TWI809686B - Method and apparatus for assessing fall risk - Google Patents

Abstract

A method and an apparatus for assessing fall risk are provided. The method includes following steps: receiving gait signals generated by a sensor detecting movement caused by a testee; performing a feature extracting analysis on the gait signals to obtain first feature values; performing a feature selecting analysis on the first feature values to obtain second feature values; performing classification analysis on the second feature values according to historical diagnostic data to obtain a fall risk assessment result of the testee.

Description

Fall risk assessment method and device

The present invention relates to a method and device for detecting the gait condition of a subject, and in particular to a fall risk assessment method and device.

The current common fall risk assessment method is to judge the fall risk assessment through physical therapists or professional medical personnel to observe the subject's movements of maintaining standing posture, walking, and changing posture. Rehabilitation practitioners usually assess fall risk based on existing measurement forms and steps, such as the Berger's Balance Scale. The scoring of existing measurement forms and the final assessment of fall risk mainly rely on the rehabilitation practitioner's personal experience and its subjective judgment.

However, the above methods need to rely on the observation and judgment of the rehabilitation practitioner one by one to complete the assessment of the risk of falls. Moreover, since the fall risk assessment is a subjective judgment, the subject usually has to go to a clinic or hospital for testing. Therefore, the existing methods not only increase the medical manpower burden and lead to a shortage of medical manpower, but also reduce the willingness of the subjects to go to the clinic for testing regularly, resulting in high inconvenience for fall risk assessment and testing.

The present invention provides a method and device for assessing the risk of falling, which uses the data of professionals' experience and analyzes the gait signal of gait changes to obtain the result of risk assessment for falling.

The present invention provides a fall risk assessment method, which is suitable for an electronic device with a processor to detect movement gait changes by using sensors arranged on the subject's body. The method includes the following steps: extracting the gait signal generated by the displacement and angle offset generated by the sensor detecting movement changes; performing feature extraction and analysis on the gait signal to obtain a plurality of first features of the gait signal value; according to the historical analysis data, the first eigenvalue is subjected to feature selection analysis to select a plurality of second eigenvalues; according to the historical analysis data and historical gait signals, the second eigenvalues are classified and analyzed to obtain the gait signal Fall risk assessment results.

The invention provides a fall risk assessment device, which includes a data acquisition device, a storage medium and a processor. The data acquisition device is used for connecting with the sensor configured on the subject, and the sensor detects the center of mass deviation and deviation angle of the subject to generate a gait signal. The storage medium is used to store a pre-established classification model and a plurality of learning parameters using machine learning, wherein the classification model uses a plurality of gait signal feature values and corresponding fall risk assessment data, and the fall risk assessment data is based on the subject The evaluations made, for example, fall risk assessment data are, for example, the evaluation results of the Berger Scale, the Simple Berg Scale, the 3-meter up-and-go test, and the risk of dementia. The processor is coupled to the data acquisition device and the storage medium, configured to acquire the sensor detection through the data acquisition device The gait signal generated by measuring the gait condition of the subject is analyzed statistically, in the frequency domain and in the time domain to obtain a plurality of first characteristic values of the gait signal, and using historical analysis to analyze the first characteristic Values are subjected to feature selection analysis to screen out a plurality of second feature values, and the second feature values are input into the classification model to perform fall risk assessment of the subject.

In order to make the above-mentioned features and advantages of the present invention more comprehensible, the following specific embodiments are described in detail together with the accompanying drawings.

100: Fall Risk Assessment Device

110: Data acquisition device

120: storage media

130: Processor

140: Transceiver

200: sensor

S310~S360, S410~S440: steps

FIG. 1 is a block diagram of a fall risk assessment device according to an embodiment of the present invention.

FIG. 2 is a first flowchart of a fall risk assessment method according to an embodiment of the present invention.

FIG. 3 is a second flowchart of a fall risk assessment method according to an embodiment of the present invention.

The embodiment of the present invention proposes a fall risk assessment method and device, which detects the gait signals of the subject when standing, walking, and posture changes by installing sensors on the subject, and analyzes the gait The signal is subjected to eigenvalue extraction and eigenvalue selection to obtain multiple eigenvalues of the gait signal, and then input these eigenvalues into a pre-built and trained classification model using machine learning to estimate the subject's risk of falling risk assessment results. In this way, the embodiment of the present invention can quickly and accurately evaluate the result of the fall risk of the subject without the medical staff personally observing the gait of the subject one by one.

In detail, FIG. 1 is a block diagram of a fall risk assessment device according to an embodiment of the present invention. Please refer to FIG. 1 , the fall risk assessment method device 100 of the embodiment of the present invention is, for example, a computer device such as a file server, a database server, an application program server, a workstation or a personal computer, or a mobile phone or a tablet computer with computing power. and other mobile devices, which include components such as a data capture device 110, a storage medium 120, and a processor 130. Any connected wired or wireless interface device is used to capture the gait signal generated when the sensor 200 detects the subject's movement. For the wired method, the data acquisition device 110 can be a universal serial bus (universal serial bus, USB), RS232, a universal asynchronous receiver/transmitter (universal asynchronous receiver/transmitter, UART), an internal integrated circuit (I2C ), serial peripheral interface (serial peripheral interface, SPI), display port (display port) or thunderbolt port (thunderbolt) and other interfaces, but not limited thereto. For the wireless mode, the data acquisition device 110 may support wireless fidelity (wireless fidelity, Wi-Fi), RFID, bluetooth, infrared, near-field communication (near-field communication, NFC) or device-to-device (device -to-device, D2D) and other communication protocol devices are not limited thereto.

In one embodiment, the sensor 200 is, for example, configured on the body of the subject Accelerometers, gyroscopes, inertial sensors, or a combination of any two or more of the above. After several experiments and statistics, when the sensor 200 of the present invention is set at the third section, the fourth section, the fifth section, or the position between the third section and the fifth section of the subject's spine, the present invention The fall risk assessment method and device 100 have more accurate fall risk assessment results.

The storage medium 120 is, for example, any type of fixed or removable random access memory (Random Access Memory, RAM), read-only memory (Read-Only Memory, ROM), flash memory (Flash memory), hard disk A disk or similar components or a combination of the above components are used to store computer programs executable by the processor 130 . In some embodiments, the storage medium 120 also stores learning parameters of a classification model pre-established and trained by the processor 130 using machine learning. The machine learning includes logistic regression (Logistic Regression), support vector machine (Support Vector Machine, SVM), stacked autoencoder (Stacked Autoencoder, SAE), convolutional neural network (Convolutional Neural Network, CNN) or deep neural network Road (Deep Neural Networks, DNNs), but not limited to this.

The processor 130 is, for example, a central processing unit (Central Processing Unit, CPU), or other programmable general purpose or special purpose microprocessor (Microprocessor), microcontroller (Microcontroller), digital signal processor (Digital Signal Processor) Processor, DSP), programmable controller, Application Specific Integrated Circuits (Application Specific Integrated Circuits, ASIC), programmable logic device (Programmable Logic Device, PLD) or other similar devices or combinations of these devices, the present invention does not this limit. In this embodiment, the processor 130 can load a computer program from the storage medium 120 to execute the A method for fall risk assessment with reinforcement learning.

In one embodiment, the fall risk assessment method device 100 of the embodiment of the present invention further includes a transceiver 140 for communicating with the sensor 200 , and the transceiver 140 transmits and receives signals in a wireless or wired manner. The transceiver 140 may also perform operations such as low noise amplification, impedance matching, frequency mixing, up or down frequency conversion, filtering, amplification, and the like.

FIG. 2 is a first flowchart of a fall risk assessment method according to an embodiment of the present invention. FIG. 3 is a second flowchart of a fall risk assessment method according to an embodiment of the present invention. Please refer to FIG. 1 and FIG. 3 at the same time. The method of this embodiment is applicable to the fall risk assessment device 100 mentioned above. The detailed steps of the fall risk assessment method of this embodiment will be described below with various components of the fall risk assessment device 100 .

In step S310 , the data acquisition device 110 of the fall risk assessment device 100 uses the sensor 200 to detect the displacement of the subject when the movement changes. Next, the sensor 200 generates a corresponding gait signal according to the displacement of the subject, where the gait signal is, for example, when the subject moves, and the sensor 200 generates a center-of-mass offset signal according to the tilt angle and the center-of-mass offset , tilt signal, acceleration signal and angular acceleration signal, or a combination of any two or more of the above. In other words, the gait signal is, for example, the normalized energy in the time domain measured by the sensor 200 when the subject moves, and then a plurality of motions corresponding to each movement of the subject and relative to the previous movement are captured. Gait signal.

In step S320, the processor 130 performs feature extraction analysis on the gait signal to obtain a plurality of first feature values of the gait signal. More details To describe in detail, step S320 can be subdivided into step S410 and step S420. In step S410, the processor 130 performs data pre-processing on the gait signal to obtain axial data of the gait signal. Specifically, the axial data includes x-axis acceleration, y-axis acceleration and z-axis acceleration. In one embodiment, the x-axis acceleration is the acceleration generated by the subject moving in the vertical direction, and the x-axis acceleration is also called V-axis (Vertical Axis) acceleration. The y-axis acceleration is the acceleration generated by the subject moving along the left and right horizontal directions, and the y-axis acceleration is also called the medial lateral (ML) axis acceleration. In addition, the z-axis acceleration is the acceleration generated by the subject moving along the front-back direction, and the z-axis acceleration is also called the Anterior-posterior (AP) axis acceleration.

In step S420, the processor 130 performs feature extraction analysis on the data of each axis to obtain a plurality of first feature values. Specifically, the first feature value includes a total gait feature value and each segment feature value, and the total gait feature value is, for example, the gait feature value of a subject performing a step detection (for example, a 3-meter up-and-go detection) , each segment feature value is, for example, a statistical feature value, a frequency domain feature value, and a randomness feature value about the subject performing a straight line, turning, standing to sitting, sitting to standing. In this embodiment, the statistical characteristic value is, for example, the offset of the body mass center at each sampling time point (t), the average offset velocity of the body mass center at each sampling time point (t), the above-mentioned offset Individual mean, standard deviation, time vs. offset slope, etc. relative to offset velocity. In the present invention, the extraction and analysis of the feature values includes the entire gait signal of the subject and the gait signals of the subject's segments (for example, moving straight, turning, standing to sitting, and sitting to standing).

Next, in this embodiment, the frequency domain feature value is, for example, the offset and the average offset velocity of the body mass center at each sampling time point, and the time domain (Time Domain) to frequency domain (Frequency Domain) conversion to obtain multiple frequency domain eigenvalues. In an embodiment, the conversion from the time domain to the frequency domain may be Fourier transform and/or wavelet transform. Confusion, also known as entropy, is used to describe the degree of confusion of gait signals. The random characteristic value is, for example, the trajectory random characteristic value of x-axis acceleration, y-axis acceleration, and z-axis acceleration; for example, a larger random characteristic value represents a more chaotic signal waveform, and a smaller random characteristic value represents The more stable the signal waveform is.

In step S330, a feature selection analysis (feature selection) is performed on the first feature value according to the historical analysis data to select a plurality of second feature values. The historical analysis data include: the diagnosis/judgment data of the rehabilitation practitioner or medical personnel on the fall risk of other subjects, the diagnosis data and judgment methods of the fall risk in the literature, and the gait signal of each subject and Feature values extracted from gait signals. Specifically, the processor 130 filters out the historical analysis data related to the subject this time according to the characteristic value of the gait signal in the historical data and the corresponding fall risk assessment result. Then, according to the relevant historical analysis data, the characteristic value categories that are related to the fall risk result and/or have a greater influence are obtained, and then a plurality of related second characteristic values are selected from the plurality of first characteristic values of the subject. .

On the other hand, feature selection analysis can be the filtering method, embedding method and packaging method in feature engineering. For example, through the chi-square test in the filtering method, the chi-square of each first feature value and each historical gait information can be calculated value, p-value, degrees of freedom, and the theoretical value of the original data, and then filter out the second eigenvalue related to the gait condition of the subject from the first eigenvalue.

In other words, the processor 130 screens out the The second eigenvalue related to the gait signal of the subject is selected, and the selected second eigenvalue has a greater impact on the fall risk assessment of the subject. In one embodiment, the processor 130 may select a plurality of related second feature values according to a feature selection algorithm in feature engineering of machine learning. The feature selection algorithm is, for example, regression model learning, random forest, chi-square test, or decision tree, etc., and the present invention is not limited thereto. In another embodiment, the processor 130 uses a classification model to perform feature selection analysis on the first feature value according to the fall risk assessment results in the historical analysis data and the corresponding historical gait signals, so as to select the second feature value. The establishment of the classification model is explained as follows.

In step S340, the processor 130 will use the sensor 200 to detect a plurality of first action signals generated by the first action gait change of the subject with the risk of falling and the second action step of the subject without the risk of falling. A plurality of second action signals generated by state changes. performing feature extraction analysis and feature selection analysis on the first motion signal and the second motion signal, respectively, to obtain a plurality of third feature values of the first motion signal and the second motion signal; using the third feature values as a classification model input, and use the corresponding fall risk assessment result as the output of the classification model to train the classification model, and record a plurality of learning parameters of the trained classification model. In this embodiment, the processor 130 may implement Logistic Regression (Logistic Regression), Support Vector Machine (Support Vector Machine, SVM), Stacked Autoencoder (Stacked Autoencoder, SAE), Convolutional Neural Network (Convolutional Neural Network, CNN) or Deep Neural Networks (DNNs) machine learning algorithms to build classification models.

In step S350, the processor 130 classifies and analyzes the second feature value according to the historical analysis data and the historical gait signal, so as to obtain a fall risk assessment result of the gait signal. Specifically, the processor 130 generates the classification factors of each historical subject according to the characteristic values of the historical subjects in the historical analysis data, and groups the historical subjects into multiple test subjects according to the classification factors of each historical subject. group. In this way, the processor 130 can perform statistical analysis on the characteristic values of the subject to generate the classification factor of the subject, and compare the classification factor of each historical patient with at least one classification threshold to determine This time the subject is classified into one of a plurality of subject groups. In an embodiment of the present invention, the above classification factors may include correlation. That is to say, the processor 130 may calculate the correlation of each subject according to the feature values of each subject. Then, the processor 130 can group the test subject into one of the plurality of test subject groups by comparing the correlation of each test subject. In more detail, the subject groups are, for example, subjects with fall risk and subjects without fall risk, and the correlation is a measure of interdependence among variables. In other words, the processor 130 uses the historical analysis data and the fall risk results as the variable Y in the function Y=f(X). The variable Y is, for example, the original Berg Balance Scale (BBS ) or the scores of the Short-Form Berg Balance Scale (SFBBS) and the scores of the three-meter up-and-go (3M-TUG) test, the independent variable X is the The eigenvalues extracted from the state signal are used to establish a model f( ), based on which the second eigenvalue of the subject is input as the independent variable X into the classification model f( ) to obtain the dependent variable Y, which should be Variable Y is the result of the subject's fall risk assessment (for example, whether there is a risk of falling, the score of the Burger's Balance Scale, the total number of seconds to walk from three meters wait).

In step S360, the processor 130 outputs the fall risk result obtained by the classification model to the output device. In another embodiment, the processor 130 transmits the fall risk result obtained by the classification model to the output device through the transceiver 140 and stores it in the storage medium 120 . The output device is, for example, a screen or a terminal device, wherein the terminal device is, for example, an electronic device, a smart phone, a desktop computer, or a notebook computer of a subject or medical personnel.

In step S370, it can be divided into step S430 and step S440. In step S430, the processor 130 performs eigenvalue data analysis on the first eigenvalue to obtain a plurality of gait indicators of the gait signal. In step S440, the processor 130 performs an improvement analysis on the gait index to obtain an improvement suggestion for the gait signal. Specifically, the processor 130 uses the numerical values of the plurality of first feature values of the subject to correspond to and explain the postural stability. For example, one of the plurality of first eigenvalues is the turn eigenvalue of the subject when turning around, and when there is a large difference between the turn eigenvalue and the eigenvalue without the risk of falling in the historical analysis data, then It means that there is a problem with the transfer of the subject's center of gravity this time. At this time, the processor 130 outputs an examination suggestion, which is to suggest that the subject undergo further advanced examination according to Parkinson's disease. In another example, when the turbulence characteristic value of the subject in the y-axis direction is low during sitting-to-stand/stand-to-sit, the processor 130 outputs a suggestion suggesting that the subject perform lower limb muscle strength signals. Accordingly, through the fall risk assessment method and device of the present invention, the subject/user only needs to wear the sensor 200 of the fall risk assessment device 100 on the waist and walk, stand, sit and/or turn around You can quickly and accurately know the fall risk assessment results and prediction results of the subject. Recommendations for preventing falls and improving gait (eg, training recommendations, specific testing recommendations, falls prevention recommendations).

To sum up, in the fall risk assessment method and device 100 of the embodiment of the present invention, the sensor 200 is used to detect the center of mass deviation and acceleration of the subject when moving, and use historical analysis data and corresponding Gait Signals Build a classification model that responds to the characteristics of each gait signal. In this way, whenever a subject with a similar situation wants to assess the risk of falling, the characteristic value of the gait signal can be input into the classification model to calculate the corresponding classification result (ie, the assessment result of whether there is a risk of falling). In addition, the fall risk assessment method and device 100 of the present invention can further generate training focus and specific diagnosis and treatment for follow-up balance improvement corresponding to the condition of the subject according to the gait index corresponding to the characteristic value of the subject.

Although the present invention has been disclosed above with the embodiments, it is not intended to limit the present invention. Anyone with ordinary knowledge in the technical field may make some changes and modifications without departing from the spirit and scope of the present invention. The scope of protection of the present invention should be defined by the scope of the appended patent application.

S310~S360: Steps

Claims (13)

A fall risk assessment method, which is suitable for an electronic device with a processor to use a sensor to detect a plurality of motion changes of a subject, the method includes the following steps: A gait signal generated by the displacement and angle offset generated by the movement change; performing feature extraction analysis on the gait signal to obtain a plurality of first eigenvalues of the gait signal; according to a historical analysis data performing feature selection analysis on the first eigenvalues to select a plurality of second eigenvalues; classifying and analyzing the second eigenvalues according to the historical analysis data and a historical gait signal to obtain the gait A fall risk assessment result of the signal; wherein the step of calculating the fall risk assessment result for the second feature value according to the historical analysis data and the historical gait signal includes: combining the second feature value and the The historical analysis data is input into a classification model pre-established using machine learning to estimate the fall risk assessment result of the gait signal, wherein the classification model uses feature information of a plurality of gait signals and a corresponding plurality of Fall risk assessment results training. The method according to claim 1, further comprising: performing eigenvalue data analysis on the first eigenvalue to obtain multiple gait indicators of the gait signal; performing improved analysis on the gait indicators to obtain Get a change in the gait signal good advice. The method according to claim 1, further comprising: capturing multiple first motion signals generated by the sensor detecting multiple first motion gait changes with a risk of falling and multiple motion signals without a risk of falling A plurality of second action signals generated by the gait change of the second action; performing feature extraction analysis and feature selection analysis on the first action signal and the second action signal respectively, so as to obtain the first action signal and the second action signal. A plurality of third feature values of the second action signal; using the third feature values as the input of the classification model, and using the output of the classification model corresponding to the fall risk assessment result to train the Classification model described. The method as described in claim item 1, wherein said machine learning includes Logistic Regression (Logistic Regression), Support Vector Machine (Support Vector Machine, SVM), Stacked Autoencoder (Stacked Autoencoder, SAE), convolutional neural network ( Convolutional Neural Network, CNN) or deep neural network (Deep Neural Networks, DNNs). The method according to claim 1, wherein the step of performing feature extraction and analysis on the gait signal to obtain the first feature value of the gait signal includes: performing data preprocessing on the gait signal , to respectively obtain a plurality of axial data of the gait signal; performing feature extraction analysis on each of the axial data to obtain the first feature value. The method according to claim 3, wherein the step of performing feature selection analysis on the first feature value according to the historical analysis data to select the second feature value includes: using the classification model according to the The fall risk assessment results in the historical analysis data and the corresponding historical gait signals are subjected to feature selection analysis on the first feature value to select the second feature value. A fall risk assessment device, comprising: a data acquisition device connected to a sensor disposed on a subject, and the sensor detects multiple movement changes of the subject to generate multiple gaits a signal; a storage medium; and a processor, coupled to the data capture device and the storage medium, configured to: capture the quality of the subject detected by the sensor through the data capture device center offset and offset angle to generate the gait signal; perform feature extraction analysis on the gait signal to obtain a plurality of first eigenvalues of the gait signal; analyze the gait signal according to a historical analysis data performing feature selection analysis on the first eigenvalues to select a plurality of second eigenvalues; classifying and analyzing the second eigenvalues according to the historical analysis data and a historical gait signal to obtain the gait A fall risk assessment result of the signal; wherein the processor further uses the sensor to detect a plurality of first movement signals generated by the gait changes of a plurality of first movements having a risk of falling and not having a risk of falling A plurality of second action signals generated by a plurality of second action gait changes, and performing feature extraction analysis and feature selection analysis on the first motion signal and the second motion signal, respectively, to obtain a plurality of third feature values of the first motion signal and the second motion signal, and The third feature value is used as an input of the classification model, and the output of the classification model corresponding to the fall risk assessment result is used to train the classification model. The fall risk assessment device according to claim 7, wherein the processor performs eigenvalue data analysis on the first eigenvalue to obtain a plurality of gait indicators of the gait signal, and analyzes the gait Indicators are analyzed for improvement to obtain an improvement suggestion for the gait signal. The fall risk assessment device according to claim 7, wherein the processor further uses the sensor to detect a plurality of first action signals generated by detecting a plurality of first action gait changes with a risk of falling and Multiple second motion signals generated by multiple second motion gait changes that do not have a risk of falling, and performing feature extraction analysis and feature selection analysis on the first motion signal and the second motion signal, so as to Obtaining a plurality of third feature values of the first motion signal and the second motion signal, and using the third feature values as an input of the classification model, and using the corresponding fall risk assessment result as described in The output of the classification model is used to train the classification model. Fall risk assessment device as described in claim item 7, wherein said machine learning includes Logistic Regression (Logistic Regression), Support Vector Machine (Support Vector Machine, SVM), Stacked Autoencoder (Stacked Autoencoder, SAE), convolutional neural network (Convolutional Neural Network, CNN) or deep neural network (Deep Neural Networks, DNNs). The fall risk assessment device as described in claim 7, wherein the processor further performs data pre-processing on the gait signal, so as to separately obtain multiple axis data of the gait signal, and for each of the axes Perform feature extraction analysis on the data to obtain the first feature value. The fall risk assessment device according to claim 9, wherein the processor further uses the classification model to perform an evaluation of the first The eigenvalues are subjected to feature selection analysis to select the second eigenvalues. The fall risk assessment device according to claim 9, wherein the sensor is disposed between the third to fifth vertebrae of the subject.

Images