Wearable body motion recorder-based circadian rhythm real-time extraction method

Abstract

The invention belongs to the technical field of medical care, and relates to a circadian rhythm real-time extraction method based on a wearable body motion recorder. The method comprises the following steps: acquiring physiological data related to a human body circadian rhythm by using a wearable body motion recorder; setting the highest sub-harmonic number of a to-be-extracted circadian rhythm component, and constructing a state equation of a constant deviator, a main frequency and each harmonic component of the adaptive filter; training data are selected, a circadian rhythm frequency band range extracted by the filter is set, and a target function used for determining undetermined parameters of the adaptive filter is established; optimizing the objective function, and determining an adaptive filtering optimal parameter; and inputting the test data into the filter with the optimal parameter, and determining the circadian rhythm amplitude-frequency parameter and phase at the current moment by using the historical state of the filter and the data at the current moment. The method is good in real-time performance, wide in application range, high in accuracy and low in testing cost.

Description

Technical Field

[0001] The invention belongs to the technical field of medical care and relates to a real-time extraction method of circadian rhythm based on a wearable body action recorder. Background Art

[0002] Circadian rhythm is of great significance to human physiology, psychology, health and cognition. In the process of circadian rhythm research, it is necessary to test and characterize the amplitude and frequency parameters of a given individual rhythm. The traditional circadian rhythm phase test method uses the initial release time of dim light melatonin onset (DLMO) or the minimum time of core body temperature (CBT). min As a time calibration point for circadian rhythm, DLMO requires the collection of plasma and saliva samples at specific time intervals under dim light conditions in the evening and is determined by separation and testing. The testing process is complicated and the testing cost is high. min The level of CBT throughout the day needs to be measured and a minimum value determined, based on DLMO and CBT min The circadian rhythm determination methods can only estimate the time of the circadian rhythm phase mark point within a specific time, and cannot determine the circadian rhythm at a certain moment in real time based on the rhythm data of the previous moment.

[0003] Chinese patent documents CN 105981019 B and CN 106231993 B respectively disclose methods for determining circadian rhythm using one or more wearable sensors, which rely on using sensors to determine the exposure of the predicted object, and then combine acceleration sensors or sleep characteristics to predict circadian rhythm. However, the exposure data measured by the sensor cannot be guaranteed to be consistent with the exposure received by the retina and circadian rhythm system of the predicted object, and the light sensor also increases the complexity of the test system and method. Summary of the invention

[0004] In order to overcome the limitations of traditional rhythm testing methods, the present invention provides a method for extracting harmonic components and calculating human rhythms in real time based on wearable body actigraphy signals collected based on adaptive signal processing technology, so as to achieve the purpose of improving the effectiveness and convenience of human rhythm feature extraction. The body actigraphy sensor is used to test rhythm signals such as body movement level, temperature, heart rate, etc., and the rhythm amplitude-frequency characteristics at the current moment are determined in real time based on the sensor signal at the current moment, which can be used as an effective online detection method for human circadian rhythm.

[0005] The technical solution provided by the present invention is: a real-time rhythm feature extraction method based on a wearable body actigraphy recorder, comprising the following steps: (1) Use wearable actigraphy to collect physiological data related to human circadian rhythm; (2) Set the highest harmonic number of the circadian rhythm component to be extracted and construct the state equations of the constant bias, main frequency and harmonic components of the adaptive filter; (3) Selecting training data, setting the circadian rhythm frequency band range extracted by the filter, and establishing an objective function for determining the parameters to be determined by the adaptive filter; (4) Optimize the objective function and determine the optimal parameters of the adaptive filter; (5) Input the test data into the filter with the optimal parameters, and use the filter's historical state and the current data to determine the circadian rhythm amplitude-frequency parameters and phase at the current moment.

[0006] Preferably, the data collected in step (1) is expressed as: (1); in, To collect data, Indicates the number of types of physiological data related to human circadian rhythm; represents the fundamental frequency of circadian rhythm; represents the number of harmonics, and , Respectively represent Rhythmic signal The amplitude and phase offset of subrhythmic harmonics, Indicates Constant bias of rhythm-like signals; Indicates time.

[0007] Preferably, step (2) specifically includes: setting the highest harmonic order extracted by the filter to Second, establish the adaptive filter state The state equation is: (2); by is the filter input signal, and the filter output signal Express Input signal at time The predicted value of also represents the rhythmic component of the original signal extracted by the filter. represents the filter frequency, and the submatrix in the state equation (2) and It is expressed as: (3); in, represents the constant bias adaptation coefficient of the adaptive filter, and the submatrix in matrix (3) is and It is expressed as: (4); Add a frequency adaptive term to the filter, and the filter frequency satisfies the differential equation: (5); in, represents the filter frequency adaptation coefficient, and the initial conditions of the filter state equations (2) and (5) are assumed to be and , the filter is Circadian component of time prediction for: (6); in, represents the filter damping coefficient; Substituting equations (1), (3), (4) and (6) into differential equations (2) and (5), the stable solution of the filter state equation is obtained as: (7); Filter frequency Converge to the fundamental frequency of the circadian rhythm, that is: (8); No. Real-time prediction value of rhythm phase of circadian rhythm signals , expressed as: (9).

[0008] Preferably, in step (3), the method for determining the optimal constant bias adaptation coefficient of the filter is , filter frequency adaptation coefficient and the filter damping coefficient The objective function form is as follows: (10); in, and Represent the filter input signal and output signal The discrete Fourier transform result of ; is the frequency band range extracted by the filter in the training data, Indicates except Other frequency components.

[0009] Preferably, in step (4), the optimal problem of the objective function is solved by a genetic algorithm to determine the optimal constant bias adaptation coefficient of the filter. , filter frequency adaptation coefficient and the filter damping coefficient : (11).

[0010] Preferably, step (5) specifically includes: setting the filter parameters to the calculation results of step (4), using the initial conditions selected in step (2), inputting the signal , numerically calculate the filter system equations (2) and (5), and use formula (6) to calculate The circadian rhythm amplitude of each signal at time is calculated using formula (8): The circadian rhythm frequency at the time is calculated using formula (9): The circadian rhythm phase of each signal at the time.

[0011] Compared with the prior art, the present invention has the following beneficial effects: (1) Only wearable actigraphy data is used to extract circadian rhythms, and the testing cost is low; (2) The current rhythm state is determined online using the adaptive filter's historical state and current data, which has good real-time performance; (3) The rhythmic components are extracted by using filters with optimized parameters and high harmonics, which has a wide range of applications and high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A schematic flow chart of a method for real-time extraction of circadian rhythm based on a wearable body actigraphy device provided in an embodiment of the present invention; Figure 2 The time-frequency domain distribution of the original signal collected by the embodiment of the present invention; wherein, (a) frequency domain; (b) time domain; Figure 3 A time-frequency domain comparison diagram of input and output signals of an adaptive filter in an embodiment of the present invention; wherein, (a) frequency domain; (b) time domain; Figure 4 Frequency variation curve of the adaptive filter in the embodiment of the present invention; Figure 5 The circadian rhythm phase and phase offset curves of each signal predicted by the adaptive filter in the embodiment of the present invention; wherein, (a) phase; (b) phase offset; Figure 6 Melatonin concentration (pg / mL) level curve collected in the embodiment of the present invention and the corresponding time of DLMO; wherein, (a) 7th day; (b) 14th day. DETAILED DESCRIPTION

[0013] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0014] The present invention provides a method for real-time extraction of circadian rhythm based on a wearable body actigraphy device, the process is as follows: Figure 1 As shown, the specific steps are as follows:

[0015] The first step is to collect data

[0016] In this embodiment, the GT3X+ energy monitor is used to collect the activity statistics of the experimental subjects every minute for two consecutive weeks. The components x(t)=[x 1 (t), x 2 (t), x 3 (t)] T , the data is collected when the circadian rhythm is stable and is not affected by factors such as season and work pressure. The frequency domain and time domain distribution of the activity signal used in the embodiment of the present invention are as follows: Figure 2 shown.

[0017] The second step is to establish the filter state equation

[0018] This embodiment considers using a filter to extract the constant bias, fundamental to 5th harmonic components of the three-axis activity statistical signal. Therefore, the highest harmonic order extracted by the filter is set to , the rhythm signal to be extracted is expressed as: (1); in, To collect data, Indicates the number of types of physiological data related to human circadian rhythm; represents the fundamental frequency of circadian rhythm; represents the number of harmonics, and , Respectively represent Rhythmic signal The amplitude and phase offset of subrhythmic harmonics, Indicates Constant bias of rhythm-like signals; Indicates time.

[0019] Set the adaptive filter state The state equation is: (2); The filter output signal Represents the filter pair Input signal at time The predicted value of the state equation (2) is the neutron matrix and It is expressed as: , (3); in, Represents the constant bias adaptation coefficient of the adaptive filter, and the filter frequency satisfies the differential equation: (4); in Represents the filter frequency adaptation coefficient, at time Filter output signal , that is, the rhythm predicted by the filter is expressed as: (5); in Represents the filter damping coefficient.

[0020] Filter initial state are all set to 0, and the initial frequency is set to , by substituting equations (1), (3) and (5) into the system differential equations (2) and (4) and performing numerical integration, the stable solution of the filter state equation (i.e., the filter state convergence value) can be obtained as: (6).

[0021] Filter frequency Converges to: (7).

[0022] No. Real-time prediction of the phase of circadian rhythm signals It is expressed as: (8).

[0023] The third step is to establish the objective function

[0024] Take part of the collected signal as training data and use genetic algorithm to determine the filter parameters , set the objective function to: (9).

[0025] Frequency range of circadian rhythm related signals in this embodiment It contains a constant offset and first to fifth harmonic components, the first harmonic component corresponds to a frequency near the 24-hour day and night cycle, and in this embodiment, it is set Range: (10).

[0026] The passband range is Figure 3 As shown in the gray area, Indicates except The frequency components other than the above are regarded as noises unrelated to the rhythm and are attenuated and filtered out by filters.

[0027] Step 4: Optimize adaptive filter parameters

[0028] Solve the optimal problem of objective function through genetic algorithm: (11).

[0029] The optimal constant bias adaptation coefficient of the adaptive filter in this embodiment is calculated as follows: , filter frequency adaptation coefficient , and the filter damping coefficient .

[0030] Step 5: Calculate circadian rhythm amplitude and frequency parameters

[0031] The optimal parameters calculated in the fourth step are used as the adaptive filter parameters, and the data collected in the first step is used as the input signal of the filter. The calculated filter output signal is compared with the input signal. Figure 3 As shown in Figure 2, the adaptive filter retains the constant bias and 1st to 5th harmonic components in the original signal, filters out the noise signal in the original signal, and the filter output can be used as the circadian rhythm component.

[0032] The adaptive filter frequency (predicted value of circadian rhythm frequency) and the predicted value of the rhythm phase of each signal are as follows Figure 4 and Figure 5 As shown, the circadian rhythm frequency showed an overall oscillatory downward trend, based on the phase shift (instantaneous phase With initial phase The circadian rhythm of the subject in this embodiment showed an overall trend of delayed rhythm within 14 days.

[0033] In order to verify the accuracy of the method proposed in the present invention, melatonin was collected from the experimental subjects on the 7th and 14th days of the experiment. Figure 6 As shown in the blue curve in the figure, based on the method of determining the DLMO time proposed in the paper "The dim light melatonin onset following fixed and freesleep schedules" published by HJ Burgess and CI Eastman (Journal of Sleep Research, Vol. 14, pp. 229-237, 2005), the corresponding DLMO times of the 7th and 14th days are as follows Figure 6 As shown by the black dotted line in the figure. Based on the DLMO results, the rhythm phase of the subject on the 14th day was delayed by about 1 hour and 25 minutes compared with the 7th day. Based on the circadian rhythm phase extraction method of the present invention, the three signal phase offset values ​​are averaged to obtain the rhythm phase offset , compared to the 14th day Average value compared with the 7th day The average value was 1 hour and 32 minutes lower. The circadian rhythm predicted based on wearable sensors was delayed by 1 hour and 32 minutes on the 14th day compared with the 7th day. The predicted results were consistent with the DLMO results.

[0034] In summary, an embodiment of real-time extraction of circadian rhythm components using wearable actigraphy has been described, as well as consistency with traditional DLMO methods for predicting circadian rhythms. The method of the present invention is based only on actigraphy data, and does not require exposure data and predicted sleep. Real-time tracking and extraction of rhythm frequency, harmonic amplitude and phase are achieved using adaptive filters, and filter parameters are optimized using genetic algorithms, aiming to improve the accuracy of online estimation of circadian rhythms using wearable actigraphy.

Claims

1. A real-time extraction method of circadian rhythm based on a wearable actigraphy recorder, characterized in that: The steps include: (1) Use wearable actigraphy to collect physiological data related to human circadian rhythm; (2) Set the highest harmonic number of the circadian rhythm component to be extracted and construct the state equations of the constant bias, main frequency and harmonic components of the adaptive filter; (3) Selecting training data, setting the circadian rhythm frequency band range extracted by the filter, and establishing an objective function for determining the parameters to be determined by the adaptive filter; (4) Optimize the objective function and determine the optimal parameters of the adaptive filter; (5) Input the test data into the filter with the optimal parameters, and use the filter's historical state and the current data to determine the circadian rhythm amplitude-frequency parameters and phase at the current moment.

2. The method for real-time extraction of circadian rhythm based on a wearable body actigraphy according to claim 1, characterized in that: The data collected in step (1) is expressed as: （1）； in, To collect data, Indicates the number of types of physiological data related to human circadian rhythm; represents the fundamental frequency of circadian rhythm; represents the number of harmonics, and , Respectively represent Rhythmic signal The amplitude and phase offset of the subrhythmic harmonics, Indicates Constant bias of rhythm-like signals; Indicates time.

3. The method for real-time extraction of circadian rhythm based on a wearable body actigraphy according to claim 2, characterized in that: Step (2) specifically includes: Set the highest harmonic order extracted by the filter to Second, establish the adaptive filter state The state equation is: （2）； by is the filter input signal, and the filter output signal Express Input signal at time The predicted value of also represents the rhythmic component of the original signal extracted by the filter. represents the filter frequency, and the submatrix in the state equation (2) and It is expressed as: （3）； in, represents the constant bias adaptation coefficient of the adaptive filter, and the submatrix in matrix (3) is and It is expressed as: （4）； Add a frequency adaptive term to the filter, and the filter frequency satisfies the differential equation: （5）； in, represents the filter frequency adaptation coefficient, and the initial conditions of the filter state equations (2) and (5) are assumed to be and , the filter is Circadian component of time prediction for: （6）； in, represents the filter damping coefficient; Substituting equations (1), (3), (4) and (6) into differential equations (2) and (5), the stable solution of the filter state equation is obtained as: （7）； Filter frequency Converge to the fundamental frequency of the circadian rhythm, that is: （8）； No. Real-time prediction value of rhythm phase of circadian rhythm signals , expressed as: （9）。 4. The method for real-time extraction of circadian rhythm based on a wearable body actigraphy according to claim 3, characterized in that: In step (3), it is used to determine the optimal constant bias adaptation coefficient of the filter , filter frequency adaptation coefficient and the filter damping coefficient The objective function form is as follows: （10）； in, Represents the discrete Fourier transform result of the filter input signal; Represents the discrete Fourier transform result of the filter output signal; is the frequency band range extracted by the filter in the training data, Indicates except Other frequency components.

5. The method for real-time extraction of circadian rhythm based on a wearable body actigraphy according to claim 4, characterized in that: The optimal problem of the objective function is solved by genetic algorithm, and the optimal constant bias adaptation coefficient of the filter is determined , filter frequency adaptation coefficient and the filter damping coefficient : （11）。 6. The method for real-time extraction of circadian rhythm based on a wearable body actigraphy according to claim 3, characterized in that: Step (5) specifically includes: setting the filter parameters to the calculation results of step (4), using the initial conditions selected in step (2), and inputting the signal , numerically calculate the filter system equations (2) and (5), and use formula (6) to calculate The circadian rhythm amplitude of each signal at time is calculated using formula (8): The circadian rhythm frequency at the time is calculated using formula (9): The circadian rhythm phase of each signal at the time.

Images