potential on the y-axis, and editing each electrocardiogram data at intervals of 2.5 seconds.

When ECG data is measured through Lead I method, Lead II method, Lead III method, aVR method, aVL method, aVF method, V1 method, V2 method, V3 method, V4 method, V5 method, and V6 method, the measured ECG The data includes Lead I electrocardiogram data measured through the Lead I method, Lead II electrocardiogram data measured through the Lead II method, Lead III electrocardiogram data measured through the Lead III method, aVR electrocardiogram data measured through the aVR method, and aVL method. aVL ECG data measured through aVF method, aVF ECG data measured through aVF method, V1 ECG data measured through V1 method, V2 ECG data measured through V2 method, V3 ECG data measured through V3 method, V4 method V4 electrocardiogram data measured through, V5 electrocardiogram data measured through the V5 method, and V6 electrocardiogram data measured through the V6 method may be included.

Existing learning methods generate a dedicated deep learning model customized for a specific training data set by inputting a specific training data set to the deep learning model and learning the deep learning model. In this case, the specific learning data set may be composed of specific ECG data measured through a specific ECG induction method.

For example, the Lead I method, the Lead II method, the Lead III method, the aVR method, the aVL method, the aVF method, the V1 method, the V2 method, the V3 method, the V4 method, the V5 method, and the V6 method are all used for measurement at a specific point in time. ECG data obtained may be referred to as standard 12-lead ECG data. Standard 12 lead ECG data is Lead I ECG data, Lead II ECG data, Lead III ECG data, aVR ECG data, aVL ECG data, aVF ECG data, V1 ECG data, V2 ECG data, V3 ECG data, V4 ECG data, V5 ECG data and V6 ECG data may be included.

Existing learning methods generate a dedicated deep learning model customized for standard 12-lead ECG data by learning deep learning data by simultaneously inputting all of the ECG data included in standard 12-lead ECG data into a deep learning model.

However, the dedicated deep learning model learned through the existing learning method has a problem in analyzing only the ECG data corresponding to the dedicated deep learning model. For example, a dedicated deep learning model tailored to the standard 12-lead ECG data must analyze the standard 12-lead ECG data only when all of the ECG data included in the standard 12-lead ECG data are simultaneously entered into the dedicated deep learning model to analyze the heart rate of the object. disease can be diagnosed.

Therefore, since the existing learning method cannot analyze various ECG data through a single deep learning model, there is a problem of generating a plurality of dedicated deep learning models customized for each of a plurality of specific training data sets, and each dedicated deep learning model. There is a problem of analyzing only the ECG data corresponding to each dedicated deep learning model through the model.

7 shows an example for explaining a health diagnosis system according to an embodiment of the present disclosure.

The health diagnosis system 10 includes a data providing device 460 , a data measuring device 470 and a health diagnosis device 480 .

The data providing device 460 may store a plurality of electrocardiogram data of a plurality of objects in a data base (DB) of the data providing device 460 . In this case, the plurality of ECG data may be ECG data provided from an arbitrary device that measures ECG data, such as the data measuring device 470 and/or the data measuring device 470 . The plurality of ECG data may be ECG data measured from a plurality of objects through a plurality of ECG induction methods. The number of the plurality of ECG data may be greater than the number of the plurality of ECG derivation schemes.

For example, the data measuring device 470 derives a heart movement of each of a plurality of objects through a plurality of electrocardiogram induction schemes at each of a plurality of points in time and measures a plurality of electrocardiogram data for each of a plurality of objects (or can be created). In this case, the data measurement device 470 may measure ECG data for each object using the same or different ECG derivation methods at each time point.

The data measuring device 470 may perform all of the Lead I method, Lead II method, Lead III method, aVR method, aVL method, aVF method, V1 method, V2 method, V3 method, V4 method, V5 method, and V6 method at any point in time. When electrocardiogram data is measured from an arbitrary object, the data measuring device 470 measures Lead I electrocardiogram data, Lead II electrocardiogram data, Lead III electrocardiogram data, aVR electrocardiogram data, aVL electrocardiogram data, and aVF electrocardiogram data for the arbitrary object. data, V1 ECG data, V2 ECG data, V3 ECG data, V4 ECG data, V5 ECG data, and V6 ECG data.

The data measurement device 470 may transmit a plurality of ECG data for a plurality of objects measured through a plurality of ECG induction methods to the data providing device 460 .

The data providing device 460 may store a plurality of ECG data of a plurality of objects transmitted from the data measuring device 470 in a database.

Each of the plurality of electrocardiogram data stored in the data providing device 460 may be labeled with environment information, which is information about a measurement environment when electrocardiogram data is measured. In this case, the environmental information may be information obtained from the data measuring device 470 and various devices (eg, sensors) capable of communicating with the data measuring device 470 when measuring the ECG data.

For example, the environment information includes first identifier information representing an ECG induction method corresponding to each of a plurality of ECG data, second identifier information representing an object corresponding to each of the plurality of ECG data, and each of the plurality of ECG data. It may include at least one of health information representing the health state of the corresponding object.

The first identifier information may be information indicating an ECG induction method, such as a name of the ECG induction method. For example, Lead I electrocardiogram data may be data labeled as 'Lead I method', Lead II electrocardiogram data may be data labeled as 'Lead II method', and Lead III electrocardiogram data may be data labeled as 'Lead III method'. . The aVR electrocardiogram data may be labeled with 'aVR method', the aVL electrocardiogram data may be labeled with 'aVL method', and the aVF electrocardiogram data may be labeled with 'aVF method'. For V1 ECG data, 'V1 method' is data labeled, for V2 ECG data, 'V2 method' is data labeled, for V3 ECG data, 'V3 method' is data labeled, and for V4 ECG data, 'V4 method' is data labeled 'V5 method' may be data labeled for V5 ECG data, and 'V6 method' may be data labeled for V6 ECG data.

The second identifier information may be information indicating an object, such as an ID and/or a name of the object.

The health information may include at least one of information indicating whether the object is healthy, information about a disease possessed by the object, and information about a physical examination performed on the object.

The information indicating whether the object is healthy may indicate whether the object is any one of a healthy object (or healthy object) and an unhealthy object (or non-healthy object). In this case, the healthy object may refer to an object that does not have a disease and has normal levels of anemia and minerals. An unhealthy object may refer to an object having a disease or having an abnormal level of at least one of anemia level and mineral level. The disease may be a disease (or heart disease) that can occur in the heart, such as arrhythmia, atrial fibrillation, and/or cardiovascular disease. The cardiovascular disease may be various diseases related to the cardiovascular system, such as angina pectoris, coronary artery disease such as myocardial infarction, valve disease, heart failure, pericardial disease, hypertension, arteriosclerosis, and/or cardiomyopathy. Minerals are called minerals and may include calcium, phosphorus, magnesium, potassium, sodium, chlorine and trace elements such as iron, copper, sulfur, iodine, manganese, cobalt and/or zinc.

For example, electrocardiogram data measured from a healthy object (or electrocardiogram data corresponding to a healthy object) may be labeled with 'healthy object', which is identifier information indicating a healthy object. ECG data measured from an unhealthy object (or ECG data corresponding to an unhealthy object) may be labeled with 'unhealthy object', which is identifier information representing the unhealthy object.

The information about the disease possessed by the object may include various detailed information about the disease, such as the name of the disease, the progress of the disease, and/or the risk of the disease. In this case, the disease may be a disease that can occur in the heart as described above. For example, information about a disease possessed by an object may be data labeled on electrocardiogram data measured from a non-healthy object.

The information on the physical examination may include information on a physical examination performed within a threshold time and/or on the same day based on the time point at which the electrocardiogram data is measured. At this time, the threshold time may be preset in various ways, such as about 30 minutes and/or 12 hours. For example, the physical examination information may include information on various tests performed on the object, such as a blood test (or blood test) performed on the object, an echocardiogram test, an angiogram test, and a computed tomography test. The blood test information may include various information obtained through a blood test, such as anemia level (or hemoglobin level) and/or mineral level (or mineral content).

The environment information includes first information about an electronic device of an object corresponding to each of a plurality of electrocardiogram data, second information about motion of an object corresponding to each of a plurality of electrocardiogram data, and object corresponding to each of a plurality of electrocardiogram data. It may further include at least one piece of third information about the body of the body.

The first information may include information on a first electromagnetic signal, which is an electromagnetic signal generated by an electronic device of an object while electrocardiogram data is measured. In this case, the electronic device is an electronic device in contact with the object, and may be at least one of an external electronic device located outside the object and an internal electronic device located (or implanted) inside the object.

The second information is at least one of a second electromagnetic signal, which is an electromagnetic signal generated by muscle movement of an object while electrocardiogram data is measured, and a third electromagnetic signal, which is an electromagnetic signal generated by respiration of an object while electrocardiogram data is measured. It may contain information about electromagnetic signals. In this case, the second electromagnetic signal may be an electromagnetic signal generated when muscles of the object move, and the third electromagnetic signal may be an electromagnetic signal generated when the object breathes.

The third information may include information on at least one of information about the object's gender, age, and obesity. In this case, the degree of obesity may be determined based on the weight and/or height of the object.

The data providing device 460 may provide the health diagnosis device 480 with a plurality of electrocardiogram data for a plurality of objects stored in the data providing device 460 .

The data measuring device 470 may measure (or generate) ECG data for a target object by inducing the movement of the heart of the target object using at least two or more ECG induction methods among a plurality of ECG induction methods at a specific time point. there is. In this case, the target object may be a diagnosis target for diagnosing heart disease. Electrocardiogram data of the target object may be electrocardiogram data measured through at least two or more electrocardiogram induction methods. The number of ECG data for the target object may be equal to the number of at least two ECG derivation schemes.

The data measurement device 470 may provide the electrocardiogram data of the target object to the health diagnosis device 480 . In this case, the electrocardiogram data of the target object may be labeled with third identifier information indicating an electrocardiogram derivation method corresponding to each of the electrocardiogram data of the target object and fourth identifier information indicating the target object. The third identifier information may be information indicating an ECG induction method, such as a name of the ECG induction method. The fourth identifier information may be information representing the target object, such as ID and/or name of the target object.

The health diagnosis device 480 may train a deep learning model using a plurality of electrocardiogram data for a plurality of objects provided from the data providing device 460 . In this case, the deep learning model may be a model in which a deep learning algorithm for diagnosing heart disease is modeled (or built). A deep learning model is modeled with a deep learning algorithm, but is not limited thereto. For example, the deep learning model may be modeled through various techniques such as machine learning techniques such as random forests and/or support vector machines and/or neural networks.

For example, the health diagnosis device 480 may input each of a plurality of electrocardiogram data provided from the data providing device 460 into a deep learning model. The health diagnosis device 460 may learn the deep learning model using the ECG data whenever each ECG data is input to the deep learning model.

The health diagnosis device 480 may diagnose the health state of the target object by analyzing electrocardiogram data of the target object provided from the data measuring device 470 using the learned deep learning model. In this case, the health state of the target object may mean a health state (or heart health state, heart health state) of the heart of the target object.

The above-described data providing device 460 may be an electronic device that is independently and distinctly implemented from the massage device 100 . For example, the data providing device 470 may be an electronic device in which a plurality of ECG data is stored (or has learning data), such as a hospital, institution, or company. The aforementioned data measuring device 470 and/or health diagnosis device 480 may be implemented in the massage device 100 or may be an electronic device implemented independently and distinct from the massage device 100.

The aforementioned electronic device may be various devices such as a personal computer (PC), a server, a module, or a portable electronic device. Portable electronic devices include a laptop computer, a mobile phone, a smart phone, a tablet PC, a mobile internet device (MID), a personal digital assistant (PDA), and an enterprise digital assistant (EDA). ), digital still camera, digital video camera, portable multimedia player (PMP), personal navigation device or portable navigation device (PND), handheld game console, e-book (e-book), can be implemented as a smart device (smart device). In this case, the smart device may be implemented as a smart watch or a smart band. Preferably, the data measurement device 470 may be implemented as a module, and the data providing device 460 and health diagnosis device 480 may be implemented as a server.

8A shows an example for describing a data measurement device according to an embodiment of the present disclosure, and FIG. 8B shows another example for describing a data measurement device according to an embodiment of the present disclosure.

Referring to FIG. 8A , the data measuring device 470 may be implemented as a separate electronic device not included in the massage device 100.

For example, the data measurement device 470 may be implemented with various electronic devices such as a Holter device capable of acquiring electrode waveforms from a human body, a wearable device, and/or a medical device. In this case, the wearable device may be at least one of a body-worn device (eg, a head-worn device, a bracelet, an anklet, a necklace, a ring, a watch, etc.), a clothing-integrated device, a body-worn device, and a bio-implantable device. can

The data measuring device 470, which is not included in the massage device 100, induces the heart movement of the object using an electrode that is in contact with at least one of the arms, legs, and chest of the object to measure electrocardiogram data for the object. . For example, the data measuring device 470 may use Lead I electrocardiogram data, Lead II electrocardiogram data, Lead III electrocardiogram data, aVR electrocardiogram data, aVL electrocardiogram data, and aVL of the object by using a plurality of electrodes in contact with the arms, legs, and chest of the object. ECG data, aVF ECG data, V1 ECG data, V2 ECG data, V3 ECG data, V4 ECG data, V5 ECG data, and V6 ECG data can be measured.

Referring to FIG. 8B , the data measurement device 470 may be implemented as an electronic device included in the massage device 100.

For example, the data measurement device 470 may be included in the arm massage unit 2000 and/or the leg massage unit 3000 of the massage device 100 . At this time, the data measuring device 470 is included in at least one arm massage unit 2000 among the two arm massage units 2000 and/or at least one leg massage unit 3000 among the two leg massage units 3000. can

Specifically, the data measurement device 470 may be disposed inside the arm massage unit 2000 and/or the leg massage unit 3000 . The inside of the arm massage unit 2000 may accommodate an arm of an object, and the inside of the leg massage unit 3000 may accommodate a leg of an object.

When the data measurement device 470 is included in the arm massage unit 2000, the data measurement device 470 is disposed on one surface of the arm massage unit 2000 and is formed in an upwardly convex shape so that the palm of the object is seated. can At this time, one surface of the arm massage unit 2000 may be an area that comes into contact with the palm of the object as an inner lower surface of the arm massage unit 2000 . Through this, the object may seat its arm into the arm massage unit 2000 so that the palm of the object is positioned on the data measuring device 470 . The palm of the object is supported by the data measuring device 470 so that the arm can be comfortably seated on the arm massager 2000 .

When the data measurement device 470 is included in the leg massage unit 3000, the data measurement device 470 may be disposed on one side of the calf massage unit or one side of the foot massage unit of the leg massage unit 3000. In this case, one side of the calf massage unit may be a rear surface of the calf massage unit and may be an area that contacts the calf of the object. One surface of the foot massager may be an inner lower surface of the foot massager and/or an outer upper surface of the foot massager, which is in contact with the sole of the object. Through this, the object places the leg into the leg massage unit 3000 so that the calf or sole is located in the data measurement device 470 or the leg massage unit 3000 so that the sole is located in the data measurement device 460. Legs can be placed outside.

The data measurement device 470 included in the massage device 100 may include an electrode unit 472 formed (or disposed) on an outer surface of the data measurement device 470 . For example, the electrode unit 472 may be formed on an outer surface of the data measuring device 470 and exposed to the outside of the data measuring device 470 . Through this, the electrode unit 472 may contact the object's body (eg, the palm, sole, and/or calf) of the object. The electrode unit 472 may be formed of at least two or more electrodes.

The data measurement device 470 included in the massage device 100 may measure electrocardiogram data of the object by inducing heart movement of the object using the electrode unit 472 . For example, the data measuring device 470 uses the electrode units 472 disposed on both the two arm massage units 2000 and the two leg massage units 3000 to provide Lead I electrocardiogram data and Lead II for an object. ECG data, Lead III ECG data, aVR ECG data, aVL ECG data, and aVF ECG data can be generated.

As described above, the data measuring device 470 measures electrocardiogram data of an object through the electrode unit 472, but is not limited thereto. For example, the data measuring device 470 may additionally measure electrocardiogram data through a chest induction method using an additional electrode distinct from the electrode unit 472 . In this case, the additional electrode may be an electrode (or electrodes) contacting a body part (eg, the chest of the object) other than the arms and legs of the object.

9 illustrates an example for explaining a health diagnosis device according to an embodiment of the present disclosure.

The health diagnosis device 480 may include a memory 481 and a processor 483 .

Memory 481 may store instructions (or programs) executable by processor 483 . For example, the instructions may include instructions for executing the operation of the health diagnosis device 480 and/or the operation of each component of the health diagnosis device 480 .

The processor 483 may process data stored in the memory 481 . The processor 483 may execute computer readable code (eg, software) stored in the memory 481 and instructions triggered by the processor. The processor 483 may be a hardware-implemented data processing device having a circuit having a physical structure for executing desired operations. For example, desired operations may include codes or instructions included in a program. For example, a data processing unit implemented in hardware includes a microprocessor, a central processing unit, a processor core, a multi-core processor, and a multiprocessor. , Application-Specific Integrated Circuit (ASIC), and Field Programmable Gate Array (FPGA).

10 is a flowchart for explaining a deep learning model learning operation of a processor according to an embodiment of the present disclosure, and FIG. 11 is an example for explaining electrical signals included in electrocardiogram data according to an embodiment of the present disclosure. , and FIG. 12 shows another example for explaining a deep learning model learning operation of a processor according to an embodiment of the present disclosure.

Referring to FIG. 10 , the processor 483 may learn a deep learning model using a plurality of electrocardiogram data for a plurality of objects provided from the data providing device 460 .

For example, the processor 483 may obtain a plurality of ECG data for a plurality of objects measured through a plurality of ECG derivation schemes (S1010). In this case, each of the plurality of ECG data may be ECG data corresponding to each of the plurality of ECG induction methods.

The processor 483 may determine at least one ECG data among a plurality of ECG data as training data of the deep learning model (S1030). In this case, the learning data may be electrocardiogram data for learning the deep learning model.

For example, the processor 483 may determine all of a plurality of ECG data as learning data or may determine some of the plurality of ECG data as learning data.

The processor 483 may extract the learning data in units of sub data (S1050). In this case, the sub data unit may be a reference unit for extracting learning data. The sub data unit may be a unit corresponding to a single piece of training data.

For example, the processor 483 may divide the training data into sub-training data of sub-data units based on the sub-data units. When the sub data unit is a unit corresponding to a single training data, the processor 483 may determine each of the training data as sub training data based on the sub data unit. The sub-learning data is learning data in units of sub-data and may be a single training data.

The processor 483 may randomly extract each of the sub-learning data based on a random extraction method.

The processor 483 may learn a deep learning model based on the extracted sub-training data whenever the sub-training data is extracted (S1070). For example, the processor 483 may learn the deep learning model by inputting the extracted sub learning data to the deep learning model whenever sub learning data is extracted. In other words, the processor 483 may learn the deep learning model by inputting the extracted sub-learning data, which is the training data of the extracted sub-data unit, to the deep learning model (S1070). The learning data of the sub-data unit is any one of the sub-learning data and may be a single learning data.

The processor 483 learns the deep learning model by inputting the extracted learning data in units of sub data to the deep learning model, but is not limited thereto.

For example, the training data of the sub-data unit may be original ECG data including various noises as ECG data measured from any one object among a plurality of objects. Original ECG data may include a plurality of frequencies, and at least one frequency among the plurality of frequencies may be noise.

Specifically, the original electrocardiogram data is an electrical signal generated from an electronic device implanted in an object and/or an electronic device in contact with the object, an electrical signal generated between the skin of the object and an electrode in contact with the skin, and a wire connected to the electrode. High-frequency noise may be included due to various reasons, such as electrical signals generated inside the data providing device 460 and/or data measuring device 470 . Original electrocardiogram data may include low-frequency noise due to the reason that the object breathes while the object's chest rises and falls.

Referring to FIG. 11 , original ECG data may include a frequency of an electrical signal similar to noise among electrical signals generated from the heart of an object. At this time, the frequency of the electrical signal similar to noise is the frequency of the spike waveform as shown by the arrow shown in FIG. 11 and may be the same as or similar to the noise frequency.

For example, the electrical signal similar to noise is an electrical signal generated from a pacemaker implanted in a non-healthy object and/or an electrical signal generated from atrial fibrillation of a heart with atrial fibrillation disease, which is an example of arrhythmia, and the like. It could be various electrical signals from an abnormal heart.

Referring to FIG. 12, the processor 483 filters the training data of the extracted sub-data unit to remove noise included in the training data of the extracted sub-data unit, and produces correction data obtained by filtering the training data of the sub-data unit. and learn a deep learning model based on the calibration data.

However, the processor 483 may recognize and remove the frequency of an electrical signal generated from the heart itself included in the extracted training data of the sub data unit as noise. For example, when the frequency of the spike waveform shown in FIG. 11 is not high, the processor 483 may recognize the frequency of the spike waveform as noise and remove it.

In other words, the filtering operation of the processor 483 has a problem in that not only the noise but also the frequency of the electric signal generated by the heart itself is filtered.

In order to solve the above problem, the processor 483 may learn a deep learning model based on the training data and correction data of the extracted sub-data unit.

For example, the processor 483 filters the extracted learning data of the sub-data unit (hereinafter referred to as 'learning data of the sub-data unit') whenever the learning data of the sub-data unit is extracted so that the learning data of the sub-data unit is filtered. corrected data can be generated. The processor 483 may learn the deep learning model by simultaneously inputting training data and correction data in units of sub data to the deep learning model. In this case, the correction data may be ECG data from which a frequency corresponding to noise is filtered (or deleted) from among a plurality of frequencies included in the original ECG data.

As another example, the processor 483 may filter the correction data to generate recalibration data obtained by filtering the correction data. The processor 483 simultaneously inputs training data and recalibration data in units of sub data to the deep learning model, or simultaneously inputs training data, correction data, and recalibration data in units of sub data to the deep learning model to learn the deep learning model. can In this case, the recalibration data may be electrocardiogram data from which a specific frequency is filtered (or deleted) from among a plurality of frequencies included in the correction data.

As described above, in order to solve the problem of the existing method, the processor 483 does not simultaneously input all of the training data included in a specific training data set to the deep learning model when learning the deep learning model, and applies each of the training data to the deep learning model. You can learn a deep learning model through a single training data by inputting it to the model.

Through this, the processor 483 may provide a general-purpose deep learning model capable of analyzing various ECG data, not a deep learning model that analyzes only specific ECG data.

In addition, the processor 483 simultaneously uses correction and/or recalibration data as well as original electrocardiogram data, which is learning data in units of sub data, to solve the problem of removing noise as well as electrical signals generated from the heart itself. model can be trained.

Through this, the processor 483 may provide a general-purpose deep learning model with higher electrocardiogram analysis accuracy than a deep learning model learned using original electrocardiogram data, correction data, or recalibration data.

In addition, when learning a deep learning model by using some rather than all of a plurality of ECG data, the processor 483 provides a deep learning model learning time and learning cost (for example, a deep learning model for learning a general-purpose deep learning model). cost of purchasing equipment to train the running model, etc.) can be saved.

13A shows an example for explaining an operation of determining learning data of a processor according to an embodiment of the present disclosure, and FIG. 13B illustrates another example for explaining an operation of determining learning data of a processor according to an embodiment of the present disclosure. 13C shows another example for explaining an operation of determining learning data of a processor according to an embodiment of the present disclosure.

Referring to FIG. 13A , the processor 483 may determine all of a plurality of ECG data of a plurality of objects as learning data (S1031).

Referring to FIGS. 13B and 13C , the processor 483 may determine some of a plurality of electrocardiogram data for a plurality of objects as learning data.

For example, as shown in FIG. 13B , the processor 483 may determine two or more ECG data from among a plurality of ECG data as learning data based on the number of ECG data (S1033).

As another example, the processor 483 may determine representative ECG data representing each object from among a plurality of ECG data based on the representative ECG derivation method representing each object as shown in FIG. 13C (S1035). At this time, The representative ECG induction method may be an ECG induction method representing each object among ECG induction methods corresponding to each object. The representative ECG data may be ECG data corresponding to a representative ECG induction method representing each object among ECG data corresponding to each object.

The processor 483 may determine representative ECG data representing each object as candidate training data of the deep learning model (S1037). In this case, the candidate learning data may be electrocardiogram data that may be determined as learning data.

The processor 483 may determine at least one candidate training data among the candidate training data as training data (S1039). For example, the processor 483 may determine all of the candidate training data as training data or some of the candidate training data as training data.

14 shows an example for explaining the learning data determination operation shown in FIG. 13B.

The processor 483 may calculate the total number of ECG data, which is the number of a plurality of ECG data for a plurality of objects (S1033a). For example, the processor 483 may calculate the total number of ECG data by counting each of a plurality of ECG data.

The processor 483 may determine the number of ECG data, which is the number of ECG data to be selected as learning data, among a plurality of ECG data, based on the total number of ECG data (S1033b).

For example, the processor 483 may determine the number of ECG data based on whether the total number of ECG data is odd or even. When the total number of ECG data is an odd number, the processor 483 may determine, as the number of ECG data, any odd number among a number equal to or greater than half of the total number of ECG data and the total number of ECG data. When the total number of ECG data is an even number, the processor 483 may determine any one of an even number from a number equal to or greater than half of the total number of ECG data to the total number of ECG data as the number of ECG data.

The processor 483 may select two or more ECG data from among a plurality of ECG data based on the number of ECG data (S1033c). In this case, the two or more ECG data may include at least one ECG data among overlapping ECG data and non-overlapping ECG data.

The processor 483 may determine the selected ECG data as learning data (S1033d). In this case, the number of selected ECG data may be equal to the number of ECG data.

FIG. 15 shows an example for explaining the operation of determining candidate learning data and the operation of determining learning data shown in FIG. 13C.

The processor 483 may classify a plurality of electrocardiogram derivation methods and a plurality of electrocardiogram data by object (S1035a).

For example, the processor 483 may classify a plurality of electrocardiogram derivation schemes and a plurality of electrocardiogram data by object based on first identifier information and second identifier information labeled with each of the plurality of electrocardiogram data.

The processor 483 may determine one ECG derivation method among the ECG derivation methods corresponding to each object as a representative ECG derivation method representing each object based on the number of the plurality of ECG derivation methods and the number of the plurality of objects. Yes (1035b).

For example, the processor 483 may determine a representative ECG derivation method representing each object based on whether the number of the plurality of ECG derivation methods and the number of the plurality of objects are the same.

When the number of the plurality of ECG derivation methods and the number of the plurality of objects are the same, the processor 483 may determine a representative ECG derivation method representing each object as a different ECG derivation method for each object. In other words, a representative electrocardiogram induction method representing each object may be different for each object.

When the number of the plurality of ECG derivation methods and the number of the plurality of objects are not the same, the processor 483 selects a representative ECG derivation method representing each object to induce an ECG having the largest number of ECG data among the ECG derivation methods corresponding to each object. method can be determined.

For example, the ECG induction schemes corresponding to the first object are the Lead I scheme and the Lead II scheme, the ECG data corresponding to the first object are two Lead I ECG data and one Lead II ECG data, and the second object When the ECG induction methods corresponding to are the Lead I method and the Lead II method, and the ECG data corresponding to the second object is one Lead I ECG data and two Lead II ECG data, the processor 483 selects the first object. A representative electrocardiogram induction method representing the second object may be determined as the Lead I method, and a representative electrocardiogram induction method representing the second object may be determined as the Lead II method.

The processor 483 may determine ECG data corresponding to a representative ECG derivation method representing each object as representative ECG data representing each object from among the ECG data corresponding to each object (S1035c).

The electrocardiogram induction method corresponding to the first object is a Lead I method and a Lead II method, the electrocardiogram data corresponding to the first object is two Lead I electrocardiogram data and one Lead II electrocardiogram data, and the electrocardiogram corresponding to the second object When the induction methods are the Lead I method and the Lead II method, and the electrocardiogram data corresponding to the second object is one Lead I electrocardiogram data and two Lead II electrocardiogram data, the processor 483 performs a representative electrocardiogram representing the first object. The data may be determined as two pieces of Lead I electrocardiogram data, and representative electrocardiogram data representing the second object may be determined as one piece of Lead II electrocardiogram data.

The processor 483 may determine representative ECG data representing each object as candidate learning data (S1037).

When the representative ECG data representing the first object is two Lead I ECG data and the representative ECG data representing the second object is two Lead II ECG data, the processor 483 executes two Lead I ECG data representing the first object. Lead I ECG data and two Lead II ECG data representing the second object may be determined as candidate learning data.

The processor 483 may determine at least one candidate training data among the candidate training data as training data (S1039).

For example, the processor 483 may determine all of the candidate training data as training data (1039a).

As another example, the processor 483 may determine some of the candidate training data as training data (1039b). For example, the processor 483 may determine two or more candidate training data among the candidate training data as training data.

Specifically, the processor 483 may calculate the total number of candidate training data, which is the number of candidate training data. For example, the processor 483 may calculate the total number of candidate training data by counting each of the candidate training data.

The processor 483 may determine the number of candidate training data, that is, the number of candidate training data to be selected as training data from among the candidate training data, based on the total number of candidate training data.

For example, the processor 483 may determine the number of candidate training data based on whether the total number of candidate training data is odd or even. When the total number of candidate training data is an odd number, the processor 483 may determine an odd number from among half or more of the total number of candidate training data and the total number of candidate training data as the number of candidate training data. When the total number of candidate training data is an even number, the processor 483 may determine an even number between half or more of the total number of candidate training data and the total number of candidate training data as the number of candidate training data.

The processor 483 may select two or more candidate training data from among the candidate training data based on the number of candidate training data. In this case, the two or more candidate training data may include at least one candidate training data among overlapping candidate training data and non-overlapping candidate training data.

The processor 483 may determine the selected candidate training data as training data. In this case, the number of selected candidate training data may be equal to the number of candidate training data.

16 shows an example for explaining a filtering operation and a learning operation of the processor shown in FIG. 12, FIG. 17A shows an example for explaining original electrocardiogram data according to an embodiment of the present disclosure, and FIG. An example for explaining correction data according to an embodiment of the present disclosure is shown, FIG. 18A shows another example for explaining correction data according to an embodiment of the present disclosure, and FIG. 18B is an example for explaining correction data according to an embodiment of the present disclosure. Another example for explaining correction data according to is shown.

Referring to FIGS. 16 to 18B , the processor 483 may acquire the learning data of the sub-data unit extracted through step S1050 as training data of the deep learning model (S1071). In other words, the processor 483 may obtain training data of a deep learning model, which is training data of an extracted sub-data unit.

The processor 483 filters the training data based on at least one of first information about the electronic device of the object corresponding to the training data and second information about the motion of the object, thereby generating correction data from which the training data is filtered. It can be done (S1073). In this case, the learning data may be electrocardiogram data for any one object among a plurality of objects. At least one of the first information and the second information may be labeled as the learning data. As described above, the first information may include information on the first electromagnetic signal, which is an electromagnetic signal generated by an electronic device of an object while electrocardiogram data is measured. The second information is information on at least one electromagnetic signal from among a second electromagnetic signal that is an electromagnetic signal generated by muscle movement of an object and a third electromagnetic signal that is an electromagnetic signal generated by respiration of an object while electrocardiogram data is measured. can include

For example, the processor 483 may generate training data from which the high-frequency noise is removed by removing high-frequency noise from the training data based on at least one electromagnetic signal from among the first electromagnetic signal and the second electromagnetic signal.

Specifically, the processor 483 may determine at least one electromagnetic signal among the first electromagnetic signal and the second electromagnetic signal as the high frequency noise (or high frequency noise signal). The processor 483 may generate training data from which high-frequency noise is removed by removing a frequency corresponding to a frequency of high-frequency noise from among a plurality of frequencies included in the training data. In this case, the frequency of the high-frequency noise may be at least one of the frequencies of the first electromagnetic signal and the frequency of the second electromagnetic signal.

When the training data is the same as the original ECG data shown in FIG. 17A, the processor 483 may generate Pre-filtered ECG data 1 shown in FIG. 1B by removing high-frequency noise included in the original ECG data. Pre-filtered ECG data 1 may be training data from which high-frequency noise is removed.

The processor 483 may generate training data from which low-frequency noise has been removed by removing low-frequency noise from the training data from which high-frequency noise has been removed based on the third electromagnetic signal.

For example, the processor 483 may determine the third electromagnetic signal as low-frequency noise (or low-frequency noise signal). The processor 483 may generate training data from which high-frequency noise and low-frequency noise are removed by removing a frequency corresponding to a frequency of low-frequency noise from among a plurality of frequencies included in the training data from which high-frequency noise is removed. In this case, the frequency of the low-frequency noise may be the frequency of the third electromagnetic signal.

When training data from which high-frequency noise has been removed is the same as Pre-filtered ECG data 2 shown in FIG. 18A, the processor 483 removes low-frequency noise included in Pre-filtered ECG data 2 to post-filter ECG data can be generated. Post-filtered ECG data may be training data from which high-frequency noise and low-frequency noise are removed.

The processor 483 may determine training data from which high-frequency noise and low-frequency noise are removed as correction data.

The processor 483 may learn a deep learning model based on the training data and correction data (S1075).

For example, the processor 483 may learn the deep learning model by simultaneously inputting training data and correction data to the deep learning model.

As another example, the processor 483 may simultaneously input training data and recalibration data to the deep learning model or simultaneously input training data, correction data, and recalibration data to the deep learning model to learn the deep learning model.

For example, the processor 483 may generate recalibration data obtained by filtering the correction data by filtering the correction data based on the third information about the body of the object corresponding to the learning data. In this case, the third information may be labeled as the learning data. As described above, the third information may include information about the object's gender, age, and obesity.

Specifically, the processor 483 may determine the normal heart rate of the object corresponding to the learning data based on the third information.

[Table 1]

For example, as shown in Table 1, the processor 483 may obtain a normal heart rate range corresponding to the third information from among previously stored normal heart rate ranges. In this case, as shown in Table 1, the pre-stored normal heart rate ranges may be variously determined according to at least one of gender, age, and obesity. The normal heart rate range corresponding to the third information may be a normal heart rate range corresponding to at least one of gender, age, and obesity of the object among the normal heart rate ranges.

When the age of the subject is 5 years old, the normal heart rate range of the subject who is 5 years old may be 75 to 120 beats as shown in Table 1.

The processor 483 may determine an average heart rate, which is an average value of the acquired normal heart rate range, as the normal heart rate of the object.

The processor 483 may select a frequency corresponding to the normal heart rate of the object from among a plurality of frequencies included in the correction data. For example, the processor 483 may select a frequency identical to the normal heart rate or a frequency most similar to the normal heart rate as a frequency corresponding to the normal heart rate from among a plurality of frequencies included in the correction data.

The processor 483 may remove a frequency included in the critical frequency interval based on a frequency corresponding to a normal heart rate of the object from among a plurality of frequencies included in the correction data. In this case, the critical frequency interval may be preset and stored in the memory 481 . For example, the threshold frequency interval may be ±5 Hz based on a frequency corresponding to a normal heart rate of the object, but is not limited thereto. For example, the threshold frequency interval may be various, such as ±10 Hz or ±15 Hz based on a frequency corresponding to a normal heart rate of the object.

When the plurality of frequencies are 50 Hz to 70 Hz, the frequency corresponding to the normal heart rate of the object is 60 Hz, and the critical frequency interval is ±5 Hz, the processor 483 may remove 55 Hz to 65 Hz from 50 Hz to 70 Hz. .

The processor 483 may determine training data from which frequencies included in the critical frequency interval are removed as recalibration data. In this case, the recalibration data may be learning data that more clearly represents the normal heart rate of the object than the learning data and the correction data.

The processor 483 may simultaneously input training data and recalibration data to the deep learning model to learn the deep learning model, or simultaneously input training data, correction data, and recalibration data to the deep learning model to learn the deep learning model. .

19 illustrates an example for describing a health state diagnosis operation of a processor according to an embodiment of the present disclosure.

The processor 483 may acquire electrocardiogram data of the target object provided from the data measurement device 470 (S1910). For example, the processor 483 may obtain ECG data of a target object measured through two or more ECG derivation methods among a plurality of ECG derivation methods. In this case, each of the ECG data for the target object may be ECG data corresponding to each of at least two ECG induction methods.

The processor 483 inputs each of the electrocardiogram data of the target object into each of the analysis models generated through the learned deep learning model, and converts each of the electrocardiogram data of the target object into normal electrocardiogram data corresponding to a healthy object and unhealthy It may be determined as any one ECG data among abnormal ECG data corresponding to the object (S1930). In this case, the analysis model may be a deep learning model that analyzes electrocardiogram data. A healthy object may be an object that does not have a heart disease and has normal levels of anemia and minerals. The non-healthy object may be an object that has a possible heart disease or has an abnormal level of at least one of anemia level and mineral level.

For example, the processor 483 may generate analysis models based on the learned deep learning model and the number of electrocardiogram data for the target object (S1931). In this case, each of the analysis models may be the same deep learning model as the learned deep learning model. The number of analysis models may be the same as the number of electrocardiogram data of the target object. In other words, the analysis model may be generated as many times as the ECG data for the target object with the same model as the learned deep learning model.

The processor 483 may input each of the electrocardiogram data of the target object to each of the analysis models (S1933).

The processor 483 may determine each of the electrocardiogram data of the target object as any one of normal electrocardiogram data and abnormal electrocardiogram data based on analysis result values output from the analysis models into which the electrocardiogram data is input ( S1935). At this time, each of the electrocardiogram data for the target object may be analyzed through each analysis model. Each analysis model may analyze electrocardiogram data input to each analysis model and output an analysis result value. In other words, each of the analysis result values may be output from each of the analysis models. Each of the analysis result values may be any one value from 0 to 1. The analysis result value may mean a probability of developing a disease. As described above, the disease may be a disease that can occur in the heart. When the analysis result value is close to 1, the probability of developing the disease is low, and when the analysis result value is close to 0, the probability of developing the disease may be high, but is not limited thereto. For example, when the analysis result value is close to 1, the probability of developing the disease may be high, and when the analysis result value is close to 0, the probability of developing the disease may be low. Hereinafter, for convenience of description, it will be assumed that when the analysis result value is close to 1, the probability of developing a disease is low, and when the analysis result value is close to 0, the probability of developing the disease is high.

For example, the processor 483 obtains the rest of the analysis result values except for the analysis result value included in the critical interval among the analysis result values by deleting the analysis result value included in the critical interval based on 0.5 from among the analysis result values. can do. In this case, the critical interval may be preset and stored in the memory 481 . The critical interval may be ±0.05 based on 0.5, but is not limited thereto. For example, the critical interval may be various, such as ±0.1 or ±0.2 based on 0.5.

The processor 483 may determine electrocardiogram data corresponding to the remaining analysis result values among the electrocardiogram data for the target object as any one of the normal electrocardiogram data and the abnormal electrocardiogram data based on the remaining analysis result values. . In this case, electrocardiogram data corresponding to the remaining analysis result values may be electrocardiogram data input to analysis models outputting the remaining analysis result values.

For example, the processor 483 may determine ECG data whose analysis result value does not exceed 0.5 among ECG data corresponding to the remaining analysis result values as abnormal ECG data. The processor 483 may determine electrocardiogram data whose analysis result value exceeds 0.5 among electrocardiogram data corresponding to the remaining analysis result values as normal electrocardiogram data.

As described above, the processor 483 determines ECG data exceeding 0.5 as normal ECG data and ECG data not exceeding 0.5 as abnormal ECG data, but is not limited thereto. For example, the processor 483 may determine ECG data exceeding 0.5 as abnormal ECG data and ECG data not exceeding 0.5 as normal ECG data according to an algorithm modeling the deep learning model.

The processor 483 may diagnose (or determine) the health state of the target object based on the number of ECG data determined as normal ECG data and the number of ECG data determined as abnormal ECG data among ECG data of the target object ( S1950).

For example, the processor 483 sets the health state of the target object to one of a normal state and an abnormal state based on whether the number of ECG data determined as normal ECG data is higher than the number of ECG data determined as abnormal ECG data. can decide In this case, the normal state is a state corresponding to a healthy object, and may be a state in which anemia and mineral levels are normal without having a disease. The abnormal state is a state corresponding to a non-healthy subject, and may be a state in which a disease is possessed (or possessed) and at least one of the anemia level and the mineral level is abnormal.

When the number of ECG data determined as normal ECG data is higher than the number of ECG data determined as abnormal ECG data, the processor 483 may determine the target object as a healthy object and determine the health state of the target object as a normal state.

If the number of ECG data determined as normal ECG data among the ECG data for the target object is lower than or equal to the number of ECG data determined as abnormal ECG data, the processor 483 determines the target object as an unhealthy object, and A health condition can be determined as abnormal.

The processor 483 may provide information (or health information) on the health state of the target object to the target object and/or a manager managing the target object (S1970). For example, the processor 483 may transmit information about the health state of the target object to an electronic device used by the target object and/or a manager. At this time, the information about the health state of the target object is the heart health of the target object, such as information indicating whether the target object is a healthy object or non-healthy object and/or information indicating whether the health state of the target object is normal or abnormal. Various information about may be included.

Through this, the processor 483 may allow the target object and/or manager to manage the heart of the target object in a healthy manner.

The processor 483 is implemented to be distinct from the control unit 1200 of the massage device 100, but is not limited thereto. For example, the control unit 1200 may be implemented to include the processor 483 therein or may be implemented to perform the above-described operation of the processor 483 to perform the operation of the processor 483 described above.

Those skilled in the art will appreciate that the present invention may be implemented in combination with other program modules and/or as a combination of hardware and software. For example, the present invention may be implemented by a computer-readable medium.

Computer readable media can be any medium that can be accessed by a computer, including volatile and nonvolatile media, transitory and non-transitory media, removable and non-transitory media. Includes removable media. By way of example, and not limitation, computer readable media may include computer readable storage media and computer readable transmission media.

Computer readable storage media includes volatile and nonvolatile media, transitory and non-transitory, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. includes media Computer readable storage media may include RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital video disk (DVD) or other optical disk storage device, magnetic cassette, magnetic tape, magnetic disk storage device or other magnetic storage device. device, or any other medium that can be accessed by a computer and used to store desired information.

A computer readable transmission medium typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism. Including all information delivery media. The term modulated data signal means a signal that has one or more of its characteristics set or changed so as to encode information within the signal. By way of example, and not limitation, computer readable transmission media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, and other wireless media. Combinations of any of the above are also intended to be included within the scope of computer readable transmission media.

Those skilled in the art will understand that the various illustrative logical blocks, modules, processors, means, circuits, and algorithm steps described in connection with the embodiments disclosed herein are electronic hardware, (for convenience) , may be implemented by various forms of program or design code (referred to herein as “software”) or a combination of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the particular application and the design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

Various embodiments presented herein may be implemented as a method, apparatus, or article of manufacture using standard programming and/or engineering techniques. The term "article of manufacture" includes a computer program, carrier, or media accessible from any computer-readable device. For example, computer-readable storage media include magnetic storage devices (eg, hard disks, floppy disks, magnetic strips, etc.), optical disks (eg, CDs, DVDs, etc.), smart cards, and flash memory devices (eg, EEPROM, cards, sticks, key drives, etc.), but are not limited thereto. The term “machine-readable medium” includes, but is not limited to, wireless channels and various other media that can store, hold, and/or convey instruction(s) and/or data.

It is to be understood that the specific order or hierarchy of steps in the processes presented is an example of example approaches. Based upon design priorities, it is to be understood that the specific order or hierarchy of steps in the processes may be rearranged within the scope of the present invention. The accompanying method claims present elements of the various steps in a sample order, but are not meant to be limited to the specific order or hierarchy presented.

The description of the presented embodiments is provided to enable any person skilled in the art to use or practice the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments without departing from the scope of the present invention. Thus, the present invention is not to be limited to the embodiments presented herein, but is to be construed in the widest scope consistent with the principles and novel features presented herein.

Claims (7)

A method for learning a deep learning model, performed by a processor of a health diagnosis device,
acquiring training data of the deep learning model;
Filtering the learning data based on at least one of first information about an electronic device of an object corresponding to the learning data and second information about a motion of the object to generate correction data filtered from the learning data step;
generating recalibration data in which the correction data is filtered by filtering the correction data based on third information about the body of the object; and
learning the deep learning model by simultaneously inputting the training data and the recalibration data to the deep learning model;
including,
method. According to claim 1,
The learning data is electrocardiogram data for the object,
The learning data is data labeled with at least one of the first information to the third information,
The first information includes information on a first electromagnetic signal, which is an electromagnetic signal generated by the electronic device while electrocardiogram data is measured,
The second information is selected from among a second electromagnetic signal, which is an electromagnetic signal generated by muscle movement of the object while electrocardiogram data is being measured, and a third electromagnetic signal, which is an electromagnetic signal generated by respiration of the object while electrocardiogram data is measured. contains information about at least one electromagnetic signal;
The third information includes information on at least one of information on the sex, age, and obesity of the object,
The electronic device is an electronic device in contact with the object,
method. According to claim 2,
The step of generating the correction data is:
generating training data from which the high-frequency noise is removed by removing high-frequency noise from the training data based on at least one of the first electromagnetic signal and the second electromagnetic signal;
generating training data from which the low-frequency noise has been removed by removing low-frequency noise from the training data from which the high-frequency noise has been removed, based on the third electromagnetic signal; and
determining training data from which the high frequency noise and the low frequency noise are removed as the correction data;
including,
method. According to claim 3,
The step of removing the high-frequency noise is:
determining at least one electromagnetic signal from among the first electromagnetic signal and the second electromagnetic signal as the high-frequency noise; and
generating learning data from which the high-frequency noise is removed by removing a frequency corresponding to the frequency of the high-frequency noise from among a plurality of frequencies included in the learning data;
including,
The step of removing the low-frequency noise is:
determining the third electromagnetic signal as the low frequency noise; and
generating training data from which the high-frequency noise and the low-frequency noise are removed by removing a frequency corresponding to the frequency of the low-frequency noise from among a plurality of frequencies included in the training data from which the high-frequency noise has been removed;
including,
The frequency of the high-frequency noise is at least one of the frequencies of the first electromagnetic signal and the frequency of the second electromagnetic signal;
The frequency of the low-frequency noise is the frequency of the third electromagnetic signal,
method. According to claim 2,
The step of generating the recalibration data is:
determining a normal heart rate of the object based on the third information;
selecting a frequency corresponding to the normal heart rate from among a plurality of frequencies included in the correction data;
removing a frequency included in a critical frequency interval based on a frequency corresponding to the normal heart rate from among a plurality of frequencies included in the correction data; and
determining correction data from which frequencies included in the critical frequency interval are removed as the re-correction data;
including,
method. According to claim 5,
Determining the subject's normal heart rate comprises:
obtaining a normal heart rate range corresponding to the third information from among previously stored normal heart rate ranges; and
determining an average heart rate, which is an average value of the acquired normal heart rate range, as the normal heart rate;
including,
method. According to claim 5,
The step of selecting a frequency corresponding to the normal heart rate is:
selecting a frequency identical to the normal heart rate or a frequency most similar to the normal heart rate as a frequency corresponding to the normal heart rate from among a plurality of frequencies included in the correction data;
including,
method.

Images