WO2020166239A1 - 睡眠時無呼吸症候群判定装置、睡眠時無呼吸症候群判定方法、及び、睡眠時無呼吸症候群判定プログラム - Google Patents
睡眠時無呼吸症候群判定装置、睡眠時無呼吸症候群判定方法、及び、睡眠時無呼吸症候群判定プログラム Download PDFInfo
- Publication number
- WO2020166239A1 WO2020166239A1 PCT/JP2020/000529 JP2020000529W WO2020166239A1 WO 2020166239 A1 WO2020166239 A1 WO 2020166239A1 JP 2020000529 W JP2020000529 W JP 2020000529W WO 2020166239 A1 WO2020166239 A1 WO 2020166239A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- period
- sleep
- feature vector
- sleep apnea
- subject
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording for evaluating the cardiovascular system, e.g. pulse, heart rate, blood pressure or blood flow
- A61B5/024—Measuring pulse rate or heart rate
- A61B5/02405—Determining heart rate variability
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording for evaluating the cardiovascular system, e.g. pulse, heart rate, blood pressure or blood flow
- A61B5/024—Measuring pulse rate or heart rate
- A61B5/0245—Measuring pulse rate or heart rate by using sensing means generating electric signals, i.e. ECG signals
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/08—Measuring devices for evaluating the respiratory organs
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/352—Detecting R peaks, e.g. for synchronising diagnostic apparatus; Estimating R-R interval
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4806—Sleep evaluation
- A61B5/4818—Sleep apnoea
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Definitions
- the present disclosure relates to a sleep apnea syndrome determination device, a sleep apnea syndrome determination method, and a sleep apnea syndrome determination program.
- Sleep apnea syndrome is a disease in which respiratory arrest or respiratory volume decreases during sleep (hereinafter, this condition is also referred to as apnea). Specifically, apnea is a symptom that occurs 30 times or more during 7 hours of sleep, or 5 times or more per hour. Apnea refers to a stasis of 10 seconds or more that occurs during sleep.
- Patent Document 1 calculates a heart rate variability (HRV: Heart Rate Variability) index from heart rate data of a subject obtained by a heart rate measuring device, and based on the heart rate variability index, the subject is apnea or low for each predetermined period during sleep. It distinguishes between breathing and normal breathing. As a result, it is possible to know the temporal change in the breathing state of the subject.
- HRV Heart Rate Variability
- SAS is a risk factor for traffic accidents because it causes excessive daytime sleepiness and poor concentration.
- SAS is known to cause various lifestyle-related diseases and aggravation thereof, causing cardiovascular disorders, glucose metabolism disorders, fat metabolism disorders, and the like. Therefore, it is desired that the determination as to whether or not the SAS is performed with high accuracy.
- the inventors of the present application apply the method of determining the breathing state for each predetermined period during the sleep period as in Patent Document 1, and further, it is possible to highly accurately determine whether or not the subject has SAS. To propose.
- the sleep apnea syndrome determination apparatus determines whether or not the subject has sleep apnea syndrome using RRI data indicating a heartbeat interval (RR Interval: RRI) in the sleep period of the subject.
- a processing unit that generates a feature vector including a plurality of continuous heartbeat intervals in a sleep period, and outputs the feature vector by inputting the feature vector to a recurrent neural network, and includes the input feature vector.
- the index based on the ratio of the period of the apnea state to the period of the normal breathing state in the sleep period is calculated from the values for each of the identification target periods corresponding to the plurality of heartbeat intervals that are set, and based on the index, the subject is It operates so as to determine whether it is a respiratory syndrome.
- a sleep apnea syndrome determination method is a method of determining whether or not a subject has sleep apnea syndrome by using RRI data indicating a heartbeat interval in a sleep period of the subject. Corresponding to a plurality of heartbeat intervals included in the input feature vector, which is output by inputting the feature vector to the recurrent neural network, and a step of generating a feature vector including a plurality of consecutive heartbeat intervals in the sleep period.
- the sleep apnea syndrome determination program uses a computer to determine whether the subject has sleep apnea syndrome by using RRI data indicating a heartbeat interval in the sleep period of the subject. And a step of generating a feature vector including a plurality of continuous heartbeat intervals in a sleep period, and a feature vector that is output by inputting the feature vector to a recurrent neural network. From the values for each identification target period corresponding to a plurality of heartbeat intervals, the step of calculating an index based on the ratio of the period of the apnea state to the period of the normal respiratory state in the sleep period, and the subject sleeps based on the index. And a step of determining whether or not it is temporal apnea syndrome.
- FIG. 1 is a diagram showing an outline of a configuration of a system including a sleep apnea determination apparatus (hereinafter, determination apparatus) according to an embodiment.
- FIG. 2A is a diagram showing an example of an electrocardiographic signal
- FIG. 2B is a diagram showing R wave data.
- FIG. 3 is a diagram for explaining the feature vector.
- FIG. 4 is a flowchart showing the flow of the determination process in the determination device.
- FIG. 5 is a diagram showing the determination conditions of each method evaluated by the evaluation test by the inventors.
- FIG. 6 is a diagram showing the attributes of the subject in the evaluation test.
- FIG. 7: is the figure which showed the determination result by the 1st method in an evaluation test.
- FIG. 1 is a diagram showing an outline of a configuration of a system including a sleep apnea determination apparatus (hereinafter, determination apparatus) according to an embodiment.
- FIG. 2A is a diagram showing an example of an electrocardiographic signal
- FIG. 8 is a diagram showing a determination result by the second method in the evaluation test.
- FIG. 9 is a diagram showing the determination result of the third method in the evaluation test.
- FIG. 10 is a figure which showed the determination result by the 4th method in an evaluation test.
- FIG. 11 is the figure which showed the determination result by the 5th method in an evaluation test.
- FIG. 12 is a diagram showing the evaluation value of the determination result by each method in the evaluation test.
- the sleep apnea syndrome determination apparatus included in the present embodiment uses the RRI data indicating the heartbeat interval (RR Interval: RRI) in the sleep period of the subject to determine whether the subject has the sleep apnea syndrome.
- the processing unit that determines whether or not the processing unit generates a feature vector including a plurality of continuous heartbeat intervals in the sleep period, and outputs the feature vector by inputting the feature vector to the recurrent neural network. From the value for each identification target period corresponding to a plurality of heartbeat intervals included in the vector, calculate an index based on the ratio of the period of apnea and normal breathing state in the sleep period, based on the index, the subject It operates so as to determine whether it is sleep apnea.
- the adoption of a recursive neural network as a discriminant model is far more accurate than the input of the index value calculated from the heartbeat interval to a discriminant model other than the recursive neural network.
- the sleep apnea syndrome determination apparatus can determine the SAS patient with high accuracy.
- the recurrent neural network is preliminarily learned by machine learning so as to output a value indicating the sleep state in the identification target period in response to the input of the feature vector.
- a feature vector including a plurality of heartbeat intervals in the identification target period is used as an input value, and a value indicating the sleep state in the identification target period is output in response to the input of the feature vector.
- the recursive neural network is LSTM (Long Short-Term Memory).
- LSTM Long Short-Term Memory
- the accuracy of the determination is further improved by using the LSTM as the recursive neural network. It was verified.
- the identification target period has a fixed period length
- the feature vector is a variable length vector consisting of a time series of RRI data groups indicating each of a plurality of heartbeat intervals of the identification target period.
- the processing unit when inputting the feature vector to the recursive neural network, the processing unit recursively uses the feature vector generated from a plurality of heartbeat intervals including a heartbeat interval larger than a threshold indicating an abnormal value of the heartbeat interval. Do not input to the neural network. As a result, it is determined whether or not the input value is a normal value (normal value determination), and as a result, only the feature vector generated from only the heartbeat interval determined to be a normal value is input to the recurrent neural network. As a result of the evaluation test by the inventors of the present application, it was verified that the determination accuracy is further improved by performing the normal value determination.
- generating the feature vector includes normalizing the value of each of the plurality of heartbeat intervals.
- the index is a ratio of an apnea period to a sleep period, and in the determining step, when the ratio is larger than a threshold value indicating sleep apnea syndrome, the subject is asleep apnea. Determined to have respiratory syndrome. This makes it easy to determine whether the subject has SAS.
- the sleep apnea syndrome determination method included in the present embodiment is a method of determining whether or not a subject has sleep apnea syndrome by using RRI data indicating a heartbeat interval in the sleep period of the subject. And a step of generating a feature vector including a plurality of heartbeat intervals of the identification target period in the sleep period, and an identification target corresponding to the plurality of heartbeat intervals included in the feature vector in response to the input of the feature vector.
- the sleep apnea syndrome determination program included in the present embodiment uses a computer to determine whether the subject has sleep apnea syndrome by using RRI data indicating a heartbeat interval in the sleep period of the subject.
- a step of calculating an index based on the ratio of the period of apnea and the period of normal breathing in the sleep period, based on the index, whether the subject has sleep apnea The determining step and the executing step are executed.
- the computer can be caused to function as the sleep apnea syndrome determination device of (1) to (7). Therefore, this program has the same effect as the sleep apnea syndrome determination device of (1) to (7).
- a system 100 including the sleep apnea determination apparatus 1 according to the present embodiment shown in FIG. 1 includes a sleep apnea determination apparatus (hereinafter, referred to as “determination apparatus”) 1 and a heartbeat measuring device 2. ..
- the determination device 1 and the heartbeat measuring device 2 can communicate with each other.
- the communication may be wireless communication or wire communication.
- the heartbeat measuring device 2 is a small and lightweight wearable terminal that is attached to the body of the subject P and measures the heartbeat of the subject P.
- a plurality of (three in FIG. 1) electrodes 21A attached to the body surface of the subject P are connected to the heartbeat measuring device 2.
- the three electrodes 21A are, for example, a plus electrode, a minus electrode, and a ground electrode.
- the wearable terminal that functions as the heartbeat measuring device 2 may be, for example, a smart watch having a heartbeat measuring function.
- the wearable terminal itself may function as the determination device 1 and the heartbeat measuring device 2.
- the vertical axis of FIG. 2(a) shows the electric potential
- the horizontal axis shows the time.
- the electrocardiographic signal periodically shows a potential change composed of P to T waves.
- the peak with the highest potential in the potential change in a unit cycle is called an R wave, and the ventricular muscle contracts at the timing of the R wave.
- the heartbeat measuring device 2 transmits the heartbeat data to the determination device 1.
- the heartbeat data is, for example, RRI data.
- RRI data indicates RRI (R-R Interval).
- RRI is a characteristic value obtained from the R wave.
- the RRI data corresponding to the electrocardiographic signal in FIG. 2A has a period (the signal intensity I exceeds a predetermined intensity threshold value Ith) corresponding to the R wave in the electrocardiographic signal.
- the RRI is the interval between R waves in the rectangular pulse train in which the period) is set to “1” and the other periods are set to “0”.
- the determination device 1 receives the RRI data in the sleep period of the subject P transmitted from the heartbeat measuring device 2 and executes the determination process using the RRI data to determine whether the subject P has sleep apnea syndrome. Determine whether.
- the heartbeat data transmitted from the heartbeat measuring device 2 may be a numerical value group representing the electrocardiographic waveform or data representing the electrocardiographic waveform itself.
- the determination device 1 calculates the RRI from the data received from the heartbeat measuring device 2 and uses it for the determination process described later.
- the determination device 1 is composed of a computer including a processing unit 10 and a storage device 20.
- the storage device 20 is connected to the processing unit 10.
- the processing unit 10 is, for example, a CPU (Central Processing Unit).
- the determination device 1 also includes a communication unit 30 for communicating with the heartbeat measuring device 2.
- the communication unit 30 may be a communication mechanism for short-distance wireless communication such as Bluetooth (registered trademark) or may be a communication mechanism for wireless LAN (Local Area Network).
- the communication unit 30 is connected to the processing unit 10.
- the communication unit 30 functions as an input unit that inputs the heartbeat data from the heartbeat measuring device 2.
- the computer forming the determination device 1 is preferably a mobile terminal such as a smartphone or a tablet.
- the mobile terminal held by the subject P can be utilized as the determination device 1, which is preferable.
- the mobile terminal may be a wearable device such as a smart watch.
- a plurality of computers may be included in the determination device 1.
- the determination device 1 may be configured by cooperation of a plurality of mobile terminals.
- the plurality of mobile terminals are, for example, smartphones and smart watches.
- the computer that constitutes the determination device 1 may be a server computer on a network such as the Internet.
- the RRI data transmitted from the heartbeat measuring device 2 of the subject P is transmitted to the server computer via a network such as the Internet.
- the server computer When determining whether or not the server computer has sleep apnea, it is sufficient to notify the terminal of the subject P (mobile terminal or the like) via the network.
- the storage device 20 of the determination device 1 stores a computer program 21 for causing the processing unit 10 to perform determination processing.
- the computer functions as the determination device 1 when the processing unit 10 executes the computer program 21.
- the determination process is a process of determining whether the subject P has sleep apnea using the RRI data in the sleep period of the subject P, and includes a generation process, an identification process, and a calculation process. ..
- the generation process is a process of generating a feature vector from the RRI data group for a predetermined period in the sleep period.
- the feature vector is used as an input value in the identification process described later, and is a collection of a plurality of features indicating the sleep state of the subject in the predetermined period.
- the processing unit 10 functions as the generation unit 11 that executes the generation process. That is, a predetermined period is set as an identification target period that is a target of the identification process described later, and RRIs are arranged in time series from electrocardiographic waveforms measured within the identification target period to form a feature vector. ..
- the number of RRIs measured during the fixed period, which is the identification target period, is not a fixed number. Therefore, the number of elements of the feature vector is variable. That is, the generated feature vector is a variable length vector.
- the generation unit 11 causes the feature vector FV1 in the periods T1 and T2 shown below. , FV2 are generated.
- FV1 ⁇ 925, 905,..., 885 ⁇
- FV2 ⁇ 935,...,915 ⁇
- the generation process includes a normal value determination process for determining whether the RRI is a normal value.
- the generation unit 11 includes a normal value determination unit 12 that executes normal value determination processing.
- the normal value determination unit 12 stores a threshold value of RRI (for example, 2000 msec) in advance and compares it with RRI to determine whether it is abnormal or normal.
- An abnormal value of RRI occurs due to factors such as artifacts.
- the generation unit 11 generates a feature vector only in the identification target period in which all included RRIs are determined to be normal among the identification target periods in the sleep period, and identifies the identification target period in which at least one abnormal RRI is included. Does not generate a feature vector.
- the feature vector for the identification target period in which even one RRI is determined to be abnormal is not used as the feature vector for the determination process.
- the generated feature vector does not include an abnormal value. That is, only the feature vector having the normal value of RRI is used for the determination process. As a result, the determination accuracy can be improved.
- the feature vector may be generated using a fixed number of continuous RRIs.
- the number of elements of the feature vector is constant and the feature vector has a fixed length.
- the identification target period has a variable length.
- the determination device 1 generates a variable-length feature vector from a plurality of consecutive RRIs included in the identification target period, which is a fixed-length period, and uses it for the determination process, rather than using a fixed number of RRIs. It is preferable.
- a recursive neural network hereinafter also referred to as RNN
- RNN recursive neural network
- the generation process includes a standardization process for standardizing the RRI group in the identification target period.
- the generation unit 11 includes a standardization processing unit 13 that executes standardization processing.
- the standardization process is, for example, a normalization process, for example, a process of converting the RRI group in the identification target period into data having an average RRI of 0 and a variance of 1 in a predetermined period using the following formula. ..
- zi (xi- ⁇ )/ ⁇ (Xi represents each RRI, ⁇ represents the average value, and ⁇ represents the standard deviation.)
- the feature vector it may be considered to be a vector composed of a plurality of HRV indexes.
- the HRV index can be, for example, two or more of the following (1) to (11).
- the sleep state identification using the HRV index identifies whether the subject is apnea or normal breathing for each time during sleep, and determines whether the subject is apnea or normal breathing. It is performed in the determination method that acquires the dynamic variation.
- a discrimination model a support vector machine that discriminates between an apnea state and a normal breathing state when an HRV index is given is used.
- meanNN average of RRI.
- SDNN standard deviation of RRI.
- Total Power Total power spectrum of RRI.
- RMSSD Square root of the mean value of the square of the difference between the nth RRI and the (n+1)th RRI. The greater the RRI variation, the greater the value, which is an indicator of the activity of the autonomic nervous system.
- NN50 The number in which the difference between the nth and the n+1th RRI exceeds 50 milliseconds. Normal RRI variation is less than 50 ms and NN50 is an indicator of severe variation.
- pNN50 Ratio of the value of NN50 to the number of all R waves. Since the NN50 is the number of times, it generally has a large value if the window width is long.
- LF Power spectrum in the frequency band of the LF band (0.04 Hz to 0.15 Hz). LF is mainly used as an index of sympathetic nervous system activity.
- HF Power spectrum in the HF band (0.15 Hz to 0.4 Hz) frequency band. HF is mainly used as an index of parasympathetic nervous system activity.
- LF/HF A ratio of LF and HF defined by LF/HF, which is considered to be a ratio of activity states of the sympathetic nervous system and the parasympathetic nervous system. The value of LF/HF is large when the sympathetic nerve is dominant, and is small when the parasympathetic nerve is dominant.
- LFnu correction LF defined by LF/(LF+HF). Emphasize changes in sympathetic nerve activity.
- HFnu corrected HF defined by HF/(LF+HF). Emphasize changes in parasympathetic activity.
- the RRI is a value directly obtained from the electrocardiographic waveform of the subject
- the continuous RRI is a group of values that directly represent the features of the electrocardiographic waveform during the sleep period of the subject.
- the inventors analyzed that the HRV index is not expressed as directly as RRI because it is an index calculated from RRI or a value related to RRI. Therefore, based on this analysis, the inventors decided to generate a feature vector from the RRI itself and use it for the identification processing described later. As a result, rather than using the feature vector generated from the HRV index, a more direct feature of the electrocardiographic waveform is used for the identification process described later. As a result, the determination accuracy can be improved.
- the identification process is a process of identifying whether the sleep state is an apnea state or a normal state for each identification target period.
- the processing unit 10 functions as the identification unit 14 that executes the identification process.
- the identification unit 14 includes an identification model 15.
- the identification model 15 is preliminarily learned by machine learning in response to the input of the feature vector so as to output a value indicating the sleep state in the identification target period corresponding to the feature vector.
- the identification model 15 receives, for example, "1" indicating an apnea state with respect to the input of the feature vectors FV1 and FV2. , And “0” indicating a normal breathing state are output.
- Apnea states refer to apnea or hypopnea.
- An apnea condition is defined as a condition where ventilation is stopped for 10 seconds or longer.
- the hypopnea state is defined as a state in which the respiratory airflow is reduced by 50% or more for 10 seconds or more. Thereby, the output of the value indicating the respiratory state is obtained for each identification target period.
- a recursive neural network is used as the identification model 15.
- the RNN is, for example, Simple RNN (Simple Recursive Neural Network), LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), Bi-directional RNN, Attention RNN, QRNN (Quasi-Recurrent Neural Network). ..
- an identification model different from RNN is SVM (Support Vector Machine).
- SVM Small Vector Machine
- the SVM is used in the above-mentioned Patent Document 1. Therefore, it is conceivable to use a model other than the RNN such as SVM as the identification model 15.
- the RNN is a model in which the output of a certain layer of the intermediate layers from the input value to the output value in the neural network is used as the input of the intermediate layer from the next input value to the output value.
- the neural network has a recursive structure in which the output of the neural network is used as the input of another network.
- the intermediate layers are linearly connected in time series. Therefore, in the RNN, the input value at a certain time affects the output value for the input value after that time. That is, the RNN is a model that performs prediction based on past data, in other words, performs learning based on time series data. This point is different from the non-RNN model such as SVM.
- the determination apparatus 1 inputs the feature vector configured by arranging the RRIs of the identification target period in time series into the identification model 15. Therefore, the input value to the identification model 15 has temporal continuity. Since the continuous RRI is a value that continuously represents the sleep state of the subject, the continuous RRI has a time-series relationship. Therefore, when such feature vectors are used, the inventors have found that RNNs that perform learning based on time-series data are more suitable for continuous biological reactions than models that do not, such as SVMs. We have reached the conclusion that will be output with higher accuracy. Therefore, RNN is used as the identification model 15.
- the identification unit 14 inputs the feature vector for each predetermined period in the sleep period to the identification model 15, and obtains the output of the value for each identification target period from the identification model 15.
- the value for each identification target period is, for example, “1” indicating the apnea state and “0” indicating the normal breathing state.
- the calculation process calculates an index based on the ratio between the period of the apnea and the period of the normal breathing in the sleep period.
- the processing unit 10 functions as the calculation unit 16 that executes the calculation process.
- the index is, for example, the AS (Apnea/Sleep) ratio A, which is the ratio of the apnea period ta to the sleep period t, which is represented by the following equation.
- the period ta of the apnea state is the total of the identification target periods that are identified as the apnea state.
- the index is not limited to the AS ratio, and may be the ratio of the period of the apnea state to the period of the normal breathing state as another example.
- A ta/t ⁇ 100 [%]
- the determination processing it is determined whether or not the subject P is SAS based on the comparison between the index value (AS ratio) A of the calculated index and the threshold TH that can be determined as SAS.
- the processing unit 10 functions as a determination unit 17 that makes this determination.
- the determination unit 17 determines that the subject P is SAS when the AS ratio A is equal to or greater than the threshold value TH (A ⁇ TH). If not (A ⁇ TH), the subject P is determined not to be SAS. By using the AS ratio A, whether or not the subject P has SAS can be easily determined.
- a user of the system 100 attaches the heartbeat measuring device 2 to his/her body, places the determination device 1 at a position communicable with the heartbeat measuring device 2 such as a bedside, and sleeps. Then, the determination device 1 determines whether or not the subject P is SAS by the method shown in the flowchart of FIG.
- the RRI data of the subject P measured by the heartbeat measuring device 2 is input to the determination device 1 (step S101).
- the measurement of the heartbeat by the heartbeat measuring device 2 is performed until the subject P wakes up, and in step S101, RRI data until the subject P wakes up is input.
- the processing unit 10 of the determination device 1 arranges the RRI data for each identification target period in time series to generate a feature vector (step S103). At that time, the processing unit 10 determines whether each value (RRI) indicated by the RRI data is a normal value by comparing it with a threshold value stored in advance (step S1031). Then, the feature vector is generated for all the identification target periods that are normal values. Further, the processing unit 10 normalizes the RRI within the identification target period (step S1033) and generates a feature vector with the normalized RRI.
- the processing unit 10 inputs the feature vector generated for each identification target period into the identification model 15 which is a recursive neural network (step S105).
- the identification model 15 which is a recursive neural network.
- the processing unit refers to the apnea period ta for the entire sleep period t and calculates the AS ratio A which is the ratio to the entire sleep period t (step S107).
- the AS ratio A is equal to or greater than the threshold value TH stored in advance (A ⁇ TH) (YES in step S109)
- the processing unit 10 outputs a determination result that the subject P is SAS (SAS patient) (Ste S111). If not (A ⁇ TH) (NO in step S109), the processing unit 10 outputs a determination result that the subject P is normal (non-SAS patient) (step S113).
- the determination result of step S111 or step S113 may be transmitted to the output device. Accordingly, it is possible to know the determination result of whether the subject P is SAS by the output from the output device such as the display. Alternatively, the determination result may be transmitted from the communication unit 30 to another device. Examples of other devices include a wearable terminal of the subject P, a server device owned by a medical institution, and the like. Thereby, the determination result of whether the subject P is SAS can be known by using another device.
- the inventors conducted an evaluation test using clinical data in order to evaluate the performance of the sleep apnea syndrome determination method according to the above-described embodiment (hereinafter referred to as “proposed method”).
- the evaluation test it was determined whether the SAS was the first method to the fifth method under the determination conditions shown in FIG.
- the first method to the third method are the proposed methods, and the fourth method and the fifth method are the determination methods as comparative examples used as evaluation criteria.
- input value indicates the value used for input to the identification model
- identification model indicates the identification model used for identifying the sleep state for a predetermined period.
- the “input method” indicates how to measure the RRI used for generating the feature vector.
- a feature vector was generated from the RRI data as an “input value” and input to the identification model that is a recursive neural network (RNN).
- the discrimination model of the first method and the third method is the LSTM of the RNN.
- SimpleRNN was used as another example of RNN.
- the feature vector in the first method and the second method is a variable-length vector using continuous RRI obtained from a fixed identification target period. The length of the identification target period was 60 seconds.
- the feature vector in the third method is a fixed length vector using a fixed number of continuous RRIs. The fixed number was 60 immediately before calculation.
- a normal value determination is performed on the RRI, and the feature vector obtained from the RRI in the identification target period including only the RRI that is the normal value is used as the identification model. I input it.
- the fourth method is a determination method adopted in Patent Document 1 described above, which uses a support vector machine (SVM) for distinguishing an apnea state from a normal breathing state when an HRV index is given.
- the fifth method is a method in which the identification model used in the fourth method is changed from SVM to the RNN used in the proposed method.
- the fifth method is also a method in which the input value to the identification model of the first method is changed from RRI to HRV index.
- an HRV index different from RRI was input to the identification model.
- the HRV index input to the discriminant model was all of (1) to (11) above.
- the identification model of the fourth method is SVM different from RNN as shown in Patent Document 1.
- a heartbeat for 3 minutes was measured, and a feature vector composed of a plurality of HRV indexes calculated at each calculation was input to the SVM. At this time, the fourth method does not determine the normal value of the measured heartbeat.
- the continuous HRV index is calculated from the continuous RRI measured in the first fixed period (here, 180 seconds) to obtain the feature vector, and the second was input to the LSTM every fixed period (here, 60 seconds). Also in the fifth method, a normal value determination was performed on the measured RRI, and the HRV index obtained from the RRI in a fixed period consisting of only the normal RRI was input to the discrimination model.
- FIGS. 7 to 11 were obtained by the first method to the fifth method, respectively.
- the left side of each figure shows the AS ratio of each subject at the time of learning the discrimination model, and the right side shows the AS ratio of each subject as a determination result using the learned discrimination model.
- the threshold of the AS ratio between the SAS patient and the non-SAS patient was set at the time of learning the discrimination model. The threshold value was used in the determination process.
- the discriminant model was learned and judged using the electrocardiographic measurement results of the group of subjects whose attributes are shown in Fig. 6.
- the subjects who measured the RRI data used for learning the discriminant model in each determination method were 13 SAS patients and 18 non-SAS patients.
- the subjects who measured the RRI data used for the determination process were 11 SAS patients and 17 non-SAS patients.
- “AHI (Apnea Hypopnea Index)” in FIG. 6 is an index of the severity of SAS, and is defined as the number of occurrences of an apnea condition of 10 seconds or more per hour, and SAS is AHI ⁇ 15. Has been diagnosed.
- the evaluation values shown in FIG. 12 are the sensitivity SE, the specificity SP, the positive reactive intermediate degree PO, and the negative reactive intermediate degree NE, which are represented by the following equations (1) to (4).
- a is the number of SAS patients and “positive” subjects
- b is the number of non-SAS patients and “positive” subjects
- c is the number of SAS patients and “negative” subjects
- D represent the number of non-SAS patients and “negative” subjects.
- “positive” is a subject whose AS ratio is larger than a threshold value (for example, AS ⁇ threshold value TH or AS>threshold value TH), and “negative” is smaller than the threshold value (for example, AS ⁇ threshold value TH or AS ⁇ threshold value TH).
- SE [%] a/(a+c) ⁇ 100 Equation (1)
- SP[%] d/(b+d) ⁇ 100 Equation (2)
- PO [%] a/(a+b) ⁇ 100 Equation (3)
- NE [%] d/(c+d) ⁇ 100 Equation (4)
- the evaluation values of the first method to the third method are higher than those of the fourth method.
- the fourth method is a determination method using a support vector machine for distinguishing an apnea state and a normal breathing state when the HRV index is given in the above Patent Document 1, these first to third methods are used.
- the evaluation values of the first method and the second method exceed 90%, and the accuracy is very high. Further, the first method is extremely high in accuracy because all evaluation values show 100%. That is, the determination accuracy can be extremely improved by using the LSTM of the RNNs as the identification model. Therefore, it was verified that the first method and the second method are significantly more effective than the conventional determination method.
- the third method is a method in which the feature vector generation method in the first method is changed from a variable length vector to a fixed length vector.
- the first method is significantly higher than the third method. Therefore, the determination accuracy is higher when the feature vector is a variable length vector than when it is a fixed length vector. Therefore, it was verified that the variable length vector is more effective than the fixed length vector.
- the evaluation values of the first method and the second method among the first method to the third method are significantly higher than those of the fifth method, and the judgment accuracy is higher than that of the fifth method. Therefore, even if the identification method is changed from SVM to RNN in the fourth method, it is verified that the first method and the second method of the proposed method are significantly effective.
- the first method and the fifth method are different only in that the input to the LSTM, which is the identification model, is the RRI and the HRV index. All the evaluation values of the first method are significantly higher than those of the fifth method, and the determination accuracy is higher than that of the fifth method. Therefore, it was verified that RRI is more effective than the HRV index as the input to the LSTM. Furthermore, since the second method has the same RNN as the identification model and a different Simple RNN from the first method, the determination accuracy is slightly inferior to the first method, but all evaluation values are higher than the fifth method. Therefore, it was verified that the input to the RNN is not limited to the LSTM and the RRI is more effective than the HRV index.
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Cardiology (AREA)
- Animal Behavior & Ethology (AREA)
- Public Health (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Veterinary Medicine (AREA)
- General Health & Medical Sciences (AREA)
- Physiology (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Pulmonology (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
- Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
Description
FV1={925,905,…,885}
FV2={935,…,915}
zi=(xi-μ)/σ
(xiは各RRI、μは平均値、σは標準偏差を表す。)
(1)meanNN:RRIの平均。
(2)SDNN:RRIの標準偏差。
(3)Total Power:RRIの全パワースペクトル。
(4)RMSSD:n番目のRRIとn+1番目のRRIの差の2乗の平均値の平方根。
RRIの変動が多いと値は大きくなり、自律神経系の活動の指標となる。
(5)NN50:n番目とn+1番目のRRIの差異が50ミリ秒を超えた数。普通のRRIの変動は50ミリ秒以下であり、NN50は激しい変動の指標となる。
(6)pNN50:NN50の値の全R波の数に対する割合。NN50は回数なので、一般に窓幅が長ければ大きな値となってしまう。pNN50を考えることで、大きな変動が起こる割合を把握できる。
(7)LF:LF帯(0.04Hz~0.15Hz)の周波数帯のパワースペクトル。LFは主に交感神経系の活動の指標とされる。
(8)HF:HF帯(0.15Hz~0.4Hz)の周波数帯のパワースペクトル。HFは主に副交感神経系の活動の指標とされる。
(9)LF/HF:LF/HFで定義されるLFとHFとの比で、交感神経系と副交感神経系との活動状況の比と考えられる。LF/HFは、交感神経優位で値が大きくなり、副交感神経優位で値が小さくなる。
(10)LFnu:LF/(LF+HF)で定義される補正LF。交感神経活動の変化を強調する。
(11)HFnu:HF/(LF+HF)で定義される補正HF。副交感神経活動の変化を強調する。
A=ta/t×100[%]
SE[%]=a/(a+c)×100 …式(1)
SP[%]=d/(b+d)×100 …式(2)
PO[%]=a/(a+b)×100 …式(3)
NE[%]=d/(c+d)×100 …式(4)
2 :心拍計測器
10 :処理部
11 :生成部
12 :正常値判定部
13 :標準化処理部
14 :識別部
15 :識別モデル
16 :算出部
17 :判定部
20 :記憶装置
21 :コンピュータプログラム
21A :電極
30 :通信部
100 :システム
A :AS比
FV1 :特徴ベクトル
FV2 :特徴ベクトル
I :信号強度
P :被験者
T1 :期間
T2 :期間
TH :閾値
Claims (9)
- 被験者の睡眠期間における心拍間隔(R-R Interval: RRI)を示すRRIデータを用いて前記被験者が睡眠時無呼吸症候群であるか否かを判定する処理部を備え、
前記処理部は、
前記睡眠期間における連続する複数の心拍間隔を含む特徴ベクトルを生成し、
再帰型ニューラルネットワークに、前記特徴ベクトルを入力することによって出力される、入力された特徴ベクトルに含まれる複数の心拍間隔に対応する識別対象期間ごとの値から、前記睡眠期間における無呼吸状態の期間と正常呼吸状態の期間との比率に基づく指標を算出し、
前記指標に基づいて、前記被験者が睡眠時無呼吸症候群であるか否かを判定する、よう動作する
睡眠時無呼吸症候群判定装置。 - 前記再帰型ニューラルネットワークは、特徴ベクトルが入力されたことに応答して前記識別対象期間の睡眠状態を示す値を出力するよう予め機械学習により学習されている
請求項1に記載の睡眠時無呼吸症候群判定装置。 - 前記再帰型ニューラルネットワークはLSTM(Long Short-Term Memory)である
請求項1又は2に記載の睡眠時無呼吸症候群判定装置。 - 前記識別対象期間は固定の期間長を有し、
前記特徴ベクトルは、前記識別対象期間の前記複数の心拍間隔それぞれを示すRRIデータ群の時系列からなる可変長ベクトルである
請求項1~請求項3のいずれか一項に記載の睡眠時無呼吸症候群判定装置。 - 前記処理部は、前記再帰型ニューラルネットワークに前記特徴ベクトルを入力する際に、心拍間隔の異常値を示す閾値より大きい心拍間隔を含む前記複数の心拍間隔から生成された前記特徴ベクトルを入力しない
請求項1~請求項4のいずれか一項に記載の睡眠時無呼吸症候群判定装置。 - 前記特徴ベクトルを生成することは、前記複数の心拍間隔それぞれの値を標準化することを含む
請求項1~請求項5のいずれか一項に記載の睡眠時無呼吸症候群判定装置。 - 前記指標は、前記睡眠期間に対する前記無呼吸状態の期間の比率であり、
前記判定することは、前記比率が睡眠時無呼吸症候群であることを示す閾値より大きい場合に、前記被験者を睡眠時無呼吸症候群であると判定することを含む
請求項1~請求項6のいずれか一項に記載の睡眠時無呼吸症候群判定装置。 - 被験者の睡眠期間における心拍間隔を示すRRIデータを用いて前記被験者が睡眠時無呼吸症候群であるか否かを判定する方法であって、
前記睡眠期間における連続する複数の心拍間隔を含む特徴ベクトルを生成するステップと、
再帰型ニューラルネットワークに、前記特徴ベクトルを入力することによって出力される、入力された特徴ベクトルに含まれる複数の心拍間隔に対応する識別対象期間ごとの値から、前記睡眠期間における無呼吸状態の期間と正常呼吸状態の期間との比率に基づく指標を算出するステップと、
前記指標に基づいて、前記被験者が睡眠時無呼吸症候群であるか否かを判定するステップと、を備える
睡眠時無呼吸症候群判定方法。 - コンピュータに、被験者の睡眠期間における心拍間隔を示すRRIデータを用いて前記被験者が睡眠時無呼吸症候群であるか否かを判定する処理を実行させるプログラムであって、
前記睡眠期間における連続する複数の心拍間隔を含む特徴ベクトルを生成するステップと、
再帰型ニューラルネットワークに、前記特徴ベクトルを入力することによって出力される、入力された特徴ベクトルに含まれる複数の心拍間隔に対応する識別対象期間ごとの値から、前記睡眠期間における無呼吸状態の期間と正常呼吸状態の期間との比率に基づく指標を算出するステップと、
前記指標に基づいて、前記被験者が睡眠時無呼吸症候群であるか否かを判定するステップと、を実行させる
睡眠時無呼吸症候群判定プログラム。
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2020572122A JP7403166B2 (ja) | 2019-02-13 | 2020-01-09 | 睡眠時無呼吸症候群判定装置、睡眠時無呼吸症候群判定装置の動作方法、及び、睡眠時無呼吸症候群判定プログラム |
| US17/430,424 US12310743B2 (en) | 2019-02-13 | 2020-01-09 | Sleep apnea syndrome determination apparatus, sleep apnea syndrome determination method, and sleep apnea syndrome determination program |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2019023217 | 2019-02-13 | ||
| JP2019-023217 | 2019-02-13 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020166239A1 true WO2020166239A1 (ja) | 2020-08-20 |
Family
ID=72044810
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2020/000529 Ceased WO2020166239A1 (ja) | 2019-02-13 | 2020-01-09 | 睡眠時無呼吸症候群判定装置、睡眠時無呼吸症候群判定方法、及び、睡眠時無呼吸症候群判定プログラム |
Country Status (3)
| Country | Link |
|---|---|
| US (1) | US12310743B2 (ja) |
| JP (1) | JP7403166B2 (ja) |
| WO (1) | WO2020166239A1 (ja) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114711725A (zh) * | 2022-04-15 | 2022-07-08 | 华南师范大学 | 一种基于双重注意力机制的睡眠呼吸暂停检测方法及装置 |
| JP2022184311A (ja) * | 2021-06-01 | 2022-12-13 | 学校法人産業医科大学 | 深部体温推定装置および方法並びにプログラム |
| JP2023167351A (ja) * | 2022-05-11 | 2023-11-24 | 京セラ株式会社 | 予測装置及び予測方法 |
| WO2025204774A1 (ja) * | 2024-03-26 | 2025-10-02 | 株式会社東海理化電機製作所 | 心身状態推定装置 |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118986293B (zh) * | 2024-10-24 | 2025-01-24 | 深圳市知谱科技有限公司 | 一种睡眠数据智能采集处理方法及系统 |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2016214491A (ja) * | 2015-05-19 | 2016-12-22 | 国立大学法人京都大学 | 無呼吸識別システム及びコンピュータプログラム |
| WO2018221750A1 (ja) * | 2017-06-02 | 2018-12-06 | 学校法人慶應義塾 | 睡眠判定装置、睡眠判定方法、および、睡眠判定プログラム |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2009150744A1 (ja) * | 2008-06-13 | 2009-12-17 | ハートメトリクス株式会社 | 睡眠状態モニタリング装置、モニタリングシステムおよびコンピュータプログラム |
| JP6869167B2 (ja) * | 2017-11-30 | 2021-05-12 | パラマウントベッド株式会社 | 異常報知装置、プログラム及び異常報知方法 |
| CN109117730B (zh) * | 2018-07-11 | 2021-05-11 | 上海夏先机电科技发展有限公司 | 心电图心房颤动实时判断方法、装置、系统及存储介质 |
| WO2020047750A1 (zh) * | 2018-09-04 | 2020-03-12 | 深圳先进技术研究院 | 心律失常的检测方法、装置、电子设备及计算机存储介质 |
-
2020
- 2020-01-09 WO PCT/JP2020/000529 patent/WO2020166239A1/ja not_active Ceased
- 2020-01-09 US US17/430,424 patent/US12310743B2/en active Active
- 2020-01-09 JP JP2020572122A patent/JP7403166B2/ja active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2016214491A (ja) * | 2015-05-19 | 2016-12-22 | 国立大学法人京都大学 | 無呼吸識別システム及びコンピュータプログラム |
| WO2018221750A1 (ja) * | 2017-06-02 | 2018-12-06 | 学校法人慶應義塾 | 睡眠判定装置、睡眠判定方法、および、睡眠判定プログラム |
Non-Patent Citations (1)
| Title |
|---|
| NAKAYAMA, CHIKAO ET AL.: "Application of suppo vector machine to heart rate during sleep for sleep apnea syndrome screening", LECTURE PROCEEDINGS OF THE ANNUAL CONFERENCE OF THE INSTITUTE OF SYSTEMS, CONTROL AND INFORMATION ENGINEERS, vol. 59, 20 May 2015 (2015-05-20), pages 1 - 6 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2022184311A (ja) * | 2021-06-01 | 2022-12-13 | 学校法人産業医科大学 | 深部体温推定装置および方法並びにプログラム |
| JP7781406B2 (ja) | 2021-06-01 | 2025-12-08 | 百合子 八谷 | 深部体温予測装置および方法並びにプログラム |
| CN114711725A (zh) * | 2022-04-15 | 2022-07-08 | 华南师范大学 | 一种基于双重注意力机制的睡眠呼吸暂停检测方法及装置 |
| JP2023167351A (ja) * | 2022-05-11 | 2023-11-24 | 京セラ株式会社 | 予測装置及び予測方法 |
| WO2025204774A1 (ja) * | 2024-03-26 | 2025-10-02 | 株式会社東海理化電機製作所 | 心身状態推定装置 |
Also Published As
| Publication number | Publication date |
|---|---|
| JPWO2020166239A1 (ja) | 2021-12-16 |
| US12310743B2 (en) | 2025-05-27 |
| JP7403166B2 (ja) | 2023-12-22 |
| US20220125376A1 (en) | 2022-04-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP7403166B2 (ja) | 睡眠時無呼吸症候群判定装置、睡眠時無呼吸症候群判定装置の動作方法、及び、睡眠時無呼吸症候群判定プログラム | |
| US10321871B2 (en) | Determining sleep stages and sleep events using sensor data | |
| JP7191159B2 (ja) | コンピュータプログラム、及び、被験者の感情状態を提供する方法 | |
| US10687757B2 (en) | Psychological acute stress measurement using a wireless sensor | |
| US7621871B2 (en) | Systems and methods for monitoring and evaluating individual performance | |
| EP1711104B1 (en) | Method and apparatus for ecg-derived sleep disordered breathing monitoring, detection and classification | |
| EP2285270B1 (en) | Method and system for determining a physiological condition | |
| CN115802931B (zh) | 检测用户温度和评估呼吸系统病症的生理症状 | |
| JP6304050B2 (ja) | 生体状態推定装置 | |
| US20240415463A1 (en) | Methods, apparatus and systems for adaptable presentation of sensor data | |
| US11075009B2 (en) | System and method for sympathetic and parasympathetic activity monitoring by heartbeat | |
| JP7476508B2 (ja) | 健康状態判定システム、健康状態判定方法及びプログラム | |
| JP6691334B2 (ja) | 睡眠時無呼吸症候群の疑いを検知するためのシステム、処理部及びコンピュータプログラム | |
| CN108577830A (zh) | 一种面向用户的体征信息动态监护方法和动态监护系统 | |
| JP2016016144A (ja) | 生体情報処理システム及び生体情報処理システムの制御方法 | |
| JP2016073527A (ja) | 睡眠状態判定装置、睡眠状態判定方法および睡眠管理システム | |
| US20210259561A1 (en) | Systems and methods for evaluation of health situation or condition | |
| WO2009150765A1 (ja) | 睡眠状態モニタリング装置、モニタリングシステムおよびコンピュータプログラム | |
| EP4327743A1 (en) | Sleeping state estimation system | |
| CN118592907A (zh) | 基于智能戒指实现用户健康监测的方法及系统 | |
| JP7602250B2 (ja) | 居眠り検知装置、検知方法、及びコンピュータプログラム | |
| WO2017180617A1 (en) | Psychological acute stress measurement using a wireless sensor | |
| JP2006271474A (ja) | 睡眠状態推定装置 | |
| JP7521836B2 (ja) | 睡眠質を点数化する装置、方法及びコンピュータプログラム | |
| TWI819792B (zh) | 基於腦波訊號之睡眠異常偵測方法與其裝置 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20755899 Country of ref document: EP Kind code of ref document: A1 |
|
| ENP | Entry into the national phase |
Ref document number: 2020572122 Country of ref document: JP Kind code of ref document: A |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 20755899 Country of ref document: EP Kind code of ref document: A1 |
|
| WWG | Wipo information: grant in national office |
Ref document number: 17430424 Country of ref document: US |