KR20230139859A - Method and system for predicting vertigo-associated disease - Google Patents

Abstract

In the dizziness-related disease prediction method, a gait feature generator generates a plurality of gait features for each of a plurality of patients using measurement values measured through an IMU sensor attached to the patient while the patient walks, and a first preprocessor generates a plurality of gait features. Generate an input vector including a plurality of gait features of the patient and a plurality of bio features of the patient, determine the disease that the patient has among the first to nth diseases related to dizziness as a label for the input vector, The machine learning module generates a disease prediction model by training the artificial neural network to output the disease that the patient has among the first to nth diseases based on the patient's input vector, and the gait feature generator generates a disease prediction model for the target patient. A plurality of target patient gait features are generated, a second preprocessor generates a target patient input vector including a plurality of target patient gait features and a plurality of target patient bio features of the target patient, and the machine learning module generates a disease prediction model and a target patient input vector. Predict diseases related to dizziness in the target patient based on the target patient input vector.

Description

Dizziness-related disease prediction method and system {METHOD AND SYSTEM FOR PREDICTING VERTIGO-ASSOCIATED DISEASE}

The present invention relates to a method and system for predicting diseases related to dizziness, and more specifically, to a method for predicting diseases related to dizziness in a patient based on the patient's gait parameters and physical information. and systems.

Dizziness refers to an abnormal sensation that occurs when there is a problem with the sense of space or balance, making it feel as if the surrounding environment or oneself is moving even though there is no movement.

Dizziness can be classified into central dizziness and peripheral dizziness. Central dizziness refers to dizziness caused by abnormalities in structures from the central nervous system to the cerebellum, brainstem, and vestibular nerve nuclei, while peripheral dizziness occurs in the vestibular nerve or labyrinth. It refers to dizziness caused by: In this way, dizziness can be caused by various causes in various parts of the body.

In general, various vestibular function tests such as nystagmus test, vestibular evoked muscle potential test, dynamic posture test, rotating chair test, head impulse test, and posture test are used to determine the cause of dizziness.

However, as described above, dizziness can be caused by various causes in each part of the body, so there is a problem that it is difficult to accurately determine the cause of dizziness by evaluating the vestibular system alone.

In addition, conventionally, the cause of dizziness is determined according to the subjective standards of the doctor performing the test without a quantitative standard for determining the cause of dizziness, so there is a limit to accurately diagnosing the type of disease that causes dizziness.

One purpose of the present invention is to provide a method for predicting diseases related to dizziness in patients based on artificial intelligence (AI).

Another object of the present invention is to provide a system that can predict diseases related to dizziness in patients based on artificial intelligence.

In order to achieve the above-described object of the present invention, in the method for predicting dizziness-related diseases according to an embodiment of the present invention, the walking feature generator is an IMU attached to the patient while the patient is walking for each of a plurality of patients. (Inertial Measurement Unit) Generates a plurality of gait features related to the patient's gait using measurement values measured through a sensor, and a first preprocessor generates a plurality of gait features of the patient for each of the plurality of patients. and generate an input vector including a plurality of bio features corresponding to the patient's physical information, and determine which of the first to nth (n is a positive integer of 2 or more) diseases related to dizziness that the patient has. A disease is determined as a label for the input vector, and a machine learning module uses the input vectors for the plurality of patients and the pairs of the labels for the input vectors as learning data to generate an artificial neural network using the input vectors for the plurality of patients. Based on the vector, the artificial neural network is trained to output a disease that the patient has among the first to nth diseases to generate a disease prediction model, and the walking feature generator generates a disease prediction model for the target patient while the target patient walks. A plurality of target patient gait features related to the gait of the target patient are generated using measurement values measured through the attached IMU sensor, and a second preprocessor generates the plurality of target patient gait features and the target patient's body. A target patient input vector including a plurality of target patient bio features corresponding to the information is generated, and the machine learning module determines a disease related to dizziness of the target patient based on the disease prediction model and the target patient input vector. predict

In one embodiment, the IMU sensor may include a first IMU sensor attached to the left shoe worn by the patient when walking and a second IMU sensor attached to the right shoe worn by the patient when walking. .

In one embodiment, the IMU sensor may include a 3-axis acceleration sensor, a 3-axis gyro sensor, and a 3-axis geomagnetic sensor.

In one embodiment, the plurality of gait features include a left foot inversion-eversion range of motion, which represents the range of angles in which the left ankle joint moves in and out of the body, and a left foot inversion-eversion range of motion, which represents the range of angles in which the right foot ankle joint moves in and out of the body. It may include the right foot internal/external joint range of motion, which indicates the range of movement angles.

In one embodiment, the plurality of gait features may include a gait asymmetry (GA) index that represents the ratio of the swing time of the left foot and the swing time of the right foot when walking.

In one embodiment, the plurality of walking features include a Phase Coordination Index (PCI) corresponding to a variable combining the variability and asymmetry of left-foot gait and right-foot gait during walking. can do.

In one embodiment, the plurality of gait features may include steps per minute (cadence).

In one embodiment, the plurality of gait features are left dorsi-flexion, which represents the range of angles in which the left ankle joint moves with the instep toward the lower leg and the instep toward and away from the lower leg. It may include the plantar-flexion range of motion and the right foot dorsiflexion-plantarflexion range of motion, which represents the range of angles in which the right ankle joint moves in the direction in which the instep approaches the lower leg and in the direction in which the instep moves away from the lower leg. .

In one embodiment, the plurality of bio features may include a height.

In one embodiment, the first to nth diseases that are labels for the input vector may include at least one disease related to the vestibular system and at least one disease unrelated to the vestibular system. .

In one embodiment, the first to nth diseases that are labels for the input vector may include musculoskeletal disorders.

In one embodiment, the step of generating the plurality of walking features by the walking feature generator uses measurement values measured through the IMU sensor while the patient walks at a preferred speed for each of the plurality of patients. generating a plurality of first gait features, for each of the plurality of patients, using measurements measured through the IMU sensor while the patient is walking at a speed slower than the preferred speed. Generating, for each of the plurality of patients, a second plurality of gait features corresponding to gait features, using measurements taken through the IMU sensor while the patient is walking at a speed faster than the preferred speed. generating a plurality of third gait features corresponding to the plurality of first gait features, and generating the plurality of first gait features, the plurality of second gait features, and the plurality of third gait features. It may include outputting the plurality of walking features, including all of them.

In one embodiment, the first preprocessor generates the input vector including the plurality of gait features of the patient and the plurality of bio features of the patient for each of the plurality of patients, and the first preprocessor The step of determining the disease that the patient has among the to nth diseases as the label for the input vector includes the first preprocessor selecting the plurality of gait features of the patient for each of the plurality of patients, the Generate the input vector including the plurality of bio features of a patient and a plurality of nystagmus test result features corresponding to results of a plurality of nystagmus tests performed on the patient, and It may include determining the disease that the patient has among n diseases using the label for the input vector.

The plurality of nystagmus test result features may include a ratio between the size of the left tonic deviation and the size of the right tonic deviation measured by performing a caloric test through a video nystagmometer.

The plurality of nystagmus test result features may include the presence or absence of spontaneous nystagmus, which is measured by performing a spontaneous nystagmus test.

In order to achieve the above-described object of the present invention, the dizziness-related disease prediction system according to an embodiment of the present invention includes a walking feature generator, a first preprocessor, a second preprocessor, and a machine learning module. The gait feature generator generates, for each of a plurality of patients, a plurality of gait features related to the gait of the patient using measurement values measured through an IMU (Inertial Measurement Unit) sensor attached to the patient while the patient is walking. and generate a plurality of target patient gait features related to the target patient's gait using measurements measured through the IMU sensor attached to the target patient while the target patient walks. The first preprocessor generates, for each of the plurality of patients, an input vector including the plurality of gait features of the patient and a plurality of bio features corresponding to the body information of the patient, and a first preprocessor related to dizziness. The disease that the patient has among the 1st to nth (n is a positive integer of 2 or more) diseases is determined as the label for the input vector. The second preprocessor generates a target patient input vector including a plurality of target patient bio features corresponding to the plurality of target patient gait features and body information of the target patient. The machine learning module uses the input vectors for the plurality of patients and the pairs of labels for the input vectors as learning data to determine the first to nth diseases based on the input vectors of the patients. Among these, a disease prediction model is generated by training the artificial neural network to output the disease that the patient has, and the disease related to the target patient's dizziness is predicted based on the disease prediction model and the target patient input vector.

The dizziness-related disease prediction system and dizziness-related disease prediction method according to embodiments of the present invention are based on artificial intelligence (AI) and use only the patient's walking factors and basic physical information to predict the dizziness that the patient has. Diseases can be accurately predicted.

1 is a diagram showing a dizziness-related disease prediction system according to an embodiment of the present invention.
Figure 2 is a flowchart showing a method for predicting dizziness-related diseases according to an embodiment of the present invention.
FIG. 3 is a diagram illustrating an example of an artificial neural network included in the machine learning module of FIG. 1.
Figure 4 is a diagram showing a dizziness-related disease prediction system according to another embodiment of the present invention.
Figure 5 is a flowchart showing a method for predicting dizziness-related diseases according to another embodiment of the present invention.

Regarding the embodiments of the present invention disclosed in the text, specific structural and functional descriptions are merely illustrative for the purpose of explaining the embodiments of the present invention, and the embodiments of the present invention may be implemented in various forms. It should not be construed as limited to the embodiments described in.

Since the present invention can be subject to various changes and have various forms, specific embodiments will be illustrated in the drawings and described in detail in the text. However, this is not intended to limit the present invention to a specific disclosed form, and should be understood to include all changes, equivalents, and substitutes included in the spirit and technical scope of the present invention.

Terms such as first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The above terms may be used for the purpose of distinguishing one component from another component. For example, a first component may be referred to as a second component, and similarly, the second component may be referred to as a first component without departing from the scope of the present invention.

When a component is said to be "connected" or "connected" to another component, it is understood that it may be directly connected to or connected to the other component, but that other components may exist in between. It should be. On the other hand, when it is mentioned that a component is “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between. Other expressions that describe the relationship between components, such as "between" and "immediately between" or "neighboring" and "directly adjacent to" should be interpreted similarly.

The terms used in this application are only used to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, terms such as “comprise” or “have” are intended to designate the existence of a described feature, number, step, operation, component, part, or combination thereof, but are not intended to indicate the presence of one or more other features or numbers. It should be understood that this does not preclude the existence or addition of steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the technical field to which the present invention pertains. 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 technology, and unless clearly defined in the present application, should not be interpreted as having an ideal or excessively formal meaning. .

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. The same reference numerals are used for the same components in the drawings, and duplicate descriptions for the same components are omitted.

1 is a diagram showing a dizziness-related disease prediction system according to an embodiment of the present invention.

The dizziness-related disease prediction system 10 shown in FIG. 1 uses gait parameters measured from each of a plurality of patients with a disease related to dizziness and physical information of each of the plurality of patients as learning data. After creating a disease prediction model that predicts the disease related to the dizziness of the patient by learning an artificial neural network, the disease related to the dizziness of the target patient is used using the disease prediction model. predict.

Figure 2 is a flowchart showing a method for predicting dizziness-related diseases according to an embodiment of the present invention.

The dizziness-related disease prediction method shown in FIG. 2 can be performed through the dizziness-related disease prediction system 10 of FIG. 1 .

Hereinafter, with reference to FIGS. 1 and 2, the configuration and operation of the dizziness-related disease prediction system 10 and the dizziness-related disease prediction method performed by the dizziness-related disease prediction system 10 will be described in detail.

Referring to FIG. 1, the dizziness-related disease prediction system 10 includes an IMU (Inertial Measurement Unit) sensor 100, a walking feature generator 200, a body information database 300, a first preprocessor 400, and a machine. It includes a machine learning module 500 and a second preprocessor 600.

The IMU sensor 100 is attached to each of the plurality of patients while the patient walks and generates various types of measurement values while the patient walks.

In one embodiment, the IMU sensor 100 includes a first IMU sensor (L_IMU) 110 attached to the left shoe worn by the patient when walking and a second IMU sensor (L_IMU) 110 attached to the right shoe worn by the patient when walking. 2 It may include an IMU sensor (R_IMU) (120).

In one embodiment, the first IMU sensor 110 and the second IMU sensor 120 each include a three-axis acceleration sensor that measures acceleration in the x-axis, y-axis, and z-axis directions, and an x-axis, y-axis, and It may include a 3-axis gyro sensor that measures the angular velocity in the z-axis direction.

According to the embodiment, in order to compensate for errors in the 3-axis acceleration sensor and the 3-axis gyro sensor, each of the first IMU sensor 110 and the second IMU sensor 120 further includes a 3-axis geomagnetic sensor (magnetometer). can do.

In this case, the first IMU sensor 110 transmits 9 measurement values representing the movement of the patient's left foot when the patient walks through the 3-axis acceleration sensor, the 3-axis gyro sensor, and the 3-axis geomagnetic sensor. The second IMU sensor 120 generates 9 measurement values representing the movement of the patient's right foot when the patient walks through the 3-axis acceleration sensor, the 3-axis gyro sensor, and the 3-axis geomagnetic sensor. can be created.

For each of the plurality of patients, the walking feature generator 200 uses measurement values measured through the IMU sensor 100 attached to the patient while the patient walks a certain section to determine the patient's gait and A plurality of related walking factors may be calculated, and the plurality of walking factors may be generated as a plurality of walking features (G_Fs) (step S100).

In one embodiment, the plurality of gait features (G_Fs) generated by the gait feature generator 200 are the left foot inversion-eversion joint, which represents the range of angles in which the left foot ankle joint moves inward and outward of the body. It may include range of motion and right foot internal and eversion joint range of motion, which indicates the range of angles in which the right ankle joint moves inward and outward of the body.

In one embodiment, the plurality of gait features (G_Fs) generated from the gait feature generator 200 are a gait imbalance index ( Gait Asymmetry (GA) may be included.

In one embodiment, a plurality of gait features (G_Fs) generated by the gait feature generator 200 correspond to variables combining the variability and asymmetry of left-foot gait and right-foot gait when walking. It may include the Phase Coordination Index (PCI).

In one embodiment, the plurality of walking features (G_Fs) generated by the walking feature generator 200 may include the number of steps per minute (cadence).

In one embodiment, the plurality of walking features (G_Fs) generated by the walking feature generator 200 are the angles at which the left foot ankle joint moves in the direction in which the instep approaches the lower leg and in the direction in which the instep moves away from the lower leg. The left foot dorsi-flexion-plantar-flexion joint range of motion, which indicates the range of motion, and the right foot, which indicates the range of angles in which the right foot ankle joint moves in the direction in which the instep approaches the lower leg and in the direction in which the instep moves away from the lower leg. May include dorsiflexion-plantarflexion joint range of motion.

Depending on the embodiment, the plurality of walking features (G_Fs) generated from the walking feature generator 200 include walking speed, left foot stride length, right foot stride length, left foot step length, and right foot step length. , left foot single limb support, right foot single limb support, left foot double limb support, right foot double limb support, left foot toe push time (stance phase), and right foot toe push time. More may be included.

In general, methods for calculating various types of gait parameters using measurements generated through an IMU sensor are widely known, so here, the gait feature generator 200 uses the IMU sensor 100 attached to the patient. A detailed description of the method for calculating each of the plurality of gait features (G_Fs) related to the patient's gait using the measured values will be omitted.

The physical information database 300 stores various types of physical information for each of the plurality of patients.

In one embodiment, the physical information database 300 may store height, gender, weight, age, body mass index (BMI), etc. for each of the plurality of patients.

Additionally, the physical information database 300 may store diseases related to dizziness that each of the plurality of patients has.

The first preprocessor 400 may read the body information of each of the plurality of patients stored in the body information database 300 as a plurality of bio features (B_Fs).

In one embodiment, the plurality of bio features (B_Fs) may include the patient's height.

Depending on the embodiment, the plurality of bio features (B_Fs) may further include the patient's gender, weight, age, and BMI value.

Additionally, the first preprocessor 400 may read the disease (DS) related to dizziness that each of the plurality of patients has stored in the body information database 300.

Thereafter, the first preprocessor 400 generates an input vector (IN_VEC) including a plurality of gait features (G_Fs) of the patient and a plurality of bio features (B_Fs) of the patient for each of the plurality of patients. Generates a label (LB) for the disease (DS) that the patient has among the first to nth (n is a positive integer of 2 or more) diseases (DS1 to DSn) related to dizziness to the input vector (IN_VEC). ) can be determined (step S200).

In one embodiment, the first to nth diseases DS1 to DSn may include at least one disease related to the vestibular system and at least one disease unrelated to the vestibular system.

For example, the first to n diseases (DS1 to DSn) include dizziness, Benign Paroxysmal Positional Vertigo (BPPV), Meniere's disease, vestibular neuritis, It may include Sudden Sensory Neural Hearing Loss with Vertigo (SSNHLV), vestibular migraine, vestibular dysfunction, musculoskeletal disorder, and central dizziness.

In one embodiment, the first preprocessor 400 generates a plurality of walking features (G_Fs) and After scaling is performed on a plurality of bio features (B_Fs), an input vector (IN_VEC) can be generated.

For example, the first preprocessor 400 performs normalization on a plurality of walking features (G_Fs) and a plurality of bio features (B_Fs), and then generates a plurality of normalized walking features (G_Fs). and an input vector (IN_VEC) including a plurality of bio features (B_Fs).

Alternatively, the first preprocessor 400 performs standardization on a plurality of walking features (G_Fs) and a plurality of bio features (B_Fs), and then generates a plurality of standardized walking features (G_Fs) and a plurality of bio features (B_Fs). An input vector (IN_VEC) containing bio features (B_Fs) can be generated.

The pairs of input vectors (IN_VEC) and labels (LB) for the plurality of patients generated from the first preprocessor 400 may be provided to the machine learning module 500 as learning data.

The machine learning module 500 uses pairs of input vectors (IN_VEC) for the plurality of patients and labels (LB) for the input vectors (IN_VEC) as learning data, and the artificial neural network uses the input vectors (IN_VEC) of the patients. Based on this, a disease prediction model can be generated by training the artificial neural network to output the disease that the patient has among the first to nth diseases (DS1 to DSn) (step S300).

FIG. 3 is a diagram illustrating an example of an artificial neural network included in the machine learning module of FIG. 1.

In one embodiment, the artificial neural network may generate resultant data encoded using a one-hot encoding method.

In this case, as shown in FIG. 3, the artificial neural network includes an input layer (INPUT LAYER) including a number of input nodes equal to the number of a plurality of walking features (G_Fs) and a plurality of bio features (B_Fs), It may include at least one hidden layer (HIDDEN LAYER) including a plurality of nodes, and an output layer (OUTPUT LAYER) including n output nodes.

Each of the plurality of walking features (G_Fs) and the plurality of bio features (B_Fs) included in the input vector (IN_VEC) is input to each of the input nodes included in the input layer of the artificial neural network, and the artificial neural network One of the first to nth diseases (DS1 to DSn) may be indicated through values output from the output nodes included in the output layer.

In FIG. 3, the artificial neural network is exemplarily shown as including one hidden layer, but the present invention is not limited thereto, and depending on the embodiment, the artificial neural network may include a plurality of hidden layers.

The machine learning module 500 uses the artificial neural network to select the first to nth input vectors (IN_VEC) based on a plurality of gait features (G_Fs) and a plurality of bio features (B_Fs) included in the input vector (IN_VEC). The artificial neural network can be trained to classify among diseases (DS1 to DSn) into diseases corresponding to the label (LB) for the input vector (IN_VEC).

For example, if the label (LB) for the input vector (IN_VEC) corresponds to the ith (i is a positive integer less than n) disease (DSi) among the first to nth diseases (DS1 to DSn), The machine learning module 500 inputs a plurality of gait features (G_Fs) and a plurality of bio features (B_Fs) included in the input vector (IN_VEC) to the input nodes included in the input layer of the artificial neural network. In this case, the ith output node included in the output layer of the artificial neural network outputs 1 and the remaining output nodes output 0, so that the artificial neural network can be trained to classify the input vector (IN_VEC) as the ith disease (DSi). there is.

The machine learning module 500 completes the learning using pairs of input vectors (IN_VEC) and labels (LB) for the input vectors (IN_VEC) for the plurality of patients, and then runs the artificial neural network on which learning has been completed. It can be determined using the disease prediction model.

As described above, the first to nth diseases DS1 to DSn include at least one disease related to the vestibular system and at least one disease unrelated to the vestibular system, so the machine learning module 500 ) The disease prediction model generated by can predict not only diseases related to the patient's vestibular system but also diseases unrelated to the vestibular system.

Referring again to FIGS. 1 and 2, the dizziness-related disease prediction system 10 generates the disease prediction model through the first preprocessor 400 and the machine learning module 500, and then uses the disease prediction model to It is possible to predict diseases related to dizziness in the target patient being diagnosed.

Specifically, the gait feature generator 200 uses the IMU sensor 100 to generate a plurality of gait features (G_Fs) for each of the plurality of patients, using the same method to predict the disease. Calculate a plurality of gait factors related to the gait of the target patient using measurement values measured through the IMU sensor 100 attached to the target patient while the target patient walks a certain section, and the plurality of gait factors may be generated as a plurality of target patient gait features (TG_Fs) (step S400).

Accordingly, the plurality of gait features (G_Fs) generated for each of the plurality of patients and the plurality of target patient gait features (TG_Fs) generated for the target patient may correspond to each other on a one-to-one basis.

The second preprocessor 600 may receive a plurality of target patient gait features (TG_Fs) generated by the gait feature generator 200.

Additionally, the second preprocessor 600 may receive a plurality of target patient bio features (TB_Fs) corresponding to various types of body information about the target patient.

A plurality of bio features (B_Fs) for each of the plurality of patients and a plurality of target patient bio features (TB_Fs) for the target patient may correspond to each other on a one-to-one basis.

In Figure 1, the second preprocessor 600 is shown as receiving a plurality of target patient bio features (TB_Fs) for the target patient from outside the dizziness-related disease prediction system 10, but the present invention is limited to this. It doesn't work.

Depending on the embodiment, the physical information about the target patient is pre-stored in the physical information database 300, and the second preprocessor 600 stores the physical information about the target patient stored in the physical information database 300 to a plurality of targets. It can also be read as patient bio features (TB_Fs).

Thereafter, the second preprocessor 600 performs the first preprocessor 400 to generate a plurality of gait features (G_Fs) of the patient and a plurality of bio features (B_Fs) of the patient for each of the plurality of patients. A target patient including a plurality of target patient gait features (TG_Fs) of the target patient and a plurality of target patient bio features (TB_Fs) of the target patient in the same manner as the method of generating an input vector (IN_VEC) including a plurality of target patient gait features (TG_Fs) of the target patient. An input vector (TIN_VEC) can be generated (step S500).

In one embodiment, the second preprocessor 600 generates a plurality of target patient gait features (TG_Fs) and a plurality of target patient bio features (TB_Fs) to have the same size range. After scaling is performed on the features (TG_Fs) and a plurality of target patient bio features (TB_Fs), a target patient input vector (TIN_VEC) may be generated.

For example, the second preprocessor 600 performs normalization on a plurality of target patient gait features (TG_Fs) and a plurality of target patient bio features (TB_Fs), and then normalizes the plurality of target patient bio features (TB_Fs). A target patient input vector (TIN_VEC) including gait features (TG_Fs) and a plurality of target patient bio features (TB_Fs) may be generated.

Alternatively, the second preprocessor 600 performs standardization on a plurality of target patient gait features (TG_Fs) and a plurality of target patient bio features (TB_Fs), and then generates the standardized plurality of target patient gait features. A target patient input vector (TIN_VEC) including TG_Fs and a plurality of target patient bio features (TB_Fs) may be generated.

The second preprocessor 600 may provide the target patient input vector (TIN_VEC) generated for the target patient to the machine learning module 500.

The machine learning module 500 may predict a disease related to dizziness of the target patient based on the disease prediction model and the target patient input vector (TIN_VEC) (step S600).

In one embodiment, the machine learning module 500 uses the disease prediction model to classify the target patient input vector (TIN_VEC) as one of the first to nth diseases (DS1 to DSn) and classifies the classified disease as one of the first to nth diseases (DS1 to DSn). It can be predicted based on the disease the target patient has.

For example, the machine learning module 500 inputs a target patient input vector (TIN_VEC) into the disease prediction model and then determines the disease related to dizziness that the target patient has based on the results output from the disease prediction model. can be determined as one of the first to nth diseases (DS1 to DSn).

Meanwhile, in Figure 1, the first preprocessing unit 400 and the second preprocessing unit 600 are shown as separate components, but the present invention is not limited thereto.

Depending on the embodiment, the first preprocessor 400 and the second preprocessor 600 may be integrated and perform both the above-described operations of the first preprocessor 400 and the second preprocessor 600. there is.

As described above with reference to FIGS. 1 to 3, the dizziness-related disease prediction system 10 and the dizziness-related disease prediction method according to embodiments of the present invention are based on artificial intelligence (AI) and the patient's walking factors and Using only basic physical information, it is possible to accurately predict diseases related to the patient's dizziness, and not only diseases related to the vestibular system but also diseases unrelated to the vestibular system can be predicted.

In one embodiment, the dizziness-related disease prediction system 10 calculates a plurality of walking factors while each of the plurality of patients walks at various speeds, and comprehensively considers the plurality of walking factors calculated for each walking speed. Thus, the disease prediction model can be generated.

For example, the walking feature generator 200 uses, for each of the plurality of patients, measurement values measured through the IMU sensor 100 attached to the patient while the patient walks at a preferred speed. Thus, a plurality of gait factors related to the patient's gait may be calculated, and the plurality of gait factors may be generated as a plurality of first gait features.

Additionally, the walking feature generator 200 uses, for each of the plurality of patients, measurement values measured through the IMU sensor 100 attached to the patient while the patient walks at a speed slower than the preferred speed. Thus, a plurality of gait factors related to the patient's gait may be calculated, and the plurality of gait factors may be generated as a plurality of second gait features.

In addition, the walking feature generator 200 uses, for each of the plurality of patients, measurement values measured through the IMU sensor 100 attached to the patient while the patient walks at a speed faster than the preferred speed. Thus, a plurality of gait factors related to the patient's gait may be calculated, and the plurality of gait factors may be generated as a plurality of third gait features.

At this time, the plurality of first walking features, the plurality of second walking features, and the plurality of third walking features may be composed of the same type of walking factors. Accordingly, the plurality of first walking features, the plurality of second walking features, and the plurality of third walking features may correspond to each other on a one-to-one basis.

Thereafter, the walking feature generator 200 outputs a plurality of walking features (G_Fs) including all of the first plurality of walking features, the plurality of second walking features, and the plurality of third walking features. can do.

In this case, the gait feature generator 200 uses the IMU sensor 100 to generate a plurality of gait features (G_Fs) for each of the plurality of patients in the same manner as the target patient at various speeds. A plurality of walking factors related to the target patient's walking are calculated for each walking speed using measurement values measured through the IMU sensor 100 attached to the target patient while walking, and the plurality of walking factors calculated for each speed are calculated. The gait factors of can be generated as a plurality of target patient gait features (TG_Fs).

For example, the gait feature generator 200 uses measurements measured through the IMU sensor 100 attached to the target patient while the target patient walks at a preferred speed to determine the target patient's gait and A plurality of related gait factors may be calculated, and the plurality of gait factors may be generated into a plurality of first target patient gait features.

In addition, the walking feature generator 200 uses measurement values measured through the IMU sensor 100 attached to the target patient while the target patient walks at a speed slower than the preferred speed to determine the target patient's gait and A plurality of related gait factors may be calculated, and the plurality of gait factors may be generated into a plurality of second target patient gait features.

In addition, the gait feature generator 200 uses measurement values measured through the IMU sensor 100 attached to the target patient while the target patient is walking at a speed faster than the preferred speed to determine the target patient's gait and A plurality of related gait factors may be calculated, and the plurality of gait factors may be generated into a plurality of third target patient gait features.

At this time, the plurality of first target patient gait features, the plurality of second target patient gait features, and the plurality of third target patient gait features may be composed of the same type of Bohan factors. Accordingly, the plurality of first target patient gait features, the plurality of second target patient gait features, and the plurality of third target patient gait features may correspond to each other on a one-to-one basis.

Thereafter, the gait feature generator 200 generates a plurality of target patient gait features including all of the first plurality of target patient gait features, the plurality of second target patient gait features, and the plurality of third target patient gait features. Can be output as gait features (TG_Fs).

In this way, the dizziness-related disease prediction system 10 may generate the plurality of first gait features using measurements measured through the IMU sensor 100 attached to the patient while the patient walks at the preferred speed. , the plurality of second gait features generated using measurements taken through an IMU sensor 100 attached to the patient while the patient walks at a speed slower than the preferred speed, and Predict the disease by learning the artificial neural network using all of the plurality of third gait features generated using measurements measured through the IMU sensor 100 attached to the patient while walking at a faster than preferred speed. By creating a model, the accuracy of disease prediction can be further improved.

Figure 4 is a diagram showing a dizziness-related disease prediction system according to another embodiment of the present invention.

Referring to FIG. 4, the dizziness-related disease prediction system 20 includes an IMU (Inertial Measurement Unit) sensor 100, a walking feature generator 200, a body information database 300, a first preprocessor 400, and a machine. It includes a machine learning module 500, a second preprocessor 600, and a Nystagmus Test database 700.

The dizziness-related disease prediction system 20 shown in FIG. 4 may be configured to further include a nystagmus test database 700 in the dizziness-related disease prediction system 10 shown in FIG. 1 .

Figure 5 is a flowchart showing a method for predicting dizziness-related diseases according to another embodiment of the present invention.

The dizziness-related disease prediction method shown in FIG. 5 may be constructed by modifying the operations of some steps (S200 and S500) in the dizziness-related disease prediction method shown in FIG. 2.

The dizziness-related disease prediction method shown in FIG. 5 can be performed through the dizziness-related disease prediction system 20 of FIG. 4.

Included in the dizziness-related disease prediction system 20 of FIG. 4 are an IMU sensor 100, a walking feature generator 200, a body information database 300, a first preprocessor 400, a machine learning module 500, And the configuration and operation of the second preprocessor 600 include the IMU sensor 100, the walking feature generator 200, the body information database 300, and the first dizziness-related disease prediction system 10 of FIG. 1. The configuration and operation of the preprocessor 400, the machine learning module 500, and the second preprocessor 600 are the same.

Included in the dizziness-related disease prediction system 10 of FIG. 1 are an IMU sensor 100, a walking feature generator 200, a body information database 300, a first preprocessor 400, a machine learning module 500, And the configuration and operation of the second preprocessor 600 (steps S100, S200, S300, S400, S500, and S600) have been described in detail with reference to FIGS. 1 to 3, so here, the dizziness-related disease prediction system of FIG. 4 ( Redundant description of the configuration and operation of 20) will be omitted, and only the operation related to the nystagmus test database 700 included in the dizziness-related disease prediction system 20 will be described in detail.

The nystagmus test database 700 stores the results of a plurality of nystagmus tests performed on each of the plurality of patients.

In one embodiment, the nystagmus test database 700 may store the results of a caloric test, a spontaneous nystagmus test, etc. performed on each of the plurality of patients.

The first preprocessor 400 may read the results of the plurality of nystagmus tests performed on each of the plurality of patients stored in the nystagmus test database 700 as a plurality of nystagmus test result features (N_Fs).

In one embodiment, the plurality of nystagmus test result features (N_Fs) include the ratio between the size of the left tonic deviation and the size of the right tonic deviation measured by performing a thermonystagm test through a video nystagmometer. can do.

In one embodiment, the plurality of nystagmus test result features (N_Fs) may include the presence or absence of spontaneous nystagmus measured by performing a spontaneous nystagmus test.

Meanwhile, the first preprocessor 400 may read the body information of each of the plurality of patients stored in the body information database 300 as a plurality of bio features (B_Fs).

Additionally, the first preprocessor 400 may read the disease (DS) related to dizziness that each of the plurality of patients has stored in the body information database 300.

After the gait feature generator 200 generates a plurality of gait features (G_Fs) for each of the plurality of patients (step S100), the first preprocessor 400 generates a plurality of gait features (G_Fs) for each of the plurality of patients, Generate an input vector (IN_VEC) including a plurality of gait features (G_Fs) of the patient, a plurality of bio features (B_Fs) of the patient, and a plurality of nystagmus test result features (N_Fs) of the patient, Among the first to nth diseases (DS1 to DSn) related to dizziness, the disease (DS) that the patient has can be determined as the label (LB) for the input vector (IN_VEC) (step S210).

As described above with reference to FIGS. 1 to 3, the first to nth diseases (DS1 to DSn) include at least one disease related to the vestibular system and at least one disease unrelated to the vestibular system. can do.

For example, the first to n diseases (DS1 to DSn) include dizziness, Benign Paroxysmal Positional Vertigo (BPPV), Meniere's disease, vestibular neuritis, It may include Sudden Sensory Neural Hearing Loss with Vertigo (SSNHLV), vestibular migraine, vestibular dysfunction, musculoskeletal disorder, and central dizziness.

The pairs of input vectors (IN_VEC) and labels (LB) for the plurality of patients generated from the first preprocessor 400 may be provided to the machine learning module 500 as learning data.

The machine learning module 500 uses pairs of input vectors (IN_VEC) for the plurality of patients and labels (LB) for the input vectors (IN_VEC) as learning data, and the artificial neural network uses the input vectors (IN_VEC) of the patients. Based on this, a disease prediction model can be generated by training the artificial neural network to output the disease that the patient has among the first to nth diseases (DS1 to DSn) (step S300).

Meanwhile, as described above with reference to FIGS. 1 to 3, the gait feature generator 200 generates a plurality of target patient gait features (TG_Fs) for the target patient for which the disease is to be predicted, and generates the second preprocessor 600. ) can be provided (step S400).

In addition, the second preprocessor 600 receives a plurality of target patient bio features (TB_Fs) corresponding to various types of body information about the target patient, and selects a plurality of nystagmus tests performed on the target patient. A plurality of target patient nystagmus test result features (TN_Fs) corresponding to the results may be received.

A plurality of bio features (B_Fs) for each of the plurality of patients and a plurality of target patient bio features (TB_Fs) for the target patient may correspond to each other on a one-to-one basis.

Likewise, the plurality of nystagmus test result features (N_Fs) for each of the plurality of patients and the plurality of target patient nystagmus test result features (TN_Fs) for the target patient may correspond to each other on a one-to-one basis.

In FIG. 1, the second preprocessor 600 generates a plurality of target patient bio features (TB_Fs) and a plurality of target patient nystagmus test result features (TN_Fs) for the target patient from outside the dizziness-related disease prediction system 10. is shown as receiving, but the present invention is not limited thereto.

Depending on the embodiment, the physical information about the target patient is pre-stored in the physical information database 300, and the second preprocessor 600 stores the physical information about the target patient stored in the physical information database 300 to a plurality of targets. It can also be read as patient bio features (TB_Fs).

Likewise, the results of a plurality of nystagmus tests performed on the target patient are pre-stored in the nystagmus test database 700, and the second preprocessor 600 performs a test on the target patient stored in the nystagmus test database 700. The results of multiple nystagmus tests may be read as multiple target patient nystagmus test result features (TN_Fs).

Thereafter, the second preprocessor 600 performs the first preprocessor 400 to generate, for each of the plurality of patients, a plurality of gait features (G_Fs) of the patient, a plurality of bio features (B_Fs) of the patient, and a plurality of target patient gait features (TG_Fs) of the target patient in the same manner as the method of generating an input vector (IN_VEC) including a plurality of nystagmus test result features (N_Fs) of the target patient. A target patient input vector (TIN_VEC) including a plurality of target patient bio features (TB_Fs) and a plurality of target patient nystagmus test result features (TN_Fs) of the target patient may be generated (step S510).

The second preprocessor 600 may provide the target patient input vector (TIN_VEC) generated for the target patient to the machine learning module 500.

The machine learning module 500 may predict a disease related to dizziness of the target patient based on the disease prediction model and the target patient input vector (TIN_VEC) (step S600).

In one embodiment, the machine learning module 500 uses the disease prediction model to classify the target patient input vector (TIN_VEC) as one of the first to nth diseases (DS1 to DSn) and classifies the classified disease as one of the first to nth diseases (DS1 to DSn). It can be predicted based on the disease the target patient has.

For example, the machine learning module 500 inputs a target patient input vector (TIN_VEC) into the disease prediction model and then determines the disease related to dizziness that the target patient has based on the results output from the disease prediction model. can be determined as one of the first to nth diseases (DS1 to DSn).

Meanwhile, in FIG. 4, the first preprocessing unit 400 and the second preprocessing unit 600 are shown as separate components, but the present invention is not limited thereto.

Depending on the embodiment, the first preprocessor 400 and the second preprocessor 600 may be integrated and perform both the above-described operations of the first preprocessor 400 and the second preprocessor 600. there is.

As described above with reference to FIGS. 1 to 5, the dizziness-related disease prediction system 20 and the dizziness-related disease prediction method according to embodiments of the present invention additionally use the results of a plurality of nystagmus tests performed on the patient. By generating the disease prediction model, the accuracy of disease prediction can be further improved.

The present invention can be usefully used to predict diseases related to dizziness in a target patient using only the target patient's gait factors and basic physical information.

As described above, the present invention has been described with reference to preferred embodiments, but those of ordinary skill in the art may vary the present invention without departing from the spirit and scope of the present invention as set forth in the claims below. You will understand that it can be modified and changed.

10, 20: Dizziness-related disease prediction system
100: IMU sensor 110: first IMU sensor
120: second IMU sensor 200: walking feature creation unit
300: body information database 400: first preprocessing unit
500: machine learning module 600: second preprocessor
700: Nystagmus test database

Claims (16)

The gait feature generator creates, for each of a plurality of patients, a plurality of gait features related to the patient's gait using measurement values measured through an IMU (Inertial Measurement Unit) sensor attached to the patient while the patient is walking. generating step;
A first preprocessor generates an input vector including a plurality of bio features corresponding to the plurality of gait features of the patient and body information of the patient for each of the plurality of patients, and generates a first preprocessor related to dizziness. determining the disease that the patient has among the through nth (n is a positive integer of 2 or more) diseases as a label for the input vector;
A machine learning module uses the input vectors for the plurality of patients and the pairs of labels for the input vectors as learning data to determine the first to nth diseases based on the input vectors of the patients. Among them, generating a disease prediction model by training the artificial neural network to output the disease that the patient has;
generating, by the gait feature generator, a plurality of target patient gait features related to the gait of the target patient using measurement values measured through the IMU sensor attached to the target patient while the target patient is walking;
A second preprocessor generating a target patient input vector including a plurality of target patient bio features corresponding to the plurality of target patient gait features and body information of the target patient; and
A dizziness-related disease prediction method comprising the step of predicting, by the machine learning module, a disease related to dizziness of the target patient based on the disease prediction model and the target patient input vector. The dizziness-related method of claim 1, wherein the IMU sensor includes a first IMU sensor attached to a left shoe worn by the patient when walking and a second IMU sensor attached to a right shoe worn by the patient when walking. How to predict disease. The method of claim 1, wherein the IMU sensor includes a 3-axis acceleration sensor, a 3-axis gyro sensor, and a 3-axis geomagnetic sensor. 2. The method of claim 1, wherein the plurality of gait features comprises a left foot inversion-eversion range of motion representing a range of angles in which the left ankle joint moves in and out of the body, and a left foot inversion-eversion range of motion representing a range of angles in which the right foot ankle joint moves in and out of the body. A method for predicting dizziness-related diseases including the right foot internal and eversion joint range of motion, which indicates the range of moving angles. The method of claim 1, wherein the plurality of walking features include Gait Asymmetry (GA), which represents the ratio of the swing time of the left foot and the swing time of the right foot when walking. . The method of claim 1, wherein the plurality of walking features include a Phase Coordination Index (PCI) corresponding to a variable combining the variability and asymmetry of left-foot gait and right-foot gait during walking. How to predict dizziness-related diseases. The method of claim 1, wherein the plurality of gait features include cadence. 2. The method of claim 1, wherein the plurality of gait features represent left foot dorsi-flexion - a range of angles in which the left ankle joint moves in a direction with the instep closer to the lower leg and in a direction with the instep away from the lower leg. Dizziness-related, including plantar-flexion range of motion and right foot dorsiflexion-plantarflexion range of motion, which is the range of angles in which the right ankle joint moves with the top of the foot closer to the lower leg and the other with the top of the foot away from the lower leg. How to predict disease. The method of claim 1, wherein the plurality of bio features include height. The dizziness-related method of claim 1, wherein the first to nth diseases that are labels for the input vector include at least one disease related to the vestibular system and at least one disease unrelated to the vestibular system. How to predict disease. The method of claim 1, wherein the first to nth diseases that are labels for the input vector include musculoskeletal disorders. The method of claim 1, wherein the step of generating the plurality of walking features by the walking feature generator includes:
generating, for each of the plurality of patients, a first plurality of gait features using measurements taken via the IMU sensor while the patient walks at a preferred speed;
For each of the plurality of patients, a plurality of second gait features corresponding to the first plurality of gait features using measurements taken via the IMU sensor while the patient is walking at a slower than the preferred speed. generating them;
For each of the plurality of patients, a plurality of third gait features corresponding to the first plurality of gait features using measurements taken via the IMU sensor while the patient is walking at a speed faster than the preferred speed. generating them; and
A dizziness-related disease prediction method comprising outputting the plurality of gait features including all of the first plurality of gait features, the plurality of second gait features, and the plurality of third gait features. The method of claim 1, wherein the first preprocessor generates the input vector including the plurality of gait features of the patient and the plurality of bio features of the patient for each of the plurality of patients, and The step of determining the disease that the patient has among the to nth diseases using the label for the input vector,
The results of the plurality of gait features of the patient, the plurality of bio features of the patient, and the plurality of nystagmus tests performed by the first preprocessor on the patient for each of the plurality of patients. Dizziness comprising the step of generating the input vector including a plurality of nystagmus test result features corresponding to, and determining a disease that the patient has among the first to nth diseases as the label for the input vector. How to predict related diseases. The method of claim 13, wherein the plurality of nystagmus test result features represent a ratio between the size of the left tonic deviation and the size of the right tonic deviation measured by performing a caloric test through a video nystagmometer. Method for predicting dizziness-related diseases, including: The method of claim 13, wherein the plurality of nystagmus test result features include the presence or absence of spontaneous nystagmus measured by performing a spontaneous nystagmus test. For each of a plurality of patients, a plurality of walking features related to the patient's gait are generated using measurement values measured through an IMU (Inertial Measurement Unit) sensor attached to the patient while the patient is walking, and a goal a gait feature generator that generates a plurality of target patient gait features related to the gait of the target patient using measurement values measured through the IMU sensor attached to the target patient while the patient is walking;
For each of the plurality of patients, an input vector including the plurality of gait features of the patient and a plurality of bio features corresponding to the body information of the patient is generated, and first to nth ( n is a positive integer of 2 or more) a first preprocessor that determines a disease that the patient has among diseases as a label for the input vector;
a second preprocessor that generates a target patient input vector including a plurality of target patient bio features corresponding to the plurality of target patient gait features and body information of the target patient; and
Using the input vectors for the plurality of patients and the pairs of labels for the input vectors as learning data, an artificial neural network determines whether the patient is selected from among the first to nth diseases based on the input vector of the patient. A machine learning module that generates a disease prediction model by training the artificial neural network to output a disease, and predicts a disease related to dizziness of the target patient based on the disease prediction model and the target patient input vector. Dizziness-related disease prediction system.