Wearable identification system, method and terminal for vertigo type

Abstract

The wearable identification system, method and terminal of the vertigo type of the present invention divides the walking acceleration time series data collected during the walking process of the user to be identified into gait cycles through a wearable device, extracts the gait stability characteristics to obtain gait stability characteristic data, and finally obtains the vertigo type identification result from the gait stability characteristic data based on a machine learning recognition model that can identify BPPV and VM. The intelligent classification algorithm proposed in the present invention has high classification accuracy, recall rate and precision rate, and can be used for wearable gait and balance function monitoring equipment. Through the data collection and recognition software of the wearable device, it can monitor the patient's gait stability characteristics in real time, quickly and easily identify and feedback the two different types of vertigo, BPPV and VM, thereby effectively reducing the risk of misdiagnosis by doctors.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence medicine, and in particular to a wearable vertigo type identification system, method and terminal. Background Art

[0002] Vertigo is a kind of dynamic or positional illusion caused by the body's spatial positioning disorder. Its etiology is complex, with different clinical classifications and types. Different types of diseases that cause vertigo require symptomatic treatment to be effective, otherwise it is difficult to relieve the symptoms, which is not conducive to the patient's prognosis. Benign paroxysmal positional vertigo (BPPV) and vestibular migraine (VM) are the most common diseases that cause vertigo. Among them, BPPV, also known as otolithiasis, refers to a peripheral vestibular disease caused by changes in head position, accompanied by characteristic nystagmus as the main manifestation, which may be accompanied by autonomic nervous system symptoms such as nausea and vomiting. It is characterized by transient, recurrent and self-limited, and is easy to relapse. VM is a recurrent vertigo disease. Vertigo attacks are often accompanied by migraine headaches, photophobia and phonophobia, migraine symptoms of visual aura, and a history of migraine. The clinical features of some VM and BPPV are similar, which makes clinical diagnosis very difficult. Therefore, it is of great clinical significance to develop a convenient and effective method to identify VM and BPPV, two different vertigo syndromes.

[0003] Chinese patent CN112790939A records a body position change chair for diagnosis and treatment of vertigo, including a platform and a support column, which has the advantages of being able to detect vertigo in patients in a weightless state and being able to brake urgently when an emergency occurs, solving the problem that existing body position change chairs cannot detect vertigo in patients in a weightless state and cannot brake in time when an emergency occurs. Chinese patent CN110638418A records a vertigo disease differential diagnosis system, which solves the shortcomings of existing head impulse test detection technology and has the characteristics of strong practicality, high degree of automation, and low operator requirements. Chinese patent CN208693258U records a vertigo diagnosis and treatment device, which, under the joint action of the revolution device and the rotation device, can realize free rotation in three dimensions and hovering at any position, thereby achieving the effect of semicircular canal VOR function detection and otolith repositioning.

[0004] However, the vertigo diagnosis and treatment equipment disclosed in the above patents are all based on the clinical manifestations of vertigo and nystagmus of patients, which will cause a certain degree of pain to patients during diagnosis. In addition, when identifying diseases with similar vertigo manifestations, it is not possible to distinguish different types of vertigo diseases well, which brings difficulties to subsequent treatment. Summary of the invention

[0005] In view of the above-mentioned shortcomings of the prior art, an object of the present invention is to provide a wearable identification system, method and wearable terminal for vertigo type, so as to solve the above-mentioned problems of the prior art.

[0006] To achieve the above-mentioned purpose and other related purposes, the present invention provides a wearable vertigo type wearable identification system, the system comprising: a data acquisition module, used to acquire walking acceleration time series data collected during the walking process of the user to be identified; wherein the walking acceleration time series data comprises: synchronously collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data; a gait cycle division module, connected to the data acquisition module, used to divide the walking acceleration time series data into gait cycles to obtain gait cycle division data; wherein the gait cycle division data contains acceleration time series of multiple gait cycles; a feature extraction module, connected to the gait cycle division module, for extracting gait stability features from the gait cycle division data to obtain gait stability feature data; a vertigo type recognition module, connected to the feature extraction module, for obtaining a vertigo type recognition result of the user to be identified based on the vertigo type recognition model and the gait stability feature data; wherein the vertigo types identified by the vertigo type recognition result include: benign paroxysmal positional vertigo and vestibular migraine.

[0007] In one embodiment of the present invention, the head acceleration time series data include: the acceleration time series of the head AP axis, the acceleration time series of the head ML axis, and the acceleration time series of the head VT axis; the waist acceleration time series data include: the acceleration time series of the waist AP axis, the acceleration time series of the waist ML axis, and the acceleration time series of the waist VT axis; the left ankle acceleration time series data include: the acceleration time series of the left ankle AP axis, the acceleration time series of the left ankle ML axis, and the acceleration time series of the left ankle VT axis; the right ankle acceleration time series data include: the acceleration time series of the right ankle AP axis, the acceleration time series of the right ankle ML axis, and the acceleration time series of the right ankle VT axis.

[0008] In one embodiment of the present invention, the walking acceleration time series data is divided into gait cycles to obtain gait cycle division data, including: based on multiple peak points of the acceleration time series of the waist VT axis and the corresponding time points, the walking acceleration time series data is divided into one or more gait cycle acceleration time series.

[0009] In one embodiment of the present invention, the walking acceleration time series data is divided into one or more acceleration time series of gait cycles based on the multiple peak points of the acceleration time series of the lumbar VT axis and the corresponding time points, including: obtaining one or more gait cycles based on the adjacent peak points of the acceleration time series of the lumbar VT axis and the time points corresponding to each peak point; according to each gait cycle, dividing the walking acceleration time series data into acceleration time series corresponding to each gait cycle; wherein the acceleration time series of each gait cycle includes: the acceleration time series of two adjacent strides.

[0010] In one embodiment of the present invention, the step of extracting gait stability features from the gait cycle division data to obtain gait stability feature data includes: respectively calculating the gait stability features corresponding to each axis of the head acceleration time series data, the waist acceleration time series data, the left ankle acceleration time series data, and the right ankle acceleration time series data to obtain gait stability feature data; wherein the gait stability features include: acceleration root mean square, acceleration amplitude variability, acceleration harmonic ratio, single-step rhythmicity, gait rhythmicity, and gait symmetry.

[0011] In one embodiment of the present invention, the vertigo type recognition model is obtained by training the gait stability feature vector matrix corresponding to multiple walking acceleration time series data samples; wherein the gait stability feature vector matrix includes: gait stability feature data and vertigo type labels corresponding to each walking acceleration time series data sample; and each walking acceleration time series data sample is composed of synchronously collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data.

[0012] In one embodiment of the present invention, the method for obtaining the gait stability feature vector matrix includes: obtaining multiple walking acceleration time series data samples collected respectively during the walking process of one or more users; dividing each walking acceleration time series data sample into gait cycles respectively to obtain each gait cycle division data; wherein each of the gait cycle division data contains acceleration time series of multiple gait cycles; extracting the gait stability characteristics corresponding to each axis of the head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data in each gait cycle division data to obtain the gait stability feature data corresponding to each walking acceleration time series data sample; based on the obtained vertigo type label marked on each gait stability feature data and each gait stability feature data, the gait stability feature vector matrix is ​​obtained.

[0013] In one embodiment of the present invention, the vertigo type recognition model adopts a KNN classification algorithm.

[0014] To achieve the above-mentioned purpose and other related purposes, the present invention provides a wearable identification method for vertigo type, the method comprising: obtaining walking acceleration time series data collected during the walking of a user to be identified; wherein the walking acceleration time series data comprises: synchronously collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data; dividing the walking acceleration time series data into gait cycles to obtain gait cycle division data; wherein the gait cycle division data contains acceleration time series of multiple gait cycles; extracting gait stability features from the gait cycle division data to obtain gait stability feature data; based on a vertigo type identification model, obtaining a vertigo type identification result of the user to be identified according to the gait stability feature data; wherein the vertigo types corresponding to the vertigo type identification result include: benign paroxysmal positional vertigo and vestibular migraine.

[0015] To achieve the above-mentioned purpose and other related purposes, the present invention provides a wearable identification terminal of the vertigo type, comprising: one or more memories and one or more processors or smart phone terminals; the one or more memories are used to store monitored gait data and vertigo identification software; the one or more processors are connected to the memories, and are used to run the vertigo identification software to execute the method described, and to provide real-time feedback on the monitored gait stability characteristics and vertigo identification results.

[0016] To achieve the above-mentioned purpose and other related purposes, the present invention provides a smart phone terminal, which includes: the wearable identification terminal for vertigo type.

[0017] As described above, the present invention is a wearable identification system, method and terminal for vertigo type, which has the following beneficial effects: the present invention divides the walking acceleration time series data collected during the walking process of the user to be identified into gait cycles through a wearable device, and extracts the gait stability characteristics to obtain gait stability feature data, and finally obtains the vertigo type identification from the gait stability feature data based on a machine learning recognition model that can identify BPPV and VM, and feeds back the monitored gait stability characteristics and vertigo identification type in real time on the wearable identification terminal. The intelligent classification algorithm proposed in the present invention has high classification accuracy, recall rate and precision rate, and can be used for wearable gait and balance function monitoring equipment. Through data collection and software operation of the wearable device, it can easily monitor the patient's gait stability characteristics in real time, and quickly identify two different types of vertigo, BPPV and VM, thereby effectively reducing the risk of misdiagnosis by doctors. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Shown is a schematic structural diagram of a wearable system for identifying vertigo types in one embodiment of the present invention.

[0019] Figure 2 Shown is a flow chart of a wearable method for identifying vertigo types in one embodiment of the present invention.

[0020] Figure 3 Shown is a schematic diagram of the BPPV and VM recognition model construction process in one embodiment of the present invention.

[0021] Figure 4 Shown is a schematic structural diagram of a wearable identification terminal for vertigo type in one embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0023] It should be noted that in the following description, reference is made to the accompanying drawings, which describe several embodiments of the present invention. It should be understood that other embodiments may also be used, and that mechanical composition, structural, electrical and operational changes may be made without departing from the spirit and scope of the present invention. The following detailed description should not be considered restrictive, and the scope of the embodiments of the present invention is limited only by the claims of the published patents. The terms used herein are only for describing specific embodiments and are not intended to limit the present invention. Spatially related terms, such as "upper", "lower", "left", "right", "below", "below", "lower", "above", "upper", etc., may be used in the text to facilitate the description of the relationship between an element or feature shown in the figure and another element or feature.

[0024] Throughout the specification, when a part is said to be "connected" to another part, this includes not only the case of "direct connection" but also the case of "indirect connection" by placing other elements therebetween. In addition, when a part is said to "include" a certain constituent element, unless otherwise stated, it does not exclude other constituent elements, but means that other constituent elements may be included.

[0025] The terms first, second and third mentioned herein are used to describe various parts, components, regions, layers and / or segments, but are not limited thereto. These terms are only used to distinguish a certain part, component, region, layer or segment from other parts, components, regions, layers or segments. Therefore, the first part, component, region, layer or segment described below may refer to the second part, component, region, layer or segment within the scope of the present invention.

[0026] Furthermore, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless there is an indication to the contrary in the context. It should be further understood that the terms "comprise", "include" indicate the presence of the described features, operations, elements, components, items, kinds, and / or groups, but do not exclude the presence, occurrence or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms "or" and "and / or" used herein are interpreted as inclusive, or mean any one or any combination. Therefore, "A, B or C" or "A, B and / or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B and C". Exceptions to this definition will only occur when the combination of elements, functions or operations is inherently mutually exclusive in some way.

[0027] Clinical studies have found that vertigo can affect patients' walking state and reduce their walking stability. Therefore, there are significant differences in gait and balance function among patients with different types of vertigo.

[0028] Therefore, this application uses machine learning technology to propose a wearable identification system for vertigo types. By extracting the patient's gait stability characteristics and constructing a recognition model based on machine learning, it assists doctors in identifying two different types of vertigo, benign paroxysmal positional vertigo (BPPV) and vestibular migraine (VM), thereby effectively reducing the risk of misdiagnosis by doctors. There are currently no related reports, and it will provide a convenient and effective new solution for the clinical differential diagnosis of different types of vertigo diseases.

[0029] The following is a detailed description of the embodiments of the present invention with reference to the accompanying drawings so that those skilled in the art can easily implement the present invention. The present invention can be embodied in many different forms and is not limited to the embodiments described herein.

[0030] like Figure 1 A schematic structural diagram of a wearable identification system for vertigo type in an embodiment of the present invention is shown.

[0031] Applied to a wearable device, the system comprises:

[0032] The data acquisition module 11 is used to acquire, through the wearable device, walking acceleration time series data arranged in time and collected by the acceleration sensor during the walking process of the user to be identified; wherein the walking acceleration time series data includes: synchronously collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data; in one embodiment, the head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data synchronously collected by wearable acceleration sensors respectively worn on the head, waist, left ankle and right ankle of the user to be identified are acquired.

[0033] A gait cycle division module 12, connected to the data acquisition module 11, is used to divide the walking acceleration time series data into gait cycles to obtain gait cycle division data. The gait cycle division data contains acceleration time series of multiple gait cycles;

[0034] A feature extraction module 13, connected to the gait cycle division module 12, for extracting gait stability features from the gait cycle division data to obtain gait stability feature data;

[0035] The vertigo type recognition module 14 is connected to the feature extraction module 13, and is used to obtain the vertigo type recognition result of the user to be identified based on the gait stability feature data based on the vertigo type recognition model; wherein the vertigo types corresponding to the vertigo type recognition result include: benign paroxysmal positional vertigo and vestibular migraine. Specifically, the gait stability feature data is input into the vertigo type recognition model, and the vertigo type recognition result of the user to be identified is output, that is, the user to be identified is judged to be benign paroxysmal positional vertigo or vestibular migraine. This solution can quickly feedback the gait stability monitoring index and the recognition result of the vertigo type.

[0036] Optionally, the wearable device includes three-axis acceleration sensors respectively worn on the head, waist, left ankle and right ankle of the user to be identified; wherein the three axes are the front-back axis (AP axis), the left-right axis (ML axis) and the vertical axis (VT axis). Therefore, the head acceleration time series data includes: the acceleration time series of the head AP axis, the acceleration time series of the head ML axis and the acceleration time series of the head VT axis; the waist acceleration time series data includes: the acceleration time series of the waist AP axis, the acceleration time series of the waist ML axis and the acceleration time series of the waist VT axis; the left ankle acceleration time series data includes: the acceleration time series of the left ankle AP axis, the acceleration time series of the left ankle ML axis and the acceleration time series of the left ankle VT axis; the right ankle acceleration time series data includes: the acceleration time series of the right ankle AP axis, the acceleration time series of the right ankle ML axis and the acceleration time series of the right ankle VT axis. Preferably, the collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data are arranged in time to obtain an acceleration time series matrix.

[0037] Optionally, dividing the walking acceleration time series data into gait cycles to obtain gait cycle division data includes:

[0038] The acceleration time series of the lumbar VT axis can be displayed as continuous waveform data with the X-axis as time and the Y-axis as acceleration value; based on the waveform data, the peak points (acceleration values) on the Y-axis and the time points on the X-axis corresponding to each peak point can be confirmed; based on the multiple peak points of the acceleration time series of the lumbar VT axis and the corresponding time points, the walking acceleration time series data can be divided into one or more acceleration time series of continuous gait cycles.

[0039] Among them, the gait cycle division data includes: the acceleration time series of each continuous gait cycle; further, the acceleration time series of each gait cycle includes: the acceleration time series of the head AP axis, the acceleration time series of the head ML axis, the acceleration time series of the head VT axis, the acceleration time series of the waist AP axis, the acceleration time series of the waist ML axis, the acceleration time series of the waist VT axis, the acceleration time series of the left ankle AP axis, the acceleration time series of the left ankle ML axis, the acceleration time series of the left ankle VT axis, the acceleration time series of the right ankle AP axis, the acceleration time series of the right ankle ML axis and the acceleration time series of the right ankle VT axis in each gait cycle.

[0040] Optionally, dividing the walking acceleration time series data into one or more gait cycles according to each peak value to obtain gait cycle division data includes:

[0041] One or more gait cycles are obtained based on the adjacent peak points of the acceleration time series of the waist VT axis and the time points corresponding to each peak point; according to each gait cycle, the walking acceleration time series data is divided into acceleration time series corresponding to each gait cycle.

[0042] Specifically, based on the time points of the adjacent peak points of the acceleration time series of the waist VT axis, the walking acceleration time series data is divided into acceleration time series containing multiple gait cycles;

[0043] Among them, the acceleration time series between two adjacent peak points is a single step, and the data between two adjacent peak points is a single step time series; the acceleration time series between three adjacent peak points is the acceleration time series of a complete gait cycle, that is, the acceleration time series of two consecutive single steps. The time interval between the first peak point and the third peak point is a gait cycle, and the data between three adjacent peak points is a gait cycle sequence.

[0044] Optionally, the extracting gait stability features from the gait cycle division data to obtain gait stability feature data includes: respectively calculating the gait stability features corresponding to each axis of the head acceleration time series data, the waist acceleration time series data, the left ankle acceleration time series data, and the right ankle acceleration time series data to obtain gait stability feature data; specifically, the gait stability feature data include: the gait stability features of the head AP axis acceleration time series, the head ML axis acceleration time series, the head VT axis acceleration time series, the waist AP axis acceleration time series, the waist ML axis acceleration time series, the waist VT axis acceleration time series, the left ankle AP axis acceleration time series, the left ankle ML axis acceleration time series, the left ankle VT axis acceleration time series, the right ankle AP axis acceleration time series, the right ankle ML axis acceleration time series, and the right ankle VT axis acceleration time series.

[0045] The gait stability characteristics include: acceleration root mean square, acceleration amplitude variability, acceleration harmonic ratio, single-step rhythmicity, gait rhythmicity and gait symmetry.

[0046] Specifically, the acceleration root mean square (RMS): RMS is a measure of the dispersion of acceleration time series data relative to zero. Its calculation formula is:

[0047]

[0048] Among them, ai is the acceleration sequence, and n is the sequence length.

[0049] The acceleration amplitude variability (AV) is the standard deviation of the acceleration peak value of each gait cycle.

[0050] The acceleration harmonic ratio (HR): HR is used as an indicator to describe the smoothness and rhythmicity of acceleration. The calculation of the harmonic ratio is based on a continuous walking test. The calculation of this parameter uses the stride frequency as the fundamental component to calculate the harmonic content of the acceleration signal in each direction.

[0051] Firstly, the finite Fourier series algorithm is used to extract the harmonic components in the acceleration signal, and then the ratio of the sum of the first ten even harmonics to the sum of the first ten odd harmonics is solved.

[0052]

[0053]

[0054] The step regularity (SR1), the size of the first peak in the acceleration autocorrelation signal, reflects the repetitive pattern between single steps; that is, the autocorrelation sequence of the acceleration signal in a single experiment is calculated, and the peak closest to the zero point is extracted.

[0055] The size of the second peak in the stride regularity (SR2) acceleration autocorrelation signal reflects the repetitive pattern in the gait cycle; that is, the autocorrelation sequence of the acceleration signal in a single experiment is calculated, and the peak value second closest to the zero point is extracted.

[0056] The gait symmetry (GS) is calculated as SR1 / SR2. The closer the value is to 1, the more symmetrical the gait is, and the more similar the walking patterns in the left and right directions are.

[0057] Optionally, the gait stability characteristics are extracted from the acceleration time series data in three-axis directions of each part divided into gait cycles to obtain a gait stability characteristic data vector matrix.

[0058] Optionally, the vertigo type recognition model is obtained by training a gait stability feature vector matrix corresponding to a plurality of walking acceleration time series data samples; wherein the feature matrix includes: gait stability feature data and vertigo type labels corresponding to each walking acceleration time series data sample; and each walking acceleration time series data sample is composed of synchronously collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data, and right ankle acceleration time series data. wherein the vertigo type label is a BPPV label or a VM label.

[0059] Preferably, one data sample refers to one walking of a subject. A subject needs to walk multiple times, and walking experimental data of about 10 times are collected to obtain 10 walking acceleration time series data samples.

[0060] Optionally, the prediction accuracy of the vertigo type identification model is verified; specifically, the method for verifying the vertigo type identification model is a 10-fold cross-validation method. That is, the data set is divided into 10 mutually exclusive subsets of similar size. Each subset is obtained by stratified sampling from the data set, and the consistency of data distribution is maintained as much as possible. Each time training, the union of k-1 subsets is used as the training set, so that 10 training and tests can be performed, and finally the mean of the 10 tests is returned. The final model should use the model retrained on the entire data set.

[0061] Preferably, the personalized model verification is performed by the 10-fold cross-validation method: 90% of the constructed feature matrix is ​​used as the training set for each training and test, and the remaining 10% is used as the test set to verify the accuracy (including accuracy, recall rate and precision rate) of the vertigo type recognition model. In order to verify the robustness and population applicability of the model, a universal model verification is applied: when dividing the training set and the test set, it is ensured that no subject's data is in both the training set and the test set, that is, the data of the same subject only appears in the training set or only appears in the test set to prevent data leakage.

[0062] Optionally, the vertigo type recognition model is evaluated; specifically, the method for evaluating the vertigo type recognition model is preferably: counting the number of true positive examples (TP), false positive examples (FP), true negative examples (TN), and false negative examples (FN), and calculating the accuracy, recall rate, and precision rate of the vertigo type recognition model. (BPPV is recorded as a positive example, and VM is recorded as a negative example)

[0063] Wherein, the accuracy calculation method is preferably:

[0064]

[0065] The recall calculation method is preferably:

[0066]

[0067] The precision calculation method is preferably:

[0068]

[0069] Optionally, the vertigo type recognition model uses a K-Nearest Neighbor (KNN) classification algorithm. The KNN classification algorithm includes the following 4 steps: prepare data and preprocess the data; calculate the distance from the test sample point (that is, the point to be classified) to each other sample point; sort each distance, and then select the K points with the smallest distance; compare the categories to which the K points belong, and according to the principle of minority obeys majority, classify the test sample point into the category with the highest proportion among the K points. The vertigo type recognition model has good robustness and can accurately identify BPPV and VM. Taking BPPV as a positive example, the accuracy, recall and precision of the constructed KNN classification algorithm for identifying BPPV and VM reached 82.9%, 87.0% and 80.5% respectively.

[0070] Optionally, the method of acquiring the gait stability characteristic vector matrix includes:

[0071] Acquire multiple walking acceleration time series data samples collected during the walking process of one or more users; specifically, collect AP, ML, and VT three-axis acceleration time series data of the head, waist, and left and right ankles of each user during normal walking speed, and obtain multiple acceleration time series matrices.

[0072] Each walking acceleration time series data sample is divided into a gait cycle to obtain gait cycle division data containing multiple gait cycles corresponding to each walking acceleration time series data sample; specifically, the gait cycle division of each walking acceleration time series data sample includes: based on multiple peak points of the acceleration time series of the waist VT axis and the corresponding time points, dividing the walking acceleration time series data into one or more acceleration time series of gait cycles.

[0073] The gait stability features corresponding to each axis of the head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data in each gait cycle division data are extracted to obtain the gait stability feature data corresponding to each walking acceleration time series data sample; specifically, the gait stability features of the acceleration time series of the head AP axis, the head ML axis, the head VT axis, the waist AP axis, the waist ML axis, the waist VT axis, the left ankle AP axis, the left ankle ML axis, the left ankle VT axis, the right ankle AP axis, the right ankle ML axis and the right ankle VT axis of each walking acceleration time series data sample are extracted to obtain the gait stability feature data of each walking acceleration time series data sample.

[0074] Based on the obtained vertigo type labels and each gait stability feature data marked with each gait stability feature data, the gait stability feature vector matrix is ​​obtained. Specifically, the BPPV label or VM label marked with each gait stability feature data label is obtained according to the diagnosis result of a clinician with professional background knowledge. The vertigo type label and each gait stability feature data marked with each gait stability feature data are combined to obtain the gait stability feature vector matrix corresponding to the model.

[0075] Optionally, the method for constructing the vertigo type recognition model includes:

[0076] The AP, ML, and VT three-axis accelerations of the head, waist, and left and right ankles of each user while walking at a normal pace are collected to obtain the acceleration time series matrix.

[0077] Based on the adjacent positions of each peak value, each acceleration time series matrix is ​​divided into acceleration time series data of multiple gait cycles to obtain gait cycle division data containing acceleration time series data of each gait cycle; wherein, the acceleration time series data of each gait cycle contains acceleration time series data of two adjacent single steps. Using the acceleration time series data of walking, the walking process of the subject is divided into several gait cycles. Specifically, the acceleration time series between two adjacent peak points is a single step. The data between two adjacent peak points is a single step time series; the acceleration time series between three adjacent peak points is the acceleration time series of a complete gait cycle, that is, the acceleration time series of two consecutive single steps. The time interval between the first peak point and the third peak point is a gait cycle, and the data between three adjacent peak points is a gait cycle time series data.

[0078] All gait stability features are extracted from a single experiment, i.e., a single walking acceleration time series data sample, i.e., the gait stability features corresponding to each axis of the head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data, and right ankle acceleration time series data are calculated to obtain gait stability feature data; wherein, the gait stability features include: acceleration root mean square, acceleration amplitude variability, acceleration harmonic ratio, single-step rhythmicity, gait rhythmicity, and gait symmetry. In the generated training set, principal component analysis is used for feature dimensionality reduction, and the eigenvectors corresponding to the eigenvalues ​​with the top 95% cumulative contribution rate of the eigenvalues ​​are selected. In the test set, the same parameters are used to perform principal component analysis dimensionality reduction on the features.

[0079] The diagnosis results of clinicians with professional background knowledge are obtained, and each gait stability feature data is marked with a BPPV label or a VM label.

[0080] Based on the acquired gait stability feature data and the annotated vertigo type labels, multiple 74-dimensional feature vectors are obtained; that is, each 74-dimensional feature vector corresponds to a walking acceleration time series data sample. The feature vectors are concatenated by column to form a feature matrix of the recognition model. Among them, the first 72 dimensions of each 74-dimensional feature vector represent the RMS, AV, HR, SR1, SR2 and GS parameters of the head, waist, left ankle and right ankle on AP, ML, VT, respectively, and the last 2 dimensions of data represent the labels of BPPV and VM. The following table is the first 72-dimensional input feature number table.

[0081] Table 1: Input feature number table

[0082]

[0083]

[0084] That is, the 74-dimensional feature matrix consisting of multiple columns (each column corresponds to a walking acceleration time series data sample) is as follows:

[0085]

[0086] The vertigo type recognition model is obtained by training a 74-dimensional feature matrix corresponding to a plurality of walking acceleration time series data samples.

[0087] Similar to the principle of the above embodiment, the present invention provides a wearable method for identifying the type of vertigo.

[0088] The following provides specific embodiments in conjunction with the accompanying drawings:

[0089] like Figure 2 A schematic flow chart showing a wearable method for identifying vertigo types in an embodiment of the present invention.

[0090] Applied to a wearable device, the method comprises:

[0091] Step S21: Acquire walking acceleration time series data collected during the walking process of the user to be identified.

[0092] The walking acceleration time series data includes: synchronously collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data, and right ankle acceleration time series data.

[0093] Optionally, the head acceleration time series data include: the acceleration time series of the head AP axis, the acceleration time series of the head ML axis, and the acceleration time series of the head VT axis; the waist acceleration time series data include: the acceleration time series of the waist AP axis, the acceleration time series of the waist ML axis, and the acceleration time series of the waist VT axis; the left ankle acceleration time series data include: the acceleration time series of the left ankle AP axis, the acceleration time series of the left ankle ML axis, and the acceleration time series of the left ankle VT axis; the right ankle acceleration time series data include: the acceleration time series of the right ankle AP axis, the acceleration time series of the right ankle ML axis, and the acceleration time series of the right ankle VT axis.

[0094] Step S22: dividing the walking acceleration time series data into gait cycles to obtain gait cycle division data.

[0095] Specifically, each of the gait cycle division data contains acceleration time series of a plurality of gait cycles.

[0096] Optionally, step S22 includes: based on multiple peak points of the acceleration time series of the waist VT axis and the corresponding time points, dividing the walking acceleration time series data into one or more acceleration time series of gait cycles.

[0097] Preferably, one or more gait cycles are obtained based on adjacent peak points of the acceleration time series of the waist VT axis and the time points corresponding to each peak point; according to each gait cycle, the walking acceleration time series data is divided into acceleration time series corresponding to each gait cycle; wherein the acceleration time series of each gait cycle includes: the acceleration time series of two adjacent single steps.

[0098] Step S23: extracting gait stability characteristics from the gait cycle division data to obtain gait stability characteristic data.

[0099] Optionally, step S23 includes: calculating the gait stability characteristics corresponding to each axis of the head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data, and right ankle acceleration time series data, respectively, to obtain gait stability characteristic data; wherein the gait stability characteristics include: acceleration root mean square, acceleration amplitude variability, acceleration harmonic ratio, single-step rhythmicity, gait rhythmicity and gait symmetry.

[0100] Step S24: Based on the vertigo type recognition model, obtain the vertigo type recognition result of the user to be identified according to the gait stability characteristic data.

[0101] Among them, the vertigo types corresponding to the vertigo type recognition results include: benign paroxysmal positional vertigo and vestibular migraine.

[0102] The optional vertigo type recognition model is obtained by training the gait stability feature vector matrix corresponding to multiple walking acceleration time series data samples; wherein the gait stability feature vector matrix includes: gait stability feature data and vertigo type labels corresponding to each walking acceleration time series data sample; and each walking acceleration time series data sample is composed of synchronously collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data.

[0103] Optionally, the method for obtaining the gait stability feature vector matrix includes: obtaining multiple walking acceleration time series data samples collected respectively during the walking process of one or more users; dividing each walking acceleration time series data sample into gait cycles to obtain gait cycle division data; wherein each of the gait cycle division data contains acceleration time series of multiple gait cycles; extracting the gait stability characteristics corresponding to each axis of the head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data in each gait cycle division data to obtain gait stability feature data corresponding to each walking acceleration time series data sample; based on the obtained vertigo type label marked on each gait stability feature data and each gait stability feature data, obtaining the feature matrix of the recognition model.

[0104] Optionally, the vertigo type identification model adopts a KNN classification algorithm.

[0105] Since the implementation principle of the wearable identification method for vertigo type has been described in the above-mentioned embodiment, it will not be repeated here.

[0106] In order to better illustrate the above-mentioned wearable identification method for vertigo type, the present invention provides the following specific embodiments.

[0107] Embodiment 1: A wearable method for identifying vertigo type.

[0108] Step 1: Construct BPPV and VM recognition model; Figure 3 The figure shows a schematic diagram of the process of constructing a BPPV and VM recognition model; the method includes:

[0109] S1, data collection: The three-axis acceleration time series data of the head, waist, left and right ankles of each patient, namely the acceleration time series data of AP, ML, and VT axes, are collected by a three-axis accelerometer, with a sampling frequency of 148.48 Hz. The clinician's diagnosis of the patient's disease is used as the basis for judging BPPV and VM patients.

[0110] S2, feature segmentation: each walking acceleration signal is low-pass filtered, and filtered with a cutoff frequency of 15 Hz using a 4th-order Butterworth model. The filtered signal is divided into continuous gait cycles, where each step is divided by finding the extreme point of the walking acceleration signal in the vertical direction; the local maximum is the time when the foot touches the ground, and the moment when the foot touches the ground twice is the start and end of a single step.

[0111] S3, dataset construction:

[0112] The machine learning model of the present invention accepts 72-dimensional input, i.e., 72-dimensional features extracted from a single experiment of a subject, and outputs one of two situations - "BPPV" or "VM", i.e., the prediction of the machine learning model that the subject suffers from vertigo.

[0113] During training, a gait experiment of the subject is regarded as a sample. The sample label is obtained by a professional doctor based on the subject's clinical manifestations and medical history. All experiments collected from the subject are used as the data set. Ten-fold cross validation is used to evaluate the effectiveness of the algorithm.

[0114] S4, model training and validation: In this embodiment, all data were collected from 16 VM patients and 16 BPPV patients. Personalized model validation was performed using the 10-fold cross-validation method: 90% of the constructed feature matrix (14 VM patients, 14 BPPV patients, and 14 healthy people) was used as the training set for each training and test, and the remaining 10% (2 VM patients, 2 BPPV patients, and 2 healthy people) was used as the test set to test the accuracy of the BPPV and VM recognition models (including accuracy, recall rate, and precision rate). In order to verify the robustness and population applicability of the model, a universal model validation was applied: when dividing the training set and the test set, it was ensured that no subject's data was in both the training set and the test set, that is, the data of the same subject only appeared in the training set or only appeared in the test set to prevent data leakage. Principal component analysis was used to perform feature dimensionality reduction, and the eigenvectors corresponding to the eigenvalues ​​with the top 95% of the cumulative contribution rate of the eigenvalues ​​were selected. In this embodiment, KNN was selected as the classification algorithm.

[0115] Because the prediction result of the classification algorithm is BPPV or VM, the output of all predicted values ​​Yj constitutes the prediction sequence matrix.

[0116] Yj=(y1,y2,y3,y4,…,y n-1 ,y n ); (8)

[0117] Where n is the number of samples.

[0118] After obtaining the optimal parameters of the prediction model through training, the prediction model accuracy (accuracy, recall and precision) was calculated on the test set. According to the comparison between the prediction results and the actual results, the methods of BPPV and VM recognition models were evaluated, the number of true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN) were counted, the confusion matrix was constructed, and the accuracy, recall and precision of the prediction model were calculated.

[0119] The data set is divided into 10 mutually exclusive subsets of similar size. Each subset is obtained by stratified sampling from the data set, and the consistency of data distribution is maintained as much as possible. The union of k-1 subsets is used as the training set for each training, so that 10 training and testing can be performed, and the mean of the 10 tests is finally returned. The final model uses the model retrained on the entire data set. After actual testing on patients, the present invention can accurately identify BPPV and VM. Taking BPPV as a positive example, the accuracy, recall and precision of the algorithm reached 82.9%, 87.0% and 80.5% respectively.

[0120] Step 2: Based on the collected walking acceleration time series data and the constructed BPPV and VM recognition models, the BPPV and VM recognition results and gait stability characteristics are obtained.

[0121] This embodiment develops an intelligent recognition algorithm for BPPV and VM based on wearable and machine learning technology, which is aimed at the two most common diseases that cause vertigo in clinical practice, BPPV and VM, to assist doctors in quickly and easily identifying these two different types of vertigo, thereby effectively reducing the risk of misdiagnosis by doctors. This embodiment constructs a KNN classification algorithm for identifying BPPV and VM by extracting the characteristics of the patient's pathological gait and gait stability. The model has good robustness and can accurately identify BPPV and VM. Taking BPPV as a positive example, the accuracy, recall rate and precision rate of the constructed KNN classification algorithm for identifying BPPV and VM reached 82.9%, 87.0% and 80.5% respectively. The proposed intelligent classification algorithm has high classification accuracy, recall rate and precision rate, and can be used for wearable gait and balance function monitoring equipment. Through data collection and software operation of wearable devices, it can provide real-time feedback on the monitored gait stability characteristics, quickly and easily identify BPPV and VM, two different types of vertigo, thereby effectively reducing the risk of misdiagnosis by doctors.

[0122] like Figure 4 A schematic structural diagram of a wearable identification terminal 10 for vertigo type in an embodiment of the present invention is shown.

[0123] The vertigo type identification terminal 40 includes: a wearable memory 41 and a processor 42. The memory 41 is used to store the monitored gait data and the vertigo identification software program; the processor 42 runs the vertigo identification software program to achieve the following Figure 2 The wearable identification method of vertigo type;

[0124] Optionally, the number of the memory 41 can be one or more, the number of the processor 42 can be one or more, and Figure 4 Take one as an example.

[0125] Optionally, the processor 42 in the vertigo type identification terminal 40 will Figure 1 In the above steps, one or more instructions corresponding to the process of the application are loaded into the memory 41, and the processor 42 runs the application stored in the first memory 41, thereby achieving the following Figure 2 Various functions in the wearable identification method of vertigo type.

[0126] Optionally, the memory 41 may include but is not limited to high-speed random access memory and non-volatile memory. For example, one or more disk storage devices, flash memory devices or other non-volatile solid-state storage devices; the processor 42 may include but is not limited to a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0127] Optionally, the processor 42 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0128] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program is executed as follows: Figure 2The wearable identification method of the type of vertigo shown. The computer-readable storage medium may include, but is not limited to, a floppy disk, an optical disk, a CD-ROM (read-only disk memory), a magneto-optical disk, a ROM (read-only memory), a RAM (random access memory), an EPROM (erasable programmable read-only memory), an EEPROM (electrically erasable programmable read-only memory), a magnetic card or an optical card, a flash memory, or other types of media / machine-readable media suitable for storing machine-executable instructions. The computer-readable storage medium may be a product that is not connected to a computer device, or a component that is connected to a computer device for use.

[0129] The present invention also provides a smart phone terminal, which is applied to a wearable device, and is used to store monitored gait data and run a vertigo recognition software program, and runs the vertigo recognition software.

[0130] In summary, the wearable identification system, method and terminal of the vertigo type of the present invention, through a wearable device, divides the walking acceleration time series data collected during the walking process of the user to be identified into gait cycles, extracts the gait stability characteristics to obtain gait stability feature data, and finally obtains the vertigo type identification result from the gait stability feature data based on the machine learning recognition model that can identify BPPV and VM. The intelligent classification algorithm proposed in the present invention has high classification accuracy, recall rate and precision rate, and can be used for wearable gait and balance function monitoring equipment. Through the data collection and recognition software of the wearable device, it can monitor the patient's gait stability characteristics in real time, quickly and easily identify and feedback BPPV and VM, two different types of vertigo, thereby effectively reducing the risk of misdiagnosis by doctors. Therefore, the present invention effectively overcomes the various shortcomings in the prior art and has a high industrial utilization value.

[0131] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A wearable identification system for vertigo type, characterized in that: Applied to a wearable device, the system comprises: A data acquisition module, used to acquire walking acceleration time series data collected during the walking process of the user to be identified; wherein the walking acceleration time series data includes: synchronously collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data, and right ankle acceleration time series data; A gait cycle division module, connected to the data acquisition module, for dividing the walking acceleration time series data into gait cycles to obtain gait cycle division data; wherein the gait cycle division data contains acceleration time series of multiple gait cycles; a feature extraction module, connected to the gait cycle division module, for extracting gait stability features from the gait cycle division data to obtain gait stability feature data; A vertigo type recognition module is connected to the feature extraction module and is used to obtain a vertigo type recognition result of the user to be identified based on the vertigo type recognition model and the gait stability feature data; wherein the vertigo types corresponding to the vertigo type recognition result include: benign paroxysmal positional vertigo and vestibular migraine; the vertigo type recognition model is obtained by training a gait stability feature vector matrix corresponding to a plurality of walking acceleration time series data samples; wherein the gait stability feature vector matrix includes: gait stability feature data corresponding to each walking acceleration time series data sample and a vertigo type label, wherein the vertigo type label is a benign paroxysmal positional vertigo label or a vestibular migraine label; and each walking acceleration time series data sample is composed of synchronously collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data; the vertigo type recognition model adopts a KNN classification algorithm.

2. The wearable vertigo type identification system according to claim 1, characterized in that: The head acceleration time series data include: the acceleration time series of the head AP axis, the acceleration time series of the head ML axis, and the acceleration time series of the head VT axis; the waist acceleration time series data include: the acceleration time series of the waist AP axis, the acceleration time series of the waist ML axis, and the acceleration time series of the waist VT axis; the left ankle acceleration time series data include: the acceleration time series of the left ankle AP axis, the acceleration time series of the left ankle ML axis, and the acceleration time series of the left ankle VT axis; the right ankle acceleration time series data include: the acceleration time series of the right ankle AP axis, the acceleration time series of the right ankle ML axis, and the acceleration time series of the right ankle VT axis.

3. The wearable vertigo type identification system according to claim 2, characterized in that: The step of dividing the walking acceleration time series data into gait cycles to obtain gait cycle division data comprises: Based on the multiple peak points of the acceleration time series of the waist VT axis and the corresponding time points, the walking acceleration time series data is divided into one or more acceleration time series of gait cycles.

4. The wearable vertigo type identification system according to claim 3, characterized in that: The multiple peak points of the acceleration time series of the waist VT axis and the corresponding time points, dividing the walking acceleration time series data into one or more acceleration time series of gait cycles, includes: Acquire one or more gait cycles based on adjacent peak points of the acceleration time series of the waist VT axis and the time points corresponding to each peak point; According to each gait cycle, the walking acceleration time series data is divided into acceleration time series corresponding to each gait cycle; The acceleration time series of each gait cycle includes: the acceleration time series of two adjacent single steps.

5. The wearable vertigo type identification system according to claim 2, characterized in that: The step of extracting gait stability characteristics from the gait cycle division data to obtain gait stability characteristic data comprises: Calculating the gait stability characteristics corresponding to each axis of the head acceleration time series data, the waist acceleration time series data, the left ankle acceleration time series data, and the right ankle acceleration time series data, respectively, to obtain gait stability characteristic data; The gait stability characteristics include: acceleration root mean square, acceleration amplitude variability, acceleration harmonic ratio, single-step rhythmicity, gait rhythmicity and gait symmetry.

6. The wearable vertigo type identification system according to claim 1, characterized in that: The method of obtaining the gait stability characteristic vector matrix includes: Acquire multiple walking acceleration time series data samples collected during the walking process of one or more users; Divide each walking acceleration time series data sample into a gait cycle to obtain gait cycle division data; wherein each gait cycle division data contains acceleration time series of multiple gait cycles; Extracting the gait stability feature corresponding to each axis of the head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data, and right ankle acceleration time series data in each gait cycle division data, so as to obtain gait stability feature data corresponding to each walking acceleration time series data sample; The gait stability feature vector matrix is ​​obtained based on the vertigo type label annotated with each gait stability feature data and each gait stability feature data.

7. A wearable method for identifying the type of vertigo, characterized in that: Applied to a wearable device, the method comprises: Acquire walking acceleration time series data collected during the walking process of the user to be identified; wherein the walking acceleration time series data includes: synchronously collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data, and right ankle acceleration time series data; Dividing the walking acceleration time series data into gait cycles to obtain gait cycle division data; wherein the gait cycle division data contains acceleration time series of multiple gait cycles; Extracting gait stability characteristics from the gait cycle division data to obtain gait stability characteristic data; Based on the vertigo type recognition model, the vertigo type recognition result of the user to be identified is obtained according to the gait stability feature data; wherein the vertigo types corresponding to the vertigo type recognition result include: benign paroxysmal positional vertigo and vestibular migraine; the vertigo type recognition model is obtained by training the gait stability feature vector matrix corresponding to multiple walking acceleration time series data samples; wherein the gait stability feature vector matrix includes: the gait stability feature data and the vertigo type label corresponding to each walking acceleration time series data sample, wherein the vertigo type label is a benign paroxysmal positional vertigo label or a vestibular migraine label; and each walking acceleration time series data sample is composed of synchronously collected head acceleration time series data, waist acceleration time series data, left ankle acceleration time series data and right ankle acceleration time series data; the vertigo type recognition model adopts a KNN classification algorithm.

8. A wearable identification terminal for vertigo, characterized in that: Applicable to a wearable device, comprising: one or more memories and one or more processors; The one or more memories are used to store the monitored gait data and the vertigo recognition software; The one or more processors are connected to the memory and are used to run the vertigo identification software to execute the method as claimed in claim 7, and to provide real-time feedback on the monitored gait stability characteristics and vertigo identification results.

9. A smart phone terminal, characterized in that: Applied to a wearable device, the terminal comprises: a wearable identification terminal for vertigo type as described in claim 8.

Images