Patent Application
20050020934
The present invention relates to improvement in physiological monitoring, in particular sleep or vigilance monitoring.
Sleep studies currently involve the use of multiple electrodes connected to the subject to record the data necessary to make a diagnosis of a sleep disorder. A polysomnography study will typically require the following electrodes to be attached to the head, but may include many more:
a and 1b of the accompanying drawings show respectively the top and view from underside of a subject's head indicating the standard positions for locating electrodes for EEG monitoring. These locations are identified in
Traditionally the right or left mastoid electrode has been used as a reference electrode for the others, though other “quiet” sites are possible such as the nape, of the neck.
To obtain fill information from the study the subject may also be required to have the following sensors attached:
The data from the electrodes on the head is used to “score” the sleep stages of the patient throughout the recording. The method currently employed world-wide for scoring sleep recording is described in Rechtschaffen and Kales (1968), “A manual of standardized technology, techniques; and scoring systems for sleep stages of human subjects”. These are known as the R & K rules. Sleep scoring breaks the recording into epochs of typically 30 seconds duration and each epoch has a sleep stage classification applied to it. The six recognized classifications are: Stage wake; stage REM (rapid eye movement); stages 1, 2, 3 and 4. The classification of each epoch requires the identification of particular features in the EEG and EOG signals, and measurement of the amplitude of the EMG relative to the background EMG level. The features are identified using frequency and amplitude criteria. A set of rules is then applied to the features to obtain the classification for each epoch. Traditionally such a polysomnographic analysis is performed by a human expert visually inspecting the signal traces.
Examples of conventional EEG traces which have been assigned to: the sleep stages mentioned above are shown in
a—awake;
b—stage 1;
c—stage 2;
d—stage 3;
e—stage 4;
f—REM.
Once each epoch has been assigned a classification, cleanup rules are applied that can reclassify certain epochs according to their context. The classifications of each epoch for the entire night's recording can be plotted against time. This is known as a “hypnogram”. Summary statistics can be derived from the hypnogram that allow objective measures of the quality of sleep to be made.
Thus the traditional methods of polysomnographic, analysis rely on human experts, however there have been proposals for automated analysis of sleep signals.
The article “A likelihood based computer approach to conventional scoring of sleep”, Procs. An. Int. Conf. of the Engineering in Medicine and Biology Society, 1992 Oct 29 to November 1, Vol. 14, pp 2645 to 2646 discloses a method of scoring sleep, using a computer program which divides each 30 second epoch into one second intervals and calculates the likelihood of the signal in each interval matching one of eleven pre-defined features. These likelihoods are then combined into the likelihood of each of eighteen events for the epoch. The signals are the conventional five channel signals: EEG, 2EOG, 2EMG. The R& K rules can then be applied to the events by combining their probabilities using a weighting matrix to assign a sleep stage type to the epoch.
A problem with the techniques mentioned above is that the patient needs many electrodes to be attached to the head. This is inconvenient, particularly where the analysis is being performed on a patient whose sleep is probably poor anyway. Having five or more electrodes attached to their head during sleep is likely to reduce the quality of their sleep even more. Further, similar techniques are used also in vigilance monitoring, which is analogous to sleep monitoring. In vigilance monitoring the same signals are monitored, but rather than sleep stage type, wakefulness types are assigned to each epoch. Thus, for instance, the vigilance of a, person operating machinery, driving or performing some safety-critical task, can be monitored. If the wakefulness stage types indicate a level of vigilance lower than required, then an alarm may be triggered. Again, the need to use five or more electrodes attached to the subject may hinder the activity of the subject. Further, movement of the subject in performing their normal activities can disturb the electrodes and cause noise in the signals which confuses the analysis.
EP-A-0773504 discloses a method and apparatus for sleep and vigilance monitoring in which a trained neural network is used to analyse the signal from the mastoid, sites of a subject's head. It can generate from this signal a hypnogram or series of wakefulness stage types, and can generate from these a summary of the quality of sleep, or trigger a vigilance alarm depending on the application. The use of only the mastoid sites is advantageous in reducing the number of electrodes which have to be attached to the subject. Analysis of the signals from the mastoid sites allow, in the case of sleep, the characterisation of each epoch into one of the following three sleep stage types: Wake, REM/light or deep. Thus it does not provide a characterization of all six sleep stage types, but nevertheless requires the use of fewer electrodes, and provides for automated real-time analysis.
According to the present invention there is provided sleep monitoring apparatus comprising one or more electrodes to obtain an electro-oculogram (EOG) signal from a subject over a period of epochs, the signal being related to the sleep stage type being experienced by the subject; and a processor adapted to analyse the electro-oculogram signal and assign a sleep stage type to each epoch.
The invention also provides in a corresponding way, vigilance monitoring apparatus comprising one or more electrodes to obtain an electro-oculogram (EOG) signal from a subject over a period of epochs, the signal being related to the wakefulness stage type being experienced by the subject; and a processor adapted to analyse the electro-oculogram signal and assign a wakefulness stage type to each epoch.
In particular the present invention provides for distinguishing between the different stages including between REM and light sleep by use of the EOG signal. Preferably the EOG signal alone is analysed, this involving use of one EOG electrode (attached to the left or right EOG site) and a reference electrode. The reference electrode may be connected to the nape of the neck, or one of the mastoid sites. It is particularly advantageous for the reference site to be the mastoid site neighbouring (ie on the same side of the head) as the EQG site. This allows a very convenient siting of electrodes.
Preferably the invention provides for the characterisation of each epoch as one of at least four types, namely wake, REM sleep, light sleep, or deep sleep, on the basis of the EOG signal alone.
In the case of sleep monitoring the process may be adapted to generate a hypnogram from the assigned sleep stage types, and to analyse the hypnogram to generate a summary index of sleep quality over the period of epochs. This summary index may be displayed on a display.
In the case of vigilance monitoring the apparatus may include a wakefulness stage type monitor for monitoring the assigned wakefulness stage types to determine whether they meet predetermined criteria which represent a lowered level of vigilance. The apparatus may also include a message generator responsive to the wakefulness stage type monitor to generate a message, such as an alarm, when the predetermined criteria are met.
Preferably the process comprises a trained artificial neural network. Preferably the apparatus is a single portable unit.
The process may be further adapted to analyse the EOG signal to derive a heart rate measurement therefrom. This heart rate measurement may be recorded and optionally displayed.
The invention will be further described by way of example with reference to the accompanying drawings in which:—
a and 1b illustrate the standard sites for positioning electrodes on the head of a patient or subject;
a to 2f illustrate EEG signals and assigned sleep stage types;
A block diagram of a typical implementation of the device according to an embodiment of the invention is described below, which refers to
Signal are acquired from electrodes 1 mounted on the subject's head via switching circuitry 2 and input amplifier 3. The electrodes 1a, 1b and, where used, the indifferent electrode may be wireless electrodes thus making use of the device less obtrusive to the subject. The input amplifier has an analog bandwidth of at least 0.5 Hertz to 30 Hertz and is of a high gain, low noise instrumentation design. The signal is input to low-pass filter 21 to reduce unwanted aliasing components before analog to digital conversion using sample-and-hold circuit 22 and A/D converter 4. The resultant quantised data samples are transferred to low power microcontroller 5 for processing. In one example the sampling rate may be 128 Hertz and the quantisation of the analog to digital converter 4 be 12 bits, which provides sufficient dynamic range not to require a gain control on the input amplifier 3.
When recording one channel of EOG from the EOG site, three electrodes are typically necessary; two of them (1a, 1b) comprise the differential inputs to input amplifier 3 (one electrode being connected to an EOG site and one to a reference site such as the nearby mastoid site). The third is an “indifferent” lead (not shown) whose sole function is to allow input amplifier return currents to flow. The indifferent lead can be attached to any part of the subject's body. The indifferent lead is optional, it is possible to omit it.
Before signal acquisition begins, the impedances of the electrodes on the subject's head may be measured by causing impedance measurement circuitry 10 to drive a signal of known amplitude and source impedance via the switching circuitry 2 through the electrodes 1a, 1b in turn onto the subject's scalp. The resultant signal is measured by the microcontroller by the process described above and from it the impedances of the electrodes fitted to the subject's scalp are calculated in turn. A warning message is displayed on the LCD display 11 if the impedance of the connections to the subject's head is unacceptably high.
During data acquisition, the device continually acquires signals from the subject's head for analysis. The microcontroller 5 analyses the quantised values and from them generates results that are stored in non-volatile data memory 7. The program for the microcontroller is held in program memory 6. Real time clock 8 allows the results to be stored with a record of the time of day of acquisition.
Watchdog 20 resets the microcontroller 5 if the microcontroller fails to write to it periodically. If the microcontroller 5 is reset, it will identify whether it was in record before the reset was received and if so, go back into record so that a minimal amount of data is lost.
Alarm 61 may be activated if the device is a vigilance monitor and the processor 5 has determined that the level of vigilance of the subject is unacceptably low.
Control of the device is via switches 13, some of which can be read by the microcontroller 5. On/Off switch 14 turns the device on and off; select switch 15 displays successive prompts and results of the LCD screen 11; enter switch 16 accepts the command-currently displayed on the LCD screen 11; and record switch 17 puts the device into record which starts signal acquisition, processing and storage.
When the device is switched on the user can choose whether to view the results from a previous recording; down-load results from the previous recording into a computer or to a printer; delete the results from a previous recording; or record new data.
The low-pass filter 21 can be either part of the input amplifier 3 or can be achieved in the sampling process itself within the A/D converter 4. The sleep summary index, or other results of the lysis; can be displayed on the LCD display 11, though other types of display are possible. Further, data may be downloaded via the RS232 connection.
It will be appreciated that the above describes a specific embodiment and that variations are possible. For instance, the device may have only the means necessary to record results, with all analysis and display carried out externally.
The signals from the electrodes are processed by microprocessor 5 using a, neural network such as a multi-level perceptron (MLP). The neural network mar be trained in the same manner as described in EP-A-0773504 (which is hereby incorporated by reference) and the assignment to each epoch of a sleep or wake stage type is also carried but in the same way as described in BP-A-0773504. In EP-A-0773504 the neural network is trained to assign sleep stage and wake stage types on the basis of signal from the mastoid site. It is found that a neural network, trained on a central channel signal can be signal, can be used to assign sleep stage and wakefulness stage types on the EOG signal in accordance with the present invention without retraining However, it is also possible to train the neural network on the basis of the EOG signal directly in an analogous way to the training on the mastoid site in EP-A-0773504.
Once put into record at step 50, after checking whether the recording must continue at step 51 the microcontroller acquires data from the EOG site at step 52 until it has one second's worth of data at 53. The data is filtered at step 54 to separate the eye movement and brain activity components of the signal. Eye movements are identified within the eye movement channel at step 55. The brain activity data is processed at step 56 using a frequency domain representation to identify the dominant frequencies. There are other ways of finding the dominant frequencies. The coefficients identifying the dominant frequencies are fed at step 57 into a neural network which, in the same way as in EP-A-0773504, generates probabilities that the subject is in one of a number of sleep stages.
If the probabilities generated by the neural network indicate a high probability of sleep stage 1 or stage REM at step 58 then the eye movement data is used to distinguish between them at step 67. The distinction between stage 1 and REM sleep is possible because of the use of the EOG signal. This signal can be used to distinguish between these two types because when the subject is in REM sleep it carries low frequency eye movement structure not present during stage 1 sleep.
The probabilities of each sleep stage are combined in step 68 to form a measure of the depth of sleep. These results are then stored at step 59 and they may be displayed in any desired fashion. In the case of a vigilance monitor, the vigilance level is monitored at step 64, and if it drops below a predetermined threshold, a warning message is generated at 65. This may be used to alarm the subject, or may be stored for a later analysis.
As mentioned, above, it is also possible for the microprocessor to obtain from the signals a measurement of the heart rate because there is a breakthrough into the EOG channel of an ECG signal.
The fact that the device is compact and simple, and uses only a few electrodes, means that the accuracy of diagnosis of sleep disorders is improved because the device itself does not interfere so much with the patient's sleep. Further, it allows a first screening of patients with sleep disorders to be undertaken before the patients are referred to a specialist practitioner or sleep clinic.
The quality of sleep is also of considerable interest to elite sportsmen and women and again the ease of use of the device means that sleep monitoring is a realistic possibility for such people.
In a further embodiment the device may be used as an alarm clock which is sensitive to the quality of sleep which has been achieved and the sleep stage of the subject. Thus the device can wake the subject up after a certain number of hours of good quality sleep or alternatively wake them only during a period of light sleep or REM after a certain number of hours in bed.
As indicated above the invention is also applicable to vigilance monitoring. Vigilance research involves determining when an awake subject, normally with open eyes, starts to move from their wakeful state towards light sleep with a loss of vigilance. Many researchers have studied vigilance but no systems for alerting a loss of vigilance are in common use. The difficulty with, using EEG data to score vigilance levels is that some people can operate adequately when they have, significant quantities of theta activity (as shown in
The best approach for screening the vigilance level of an operator is to take video recordings of the operator's face, and in particular their eyes, and to have this recording scored by an expert examiner. The examiner is trained to look for signs of fatigue on the recording. These signs are reflected by behavioural patterns such as “droopy eyes”, “slow blinks”, “head nods”, “eyes shut for seconds” and “yawns”. The main drawback of a video system is that when the operator turns their head, the signal is lost. Some investigators have tried to circumvent this issue by mounting the video camera on the operator's head although this is not a particularly practical solution.
Using the same (preferably single) EOG channel, as above for distinguishing between REM and Light Sleep, low frequency eye movement data will be produced on the EOG signals when the eyes are moving and the eyelids are open. Signals are also detected when the eyelids open and close. The neural network is trained to distinguish between a ‘fast’ blink or movement of the eyes and signs of fatigue such as a slower-closing of the eyes or rolling of the eyes, or long periods of no eye movement if the operator is staring into the distance, based on low frequency eye movement data on the EOG trace.
This low frequency eye movement data is used in conjunction with the standard brain wave activity on the EOG signal to produce an accurate vigilance monitor. The data may be processed by a neural network. Because of subject-to-subject differences in vigilance levels the training of the neural network may be subject specific where necessary.
| Number | Date | Country | Kind |
|---|---|---|---|
| 01262864 | Nov 2001 | GB | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/GB02/04874 | 10/30/2002 | WO |
