SLEEP QUALITY INDICATORS

Abstract

A method for diagnosis includes acquiring a physiological signal from a patient (22) during a period of sleep, and segmenting the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum. Respective levels of membership of the segments in a plurality of frequency states are computed responsively to the respective frequency spectrum. Based on the respective levels of membership, a sleep quality indicator is determined and displayed, responsively to a statistical characteristic of the segments.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims the benefit of U.S. Provisional Patent Application 60/590,375, filed Jul. 21, 2004. It is a continuation-in-part of U.S. patent application Ser. No. 10/678,773, filed Oct. 3, 2003 (published as US 2004/0230105 A1), and is also related to co-pending U.S. patent application Ser. No. 10/677,176, filed Oct. 2, 2003 (published as US 2004/0073098 A1). The disclosures of all these related applications are incorporated herein by reference.

FIELD OF THE INVENTION

The present invention relates generally to physiological monitoring and diagnosis, and specifically to sleep recording and analysis.

BACKGROUND OF THE INVENTION

Human sleep is generally described as a succession of five recurring stages (plus waking, which is sometimes classified as a sixth stage). Sleep stages are typically monitored using a polysomnograph to collect physiological signals from the sleeping subject, including brain waves (EEG), eye movements (EOG), muscle activity (EMG), heartbeat (ECG), blood oxygen levels (SpO2) and respiration. The commonly-recognized stages include:

• • Stage 1 sleep, or drowsiness. The eyes are closed during Stage 1 sleep, but if aroused from it, a person may feel as if he or she has not slept.
  • Stage 2 is a period of light sleep, during which the body prepares to enter deep sleep.
  • Stages 3 and 4 are deep sleep stages, with Stage 4 being more intense than Stage 3.
  • Stage 5, REM (rapid eye movement) sleep, is distinguishable from non-REM (NREM) sleep by changes in physiological states, including its characteristic rapid eye movements.
    Polysomnograms show brain wave patterns in REM to be similar to Stage 1 sleep. In normal sleep, heart rate and respiration speed up and become erratic, while the muscles may twitch. Intense dreaming occurs during REM sleep, but paralysis occurs simultaneously in the major voluntary muscle groups.

Although sleep staging is most often performed by a human operator, who reads and scores the polysomnogram, there are also methods known in the art for computerized sleep staging. Penzel et al review such methods in “Computer Based Sleep Recording and Analysis,” Sleep Medicine Reviews 4:2 (2000), pages 131-148, which is incorporated herein by reference.

SUMMARY OF THE INVENTION

Although sleep staging is widely accepted as the standard method for diagnosis and classification of sleep disorders, this method provides only coarse resolution and fails to exploit the wealth of information in the polysomnogram signals. The inventors have found many cases in which traditional sleep stage analysis fails to uncover underlying sleep pathologies. In response to the shortcomings of conventional methods, the inventors have developed a family of sleep quality indicators, which assist the diagnostician in recognizing sleep-related disorders.

In some embodiments of the present invention, a sleep analysis system acquires physiological signals, such as EEG signals, during sleep, and adaptively segments the signals to identify quasi-stationary segments. The system automatically analyzes each segment to determine the relative energy in each of a number of frequency bands, and thus assigns the segments to different frequency states. Typically, the states are defined for each patient by fuzzy clustering of features extracted from each segments; and each segment is assigned a degree of membership with respect to each of the states. Based on the fuzzy clustering and membership levels, the system determines and displays sleep quality indicators relating to the distribution of the segments among the clusters.

In some embodiments, the system displays the results of the analysis so that changes in the distribution of states over time, in the course of a period of sleep, can be readily visualized by a caregiver, such as a medical sleep specialist. Additionally or alternatively, the system may display characteristic patterns of transition between different states.

Further additionally or alternatively, the system may calculate the fundamental frequency of each segment, typically expressed as the moment of the EEG power spectrum. The fundamental frequency is then displayed so as to enable the caregiver to visualize changes in the trend and standard deviation of the fundamental frequency, which are indicative of continuous changes in the patient's sleep quality.

Although the embodiments described herein relate mainly to analysis and visualization of EEG signals, the principles of the present invention may similarly be applied to polysomnogram signals of other types, such as respiration and ECG signals.

There is therefore provided, in accordance with an embodiment of the present invention, a method for diagnosis, including:

acquiring a physiological signal from a patient during a period of sleep;

segmenting the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum;

computing respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum; and

displaying a plot indicative of the levels of membership of the segments in the sequence over time.

In disclosed embodiments, computing the respective levels includes applying fuzzy clustering to the segments so as to define the states.

In one embodiment, displaying the plot includes displaying a density plot, in which the levels of membership are represented by color variations. In another embodiment, displaying the plot includes displaying an accumulation plot, showing cumulative levels of membership of the segments in the plurality of frequency states over the sequence. In yet another embodiment, displaying the plot includes displaying an accumulation plot showing cumulative durations of the segments in each of the plurality of frequency states. Typically, the method includes determining and comparing respective accumulation rates of the cumulative durations in at least two of the frequency states.

In some embodiments, displaying the plot includes assigning each of at least some of the segments to one of a waking state and a sleep state responsively to the frequency spectrum, and displaying an accumulation plot showing a cumulative assignment of the segments to the waking and sleep states over time.

There is also provided, in accordance with an embodiment of the present invention, a method for diagnosis, including:

acquiring a physiological signal from a patient during a period of sleep;

segmenting the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum;

computing a fundamental frequency of each segment in the time sequence responsively to a moment of the respective frequency spectrum of the segment; and

displaying a plot showing the fundamental frequency of the segments in the sequence over time.

Displaying the plot may include showing at least one of a trend and a variance of the fundamental frequency.

There is additionally provided, in accordance with an embodiment of the present invention, a method for diagnosis, including:

acquiring a physiological signal from a patient during a period of sleep;

segmenting the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum;

computing respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum;

based on the respective levels of membership, determining a sleep quality indicator responsively to a statistical characteristic of the segments; and

displaying the sleep quality indicator.

In disclosed embodiments, the statistical characteristic includes at least one of:

a cumulative duration of the segments associated with each of the frequency clusters;

a relative duration of the segments associated with each of the frequency clusters;

a mean duration of the segments associated with each of the frequency clusters;

a variance of a duration of the segments associated with each of the frequency clusters;

a total number of the segments associated with each of the frequency clusters; and

a relative duration of the segments associated with each of the frequency clusters.

In some embodiments, the method includes assigning the segments to predefined sleep stages responsively to the frequency spectrum, and determining the sleep quality indicator includes computing the statistical characteristic with respect to each of the sleep stages.

In one embodiment, displaying the sleep quality indicator includes displaying a plot indicative of the levels of membership of the segments in the sequence over time. In another embodiment, displaying the sleep quality indicator includes displaying a plot showing a fundamental frequency of the segments in the sequence over time. In yet another embodiment, computing the respective levels of membership includes assigning the segments in the time sequence to respective frequency states, and determining the sleep quality indicator includes computing probabilities of transition among the frequency states.

In some embodiments, the physiological signal includes an electroencephalogram (EEG) signal. Optionally, the method includes identifying transient phenomena in the EEG signal, and computing an index quantifying a frequency of occurrence of the transient phenomena. The transient phenomena may include one or more of K-complexes and spindles.

Additionally or alternatively, the physiological signal may include a respiration signal. In one embodiment, the method includes identifying respiratory events occurring during the period of sleep, and computing statistical characteristics of the respiratory events. Typically, computing the statistical characteristics includes computing and displaying a respiratory event histogram.

In another embodiment, the method includes measuring a heart rate of the patient, and computing the statistical characteristics includes computing a relative heart rate index indicative of changes in the heart rate associated with the respiratory events.

In yet another embodiment, computing the statistical characteristics includes assigning respective confidence levels to the respiratory events, and displaying the confidence levels as a function of respiration state.

There is further provided, in accordance with an embodiment of the present invention, a method for diagnosis, including:

acquiring a physiological signal from a patient during a period of sleep;

segmenting the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum;

computing respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum;

responsively to the respective levels of membership, assigning each of at least some of the segments to one of a waking state and a sleep state responsively to the frequency spectrum; and

displaying an accumulation plot showing a cumulative assignment of the segments to the waking and sleep states over time.

The method may include determining and comparing respective accumulation rates of the waking and sleep states.

There is moreover provided, in accordance with an embodiment of the present invention, diagnostic apparatus, including a sensor, which is adapted to acquire a physiological signal from a patient during a period of sleep, and a diagnostic processor, which is adapted to carry out the functions described above. In some embodiments, the sensor includes at least one electrode, and the physiological signal includes an electroencephalogram (EEG) signal. Additionally or alternatively, the sensor may include a respiration sensor and/or a heart rate sensor.

There is furthermore provided, in accordance with an embodiment of the present invention, a computer software product, including a computer-readable medium in which program instructions are stored, which instructions, when read by a computer, cause the computer to carry out the functions described above.

The present invention will be more fully understood from the following detailed description of the embodiments thereof, taken together with the drawings in which:

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic, pictorial illustration of a system for polysomnography, in accordance with an embodiment of the present invention;

FIG. 2 is a flow chart that schematically illustrates a method for determining sleep quality parameters, in accordance with an embodiment of the present invention;

FIG. 3 is a three-dimensional plot showing clusters of EEG frequency states, in accordance with an embodiment of the present invention;

FIG. 4A is a hypnogram, showing classification of sleep stages of a patient over time, in accordance with an embodiment of the present invention;

FIG. 4B is a density plot showing a distribution of frequency cluster membership of successive segments of an EEG signal as a function of time for the patient of FIG. 4A, in accordance with an embodiment of the present invention;

FIG. 5A is a hypnogram, showing classification of sleep stages of another patient over time, in accordance with an embodiment of the present invention;

FIG. 5B is a density plot showing a distribution of frequency cluster membership of successive segments of an EEG signal as a function of time for the patient of FIG. 5A, in accordance with an embodiment of the present invention;

FIGS. 6A and 6B are plots showing variations in the fundamental frequency of EEG signals over time, in accordance with an embodiment of the present invention;

FIG. 7 is a frequency state accumulation plot, showing cumulative frequency state durations of successive segments of an EEG signal over time, in accordance with an embodiment of the present invention;

FIG. 8 is a sleep/wake state accumulation plot, showing cumulative durations of sleep and wake states of a patient over time, in accordance with an embodiment of the present invention; and

FIG. 9 is a transition matrix showing probabilities of transitions among frequency states in successive segments of an EEG signal, in accordance with an embodiment of the present invention.

DETAILED DESCRIPTION OF EMBODIMENTS

FIG. 1 is a schematic, pictorial illustration of a system 20 for sleep monitoring and diagnosis, in accordance with an embodiment of the present invention. In this embodiment, system 20 is used to monitor a patient 22 in a home or hospital ward environment, although the principles of the present invention may similarly be applied in dedicated sleep laboratories. System 20 receives and analyzes physiological signals generated by the patient's body, including an EEG signal measured by scalp electrodes 23, an ECG signal measured by skin electrodes 24, and a respiration signal measured by a respiration sensor 26. (As shown in the figure, respiration sensor 26 makes electrical measurements of thoracic and abdominal movement. Additionally or alternatively, air flow measurement may be used for respiration sensing.) Alternatively, only a subset of the above electrodes and sensors may be used, and/or other sensors may be added, such as an EMG or SpO2 sensor, as are known in the art.

The signals from electrodes 23, 24 and sensor 26 are collected, amplified and digitized by a console 28. Console 28 may process and analyze the signals locally, using the methods described hereinbelow. Alternatively or additionally, console 28 may be coupled to communicate over a network 30, such as a telephone network or the Internet, with a diagnostic processor 32. This configuration permits sleep studies to be performed simultaneously in multiple different locations. Processor 32 typically comprises a general-purpose computer with suitable software for carrying out the functions described herein. This software may be downloaded to processor 32 in electronic form, or it may alternatively be provided on tangible media, such as optical, magnetic or non-volatile electronic memory. Processor 32 analyzes the signals conveyed by console 28 in order to identify sleep states of patient 22 and to extract sleep quality indicators. The results of the analysis are presented to an operator 34, such as a physician, on an output device 36, such as a display or printer.

FIG. 2 is a flow chart that schematically illustrates a method for determining sleep quality parameters, in accordance with an embodiment of the present invention. The method and examples of sleep quality indicators given below relate mainly to processing of EEG signals. The principles of this method and the types of indicators derived therefrom, however, may similarly be applied to other sorts of signals, such as respiration and ECG signals.

Processor 32 acquires an EEG signal from patient 22, at a signal acquisition step 40. Typically, for sleep studies, the signal is acquired over the course of at least several hours. The processor then adaptively segments the signal into quasi-stationary segments, at a segmentation step 42. Adaptive segmentation is described at length in the above-mentioned U.S. patent applications. Briefly, processor 32 advances a sliding window, of variable size, through the EEG signal and evaluates statistical features of the signal within the window. The statistical features typically include aspects of the frequency spectrum of each segment, which are determined by methods of spectral analysis known in the art. The processor optimizes the window boundaries so as to envelope a segment that is statistically stationary to within a predefined bound. In consequence, the EEG signal is divided into a time sequence of quasi-stationary segments of varying length, separated by shorter transient periods. This sort of adaptive segmentation is advantageous in that the segments that are chosen represent actual physiological states of the patient, as opposed to the arbitrary 30-second epochs that are used in conventional sleep scoring.

EEG signals normally comprise five major types of waves: (1) δ-wave (1.0-3.5 Hz), (2) θ-wave (4.0-7.0 Hz), (3) α-wave (7.5-12 Hz), (4) σ-wave (12-15 Hz); and (5) β-wave (15-35 Hz). Each quasi-stationary segment typically comprises one dominant wave and possibly other frequency components superimposed on the dominant wave. The frequency composition of the different types of segments determined at step 42 typically varies from patient to patient. Therefore, in order to classify the segments for each individual patient, processor 32 applies a fuzzy clustering algorithm to divide the segments into clusters, at a clustering step 44. Each cluster has a characteristic distribution of features, such as frequency components and overall segment energy. Methods of fuzzy clustering are likewise described in the above-mentioned patent applications.

FIG. 3 is a three-dimensional plot showing clustering of EEG frequency states, in accordance with an embodiment of the present invention. In this figure, the dots represent individual segments of an EEG signal, plotted on three axes corresponding to the following segment features: 1) Relative energy in the delta frequency band; 2) Relative energy in the alpha, sigma and beta frequency bands; and 3) Total segment energy, normalized to a scale of 0-100. Details of the features and clustering scheme are presented below in Appendix A. The inventors have found that this set of features provides useful differentiation between sleep states, but other sets of features, in two, three, or more dimensions may similarly be used for clustering purposes.

In the representation of FIG. 3, four clusters of segments can be identified: a high-frequency (HF) cluster 60, a low-energy mixed-frequency cluster 62 (MF1), a high-energy mixed frequency cluster 64 (MF2), and a low-frequency cluster 66 (LF). These clusters have been found to correlate respectively with deep sleep (stages 3 and 4), moderate sleep (stage 2), light sleep (stage 1/REM), and wakefulness. Alternatively, other clustering models may be used, particularly in conjunction with other feature axes. In any case, the bounds of each cluster are determined adaptively for each patient at step 44.

Returning now to FIG. 2, for each of the segments found at step 42, processor 32 computes membership levels with respect to each of the frequency states, at a membership computation step 46. In other words, rather than just assigning each segment to a respective cluster, the processor determines the similarity of each segment to each of the clusters found at step 44. The membership level of a given segment n having a feature vector x_n may be computed relative to the center vectors μ_k of the different clusters. The membership level w_kⁿ(x) of the segment in the k^th cluster is then calculated as follows:

$w_k^n\left(x\right)=\frac{D\left(x_n,{\mu }_k\right)}{\sum _{k=1}^KD\left(x_n,{\mu }_k\right)},$

wherein K is the number of clusters and D is a scalar function of distance between x_n and μ_k. For example, D can be an Euclidian distance, given by D(x_n, μ_k)=(x_n−μ_k)^H(x_n−μ_k), wherein H denotes the conjugate transpose operator. Alternatively, other methods known in the art may be used for computing cluster membership. The membership levels may be advantageously displayed as a function of time, as illustrated below in FIG. 4B.

The membership values determined at step 46 may be used by processor 32 in automatically assigning each 30-sec epoch during the monitoring period to one of the accepted sleep stages, at a sleep staging step 48. For example, the following scheme may be used, combining the states of the segments in the EEG signal with additional information from EMG and EOG signals:

1) Stage wake—Epochs more than 50% of whose duration are occupied by high-frequency EEG and/or body movements and/or eye blinks are classified as stage wake. Epochs that are not classified as stage wake are classified as sleep.
2) Sleep stages 2-4—Epochs classified as sleep, in which:

• • Predominant high-energy mixed-frequency activity is present, or
  • 1% of the epoch duration is occupied by K-complex activity (low-frequency transients), or
  • More than 15% of the epoch duration is occupied by low-frequency activity.
  • a) Sleep stage 2—Epochs classified at step (2) as stages 2-4, less than 15% of whose duration is occupied by low-frequency activity.
  • b) Stage 3—Epochs classified as stages 2-4, 15-45% of whose duration is occupied by low-frequency activity.
  • c) Stage 4—Epochs classified as stages 2-4, more than 45% of whose duration is occupied by low-frequency activity.
    3) Sleep stage 1+REM—Epochs classified as sleep, which have predominant low-energy mixed-frequency activity.
  • a) REM—Epochs classified as sleep stage 1+REM in which rapid eye movements or low-energy EMG is detected.
  • b) Sleep stage 1—Epochs classified as sleep stage 1+REM and not classified as stage REM, or epochs classified as sleep and not classified as any other sleep stage.

Alternatively or additionally, sleep stages may be determined using cardiovascular, respiratory or other physiological indicators. For example, a method for sleep staging based on cardio-respiratory signals is described in U.S. patent application Ser. No. 10/995,817, filed Nov. 22, 2004, whose disclosure is incorporated herein by reference.

In addition to or instead of standard sleep staging at step 48, processor 32 typically computes sleep quality indicators, at a sleep quality assessment step 50. For example, for each of clusters 60, 62, 64 and 66 (referred to respectively as HF, MF1, MF2 and LF states), the processor may compute the following sleep quality parameters:

• • Total (cumulative) duration (sec) of segments belonging to the state.
  • Relative total duration (%) of segments belonging to the state, compared to the total time monitored.
  • Mean duration (sec) of segments belonging to the state.
  • Standard deviation (variance) of duration of segments belonging to the state.
  • Total number of segments belonging to the state.
  • Relative total number (%) of segments belonging to the state compared to the total number of segments.
    The sleep quality parameters may be computed over all the quasi-stationary segments identified at step 42, or alternatively over a selected sequence of the segments.

Furthermore, processor 32 may combine the segmentation data with the sleep staging performed at step 48 in order to compute the above parameters separately for each identified sleep stage or group of sleep stages. For example, the relative duration of the HF state in REM may be calculated as follows:

$\frac{\mathrm{Duration}\mathrm{of}\mathrm{HF}\mathrm{in}\mathrm{REM}}{\mathrm{Total}\mathrm{REM}\mathrm{duration}}\times 100$

As another example, the relative number of HF segments in REM may be calculated as follows:

$\frac{\mathrm{Number}\mathrm{of}\mathrm{HF}\mathrm{segments}\mathrm{in}\mathrm{REM}}{\mathrm{Total}\mathrm{number}\mathrm{of}\mathrm{segments}\mathrm{in}\mathrm{REM}}\times 100$

References is now made to FIGS. 4A and 4B, which show processed results of EEG measurements made in system 20, in accordance with an embodiment of the present invention. FIG. 4A is a hypnogram, showing sleep stages of the patient over time, as derived from the polysomnogram signals at step 48. FIG. 4B is a density plot showing the distribution of membership of the sequence of EEG segments in each of the four states defined above (HF, MF1, MF2 and LF).

The term “density plot” is used herein to denote a plot in which the color at a given point is indicative of the relative value of a parameter referred to the Cartesian coordinates of the point. In other words, as can be seen in FIG. 4B, for each point in time along the horizontal axis (corresponding to the segment of the EEG signal occurring at that time), four density values are arrayed vertically, corresponding to the degree of membership of the segment in each of clusters HF, MF1, MF2 and LF, which are arrayed along the vertical axis. A density scale 70 at the bottom of the figure shows the correspondence between colors and normalized membership values. (“Color” in this context includes shades of gray.) Although the display shown in FIG. 4B uses gray-scale density values, different colors may equivalently be used to represent the membership values. Thus, the degree of membership at each point in the plot may be equivalently represented by a scale of varying hue, intensity or saturation, or a combination of these factors. All such alternative types of density plots are considered to be within the scope of the present invention.

It can be seen in FIG. 4B that the density plot correlates with the sleep stages shown in FIG. 4A, but contains much richer information about the EEG activity occurring at many points during the sleep period. The inventors have found that the information contained in the density plot (which is lost in the discrete hypnogram) permits the caregiver to recognize abnormal sleep patterns that would otherwise go unnoticed. For example, in one clinical study, the inventors identified a group of patients whose hypnograms appeared to be normal, but who showed relatively high levels of HF membership during sleep. This result is indicative of sleep fragmentation, i.e., poor sleep quality in this group.

FIGS. 5A and 5B, respectively, show a hypnogram and a density plot for another patient, who was known to suffer from a sleep disorder. During the monitoring period, a sleeping drug was administered to the patient, in an attempt to induce deep sleep. In the hypnogram, it appears that drug was ineffective, since the patient's sleep stage never drops below stage 2. In the density plot in FIG. 5B, however, a period of low-frequency activity at around 1 AM demonstrates the short-term efficacy of the drug.

FIG. 6A is a plot showing variations in the fundamental frequency of an EEG signal over time, in accordance with an embodiment of the present invention. This plot was derived from the EEG signal of the patient whose hypnogram and density plot are shown in FIGS. 4A and 4B. The fundamental frequency is determined for each segment by taking the moment of the frequency spectrum of the segment, as shown in Appendix A. The solid line in the figure shows the fundamental frequency value, while the dotted marks above the solid line show the variance. A line at 4 Hz shows the approximate boundary between deep sleep and other sleep stages.

As in the case of the density plot shown above, the fundamental frequency correlates well with the hypnogram sleep stages, but provides richer information that is lost in the discrete hypnogram. This information may be further brought out, for example, by displaying a trend line and a range of standard deviation of the fundamental frequency over time (omitted from FIG. 6A for the sake of simplicity). It will be observed in FIG. 6A, for instance, that at some points changes in frequency are precipitous, while other changes are more gradual. These variations in slope, which are lost for the most part in the hypnogram, can be useful in assessment of clinical factors such as drug effects. The fundamental frequency plot also permits the caregiver to observe local variability, even when the frequency trend (and hence the sleep stage) is flat. In this regard, note the difference between the smooth fundamental frequency plot in the neighborhood of 1 AM and the highly-variable plot at around 2 AM.

Like other sleep quality indicators, the fundamental frequency may be correlated with the patient's sleep stages. For example, processor 32 may calculate the average fundamental frequency, and possibly the variance of the fundamental frequency, over each of the sleep stages identified at step 48.

FIG. 6B is a plot of fundamental frequency taken from the EEG signal of the patient whose hypnogram and density plot are shown in FIGS. 5A and 5B. For patients with sleep disorders, the fundamental frequency drops below the 4 Hz threshold only occasionally, if at all. In the case shown in FIG. 6B, the effect of sleeping drug administration can be seen in the period of deep sleep at about 1 AM following administration of the drug, despite the negative hypnogram findings. The precipitous frequency drop at 1 AM is followed by shallower, more gradual drops thereafter, reflecting the cyclical interaction of the drug with the sleep states of the brain.

FIG. 7 is a frequency state accumulation plot, showing cumulative duration of successive segments of an EEG signal over time, in accordance with an embodiment of the present invention. Curves 80, 82, 84 and 86 respectively show the cumulative durations of HF, MF2, MF1 and LF sleep states, as a fraction of the total sleep period. The accumulation function Ac(t,s) for state s at time t is given by:

$\mathrm{Ac}\left(t,s\right)=\frac{1}{T}{\int }_0^tD\left(t^{\prime },s\right)t^{\prime },$

wherein D denotes the duration in seconds, and T is the total duration of all EEG segments. In other words, for each successive segment, the duration of the segment is added to the cumulative duration of the state to which the segment belongs, while the cumulative durations of the other states remains unchanged.

Alternatively or additionally, cumulative membership values may be computed and displayed by integrating the above-mentioned membership function w_kⁿ over successive segments. Parameters that can be extracted in this manner include:

1. Total Membership Index Up to Segment N:

$T\left(k\right)=\frac{1}{N}\sum _{n=1}^Nw_n^k;k=1,\dots ,K$

2. Cumulative Membership Index of Segment N:

$\mathrm{Af}\left(j,k\right)=\frac{1}{N}\sum _{n=1}^jw_n^k.$

The accumulation rate of each frequency state can be modeled by fitting an exponential function g(t)=1−e^−μt to the accumulation function, using least squares fitting, for example. The estimated accumulation rate μ for each curve is shown in the figure.

Changing trends in the state accumulation plot are indicative of changes and/or fragmentation of sleep states. For example, a knee 88 in HF curve 80 marks the point of transition from wakeful to sleeping states (occurring in this case about one hour after the beginning of the trial). The accumulation rate of HF states is markedly lower following the wake/sleep transition in normal patients, as can be seen in FIG. 7. By contrast, patients who suffer from sleep disorders exhibit higher values of HF accumulation during periods of sleep. The inventors have also observed that in some patients whose hypnograms appear to be normal, fragmented sleep can still be detected on the basis of an elevated HF accumulation rate.

FIG. 8 is a sleep/wake accumulation plot, showing cumulative durations of sleep and wake states of a patient over time, in accordance with an embodiment of the present invention. In this plot, curves 90 and 92 show the fractional durations of wake and sleep states, respectively. For this purpose, each EEG segment may be classified as sleep- or wake-related, according to the following criteria:

• • Wake-related—HF and noisy EEG segments.
  • Sleep related—MF1, MF2 and LF segments.
    This plot provides information similar to the frequency state accumulation plot of FIG. 7, but in a more condensed form. Accumulation rates of the sleep and wake states are computed in a manner similar to that described above. Changing trends in the sleep/wake accumulation plot may indicate changes and/or fragmentation of sleep.

FIG. 9 is a transition matrix showing probabilities of transitions among frequency states in successive segments of an EEG signal, in accordance with an embodiment of the present invention. Such a matrix can be calculated over the entire recording time for a given patient or for certain portions of the recording, for example, during a selected sleep stage. To calculate the transition values, processor 32 counts changes or persistence of the sleep state from second to second. In other words, if the duration of an HF segment is 10 sec, followed by transition to MF1, the processor will count ten transitions from the HF state to itself and then one HF:MF1 transition. (As a result, it can be seen that the values on the diagonal of the transition matrix are much larger than the off-diagonal values.) The transition probability P(i,j) from state i to state j is then calculated as follows:

$P\left(i,j\right)=\frac{N_{i,j}}{\sum _jN_{i,j}}$

wherein N_i,j is the number of transitions from state i to state j.

The transition matrix shows a pattern of frequency state dynamics during sleep, which can be used as a measure of sleep quality. For example, the inventors found that in a group of patients suffering from fragmented sleep (who nonetheless presented apparently normal hypnograms), the transition probability from state MF2 to LF state was substantially lower than in normal patients. This result reflects a deficiency in low-frequency (LF) activity that characterizes fragmented sleep.

Various other sleep quality indicators may be derived from the EEG signal and calculated over the entire sleep period or for selected sleep stages. For example, the sleep quality indicators may relate to transient phenomena in the EEG, such as K-complexes and/or spindles. A K-complex index, which quantifies the frequency of K-complex episodes during sleep, may be calculated as follows:

$\frac{\mathrm{Number}\mathrm{of}K\text{-}\mathrm{complexes}\mathrm{in}\mathrm{stage}s}{\mathrm{Total}\mathrm{duration}\mathrm{of}\mathrm{stage}s}$

A spindle index, quantifying the frequency of EEG spindles during sleep, may be calculated in like fashion. (K-complexes and spindles are well-known phenomena in EEG. Techniques for automatic identification and monitoring of these phenomena are described in the above-mentioned related patent applications.)

Although the embodiments described above relate mainly to analysis of EEG signals, the principles of these embodiments may similarly be applied to other physiological indicators. For example, a snoring index (based on identification of snoring episodes by audio analysis) may be used to indicate the number or duration of snores during one or more sleep stages.

As another example, a transition matrix of the type shown in FIG. 9 may be computed for other indicators, such as pathological respiration states. The above-mentioned related applications describe methods for automatic classification of respiration states based on respiration measurements during sleep. According to one scheme, these pathological respiration states include central breathing, obstructive breathing, mixed breathing, hypopnea and RERA (respiratory effort related arousal). A suitable transition matrix may be constructed to show transition patterns between the respiration states.

Furthermore, processor 32 may generate respiratory event histograms to describe the distribution of the duration of respiratory events during different sleep stages. (Methods for identifying respiratory events are likewise described in the above-mentioned related applications.) Additionally or alternatively, respiratory event histograms may be presented as a function of body position, time of night, or pressure titration levels of a respiratory assist device. The processor may also assign a confidence level to each suspected respiratory event (for example, from 0 for non-events to 1 for events that are certain), and the confidence levels may be displayed as a function of respiration state in a density plot similar to that shown in FIG. 4B.

Respiratory events are typically accompanied by a drop in heart rate (bradycardia), followed by heart rate elevation (tachycardia). Processor 32 may calculate sleep quality indicators based on these phenomena. For example, a relative heart rate index RHR, indicating the change (drop and/or elevation) of the heart rate associated with respiratory events, may be calculated as follows:

$\mathrm{RHR}=\frac{\mathrm{HR}\left(t\right)-\mathrm{BHR}}{\mathrm{BHR}}·100,$

Here HR(t) is the HR in the time interval of interest, and BHR is the baseline heart rate.

It will be appreciated that the embodiments described above are cited by way of example, and that the present invention is not limited to what has been particularly shown and described hereinabove. Rather, the scope of the present invention includes both combinations and subcombinations of the various features described hereinabove, as well as variations and modifications thereof which would occur to persons skilled in the art upon reading the foregoing description and which are not disclosed in the prior art.

APPENDIX A

Calculation of Features and Classification of Frequency Groups

1. Definition of Features

• • f₁=δ: Relative energy in delta frequency band (0.5-4 Hz).
  • f₂=θ: Relative energy in theta band (4-7 Hz).
  • f₃=α+σ+β: Relative energy in alpha (7-12 Hz), sigma (12-15 Hz) and beta (15-30 Hz) bands.
  • f₄=F_f: Fundamental frequency.
  • f₅=p₂p: Peak to peak amplitude.
  • f₆=Var: Normalized variance (energy) normalized to a scale of 0-100.

Example

Calculation of Relative Energy in a Given Band

If S_xx(f) denotes the estimated power spectrum of an EEG segment, then

$E_b\left(f1,f2\right)=\frac{{\int }_{f1}^{f2}S_{\mathrm{xx}}\left(f\right)}{{\int }_{-\infty }^{\infty }S_{\mathrm{xx}}\left(f\right)}·100$

is the relative energy in the frequency band that is bounded by frequencies f1 and f2.

Example

Calculation of Normalized Variance of Relative Energy

Let V(x_k) denote the energy variance of the samples within the k^th EEG segment (denoted x_k). The normalized variance is then

$\stackrel{\_}{V}\left(x_k\right)=\frac{V\left(x_k\right)}{\sum _{k=1}^KV\left(x_k\right)}·100,$

wherein K is the total number of segments.

Example

Calculation of Fundamental Frequency

The fundamental frequency of an EEG segment is the moment of the frequency spectrum, calculated as follows:

$F_f={\int }_{-\infty }^{\infty }f·S_{\mathrm{xx}}\left(f\right)f$

2. Initial Classification Using Fuzzy Clustering

The EEG segments are classified into the following classes in the feature space defined by f₁, f₂ and f₃:

C1—high frequency.

C2—mixed frequency.

C3—Low frequency.

The segments may be classified according to the criterion: C(x_n)=arg m_kax (w_n^k), wherein x_n is the feature vector of segment n, and w_n^k is the membership level of the segment in cluster k. In this case k=1, 2, 3.

3. Unification: {tilde over (C)}₂←C₂ ∪C₃

4. Partitioning of {tilde over (C)}₂ by Fundamental Frequency:

f_4,n>5 Hzε{tilde over (C)}₂

f_4,n≦5 Hzε{tilde over (C)}₃

(Here f_4,n is the fundamental Frequency of Segment n.)

5. Partitioning of {tilde over (C)}₃ Using Fuzzy Clustering in the Feature Space Defined by f₄ and f₅

Features corresponding to the maximal centroid of f₅ are returned to {tilde over (C)}₂ in two subclasses. Feature vectors classified in the subclass characterized by minimal f₅ value of the centroid are returned to {tilde over (C)}₂.

6. Definition of Validation Rules

$•R_1=\frac{\alpha +\sigma +\beta }{\theta +\delta }>0.3$ $•R_2=\frac{\alpha +\sigma +\beta }{\theta +\delta }>0.7$ $•R_3=F_f\le 3\mathrm{Hz}$

7. Implementation of Rules

• • 1. ∀x_nε{tilde over (C)}₃ if ˜R_3,n x_nε{tilde over (C)}₂
  • 2. ∀x_nεC₁ if ˜R_1,n x_nε{tilde over (C)}₂
  • 3. ∀x_n if R_2,n x_nεC₁

8. Partitioning of {tilde over (C)}₂ Using Hierarchical Fuzzy Clustering in the Feature Space Defined by f₆ into Ĉ₂ and C₄.

Hierarchical fuzzy clustering partitions the feature space in a recursive manner. Each level of recursion generates a new hierarchy level, in which a portion of the feature space attributed to one selected cluster is subdivided into M groups. In the present case, at each hierarchy level, the cluster with minimal centroid value is partitioned into two new clusters until the diversity level between clusters at the same hierarchy level drops below a predetermined threshold. The diversity level D is given by:

$D=\frac{\mathrm{maximal}\mathrm{centroid}\mathrm{value}}{\mathrm{minimal}\mathrm{centroid}\mathrm{value}}$

In one embodiment, the threshold on D is 2, i.e., when D<2 the recursion stops.

The feature vectors attributed to the cluster with minimal centroid value are assigned to C₄, while the rest of the feature vectors are assigned to Ĉ₂. C₄ corresponds to MF1, while Ĉ₂ corresponds to MF2.

Claims

1. A method for diagnosis, comprising:

acquiring a physiological signal from a patient during a period of sleep;

segmenting the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum;

computing respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum; and

displaying a plot indicative of the levels of membership of the segments in the sequence over time.

2. The method according to claim 1, wherein computing the respective levels comprises applying fuzzy clustering to the segments so as to define the states.

3. The method according to claim 1, wherein displaying the plot comprises displaying a density plot, in which the levels of membership are represented by color variations.

4. The method according to claim 1, wherein displaying the plot comprises displaying an accumulation plot, showing cumulative levels of membership of the segments in the plurality of frequency states over the sequence.

5. The method according to claim 1, wherein displaying the plot comprises displaying an accumulation plot showing cumulative durations of the segments in each of the plurality of frequency states.

6. The method according to claim 5, and comprising determining and comparing respective accumulation rates of the cumulative durations in at least two of the frequency states.

7. The method according to claim 1, wherein displaying the plot comprises assigning each of at least some of the segments to one of a waking state and a sleep state responsively to the frequency spectrum, and displaying an accumulation plot showing a cumulative assignment of the segments to the waking and sleep states over time.

8. The method according to claim 1, wherein the physiological signal comprises an electroencephalogram (EEG) signal.

9. A method for diagnosis, comprising:

acquiring a physiological signal from a patient during a period of sleep;

segmenting the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum;

computing a fundamental frequency of each segment in the time sequence responsively to a moment of the respective frequency spectrum of the segment; and

displaying a plot showing the fundamental frequency of the segments in the sequence over time.

10. The method according to claim 9, wherein displaying the plot comprises showing at least one of a trend and a variance of the fundamental frequency.

11. The method according to claim 9, wherein computing the respective levels comprises applying fuzzy clustering to the segments so as to define the states.

12. The method according to claim 9, wherein the physiological signal comprises an electroencephalogram (EEG) signal.

13. A method for diagnosis, comprising:

acquiring a physiological signal from a patient during a period of sleep;

segmenting the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum;

computing respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum;

based on the respective levels of membership, determining a sleep quality indicator responsively to a statistical characteristic of the segments; and

displaying the sleep quality indicator.

14. The method according to claim 13, wherein computing the respective levels comprises applying fuzzy clustering to the segments so as to define the states.

15. The method according to claim 13, wherein the statistical characteristic comprises at least one duration measure selected from a group of duration measures consisting of:

a cumulative duration of the segments associated with each of the frequency clusters;

a relative duration of the segments associated with each of the frequency clusters;

a mean duration of the segments associated with each of the frequency clusters;

a variance of a duration of the segments associated with each of the frequency clusters;

a total number of the segments associated with each of the frequency clusters; and

a relative duration of the segments associated with each of the frequency clusters.

16. The method according to claim 13, and comprising assigning the segments to predefined sleep stages responsively to the frequency spectrum, wherein determining the sleep quality indicator comprises computing the statistical characteristic with respect to each of the sleep stages.

17. The method according to claim 13, wherein displaying the sleep quality indicator comprises displaying a plot indicative of the levels of membership of the segments in the sequence over time.

18. The method according to claim 13, wherein displaying the sleep quality indicator comprises displaying a plot showing a fundamental frequency of the segments in the sequence over time.

19. The method according to claim 13, wherein computing the respective levels of membership comprises assigning the segments in the time sequence to respective frequency states, and wherein determining the sleep quality indicator comprises computing probabilities of transition among the frequency states.

20. The method according to claim 13, wherein the physiological signal comprises an electroencephalogram (EEG) signal.

21. The method according to claim 20, and comprising identifying transient phenomena in the EEG signal, and computing an index quantifying a frequency of occurrence of the transient phenomena.

22. The method according to claim 21, wherein the transient phenomena comprise one or more of K-complexes and spindles.

23. The method according to claim 13, wherein the physiological signal comprises a respiration signal.

24. The method according to claim 23, and comprising identifying respiratory events occurring during the period of sleep, and computing statistical characteristics of the respiratory events.

25. The method according to claim 24, wherein computing the statistical characteristics comprises computing and displaying a respiratory event histogram.

26. The method according to claim 24, and comprising measuring a heart rate of the patient, wherein computing the statistical characteristics comprises computing a relative heart rate index indicative of changes in the heart rate associated with the respiratory events.

27. The method according to claim 24, wherein computing the statistical characteristics comprises assigning respective confidence levels to the respiratory events, and displaying the confidence levels as a function of respiration state.

28. A method for diagnosis, comprising:

acquiring a physiological signal from a patient during a period of sleep;

segmenting the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum;

computing respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum;

responsively to the respective levels of membership, assigning each of at least some of the segments to one of a waking state and a sleep state responsively to the frequency spectrum; and

displaying an accumulation plot showing a cumulative assignment of the segments to the waking and sleep states over time.

29. The method according to claim 28, wherein computing the respective levels comprises applying fuzzy clustering to the segments so as to define the states.

30. The method according to claim 28, and comprising determining and comparing respective accumulation rates of the waking and sleep states.

31. The method according to claim 28, wherein the physiological signal comprises an electroencephalogram (EEG) signal.

32. Diagnostic apparatus, comprising:

a sensor, which is adapted to acquire a physiological signal from a patient during a period of sleep; and

a diagnostic processor, which is adapted to segment the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum, to compute respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum, and to display a plot indicative of the levels of membership of the segments in the sequence over time.

33. The apparatus according to claim 32, wherein the processor is adapted to apply fuzzy clustering to the segments so as to define the states.

34. The apparatus according to claim 32, wherein the plot comprises a density plot, in which the levels of membership are represented by color variations.

35. The apparatus according to claim 32, wherein the plot comprises an accumulation plot, showing cumulative levels of membership of the segments in the plurality of frequency states over the sequence.

36. The apparatus according to claim 32, wherein the plot comprises an accumulation plot showing cumulative durations of the segments in each of the plurality of frequency states.

37. The apparatus according to claim 36, wherein the processor is adapted to determine and output respective accumulation rates of the cumulative durations in at least two of the frequency states.

38. The apparatus according to claim 32, wherein the processor is adapted to assign each of at least some of the segments to one of a waking state and a sleep state responsively to the frequency spectrum, and to display an accumulation plot showing a cumulative assignment of the segments to the waking and sleep states over time.

39. The apparatus according to claim 32, wherein the sensor comprises at least one electrode, and wherein the physiological signal comprises an electroencephalogram (EEG) signal.

40. Diagnostic apparatus, comprising:

a sensor, which is adapted to acquire a physiological signal from a patient during a period of sleep; and

a diagnostic processor, which is adapted to segment the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum, to compute a fundamental frequency of each segment in the time sequence responsively to a moment of the respective frequency spectrum of the segment, and to display a plot showing the fundamental frequency of the segments in the sequence over time.

41. The apparatus according to claim 40, wherein the processor is adapted to display in the plot at least one of a trend and a variance of the fundamental frequency.

42. The apparatus according to claim 40, wherein the processor is adapted to apply fuzzy clustering to the segments so as to define the states.

43. The apparatus according to claim 40, wherein the sensor comprises at least one electrode, and wherein the physiological signal comprises an electroencephalogram (EEG) signal.

44. Diagnostic apparatus, comprising:

a sensor, which is adapted to acquire a physiological signal from a patient during a period of sleep; and

a diagnostic processor, which is adapted to segment the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum, to compute respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum, to determine, based on the respective levels of membership, a sleep quality indicator responsively to a statistical characteristic of the segments, and to display the sleep quality indicator.

45. The apparatus according to claim 44, wherein the processor is adapted to apply fuzzy clustering to the segments so as to define the states.

46. The apparatus according to claim 44, wherein the statistical characteristic comprises at least one duration measure selected from a group of duration measures consisting of:

a cumulative duration of the segments associated with each of the frequency clusters;

a relative duration of the segments associated with each of the frequency clusters;

a mean duration of the segments associated with each of the frequency clusters;

a variance of a duration of the segments associated with each of the frequency clusters;

a total number of the segments associated with each of the frequency clusters; and

a relative duration of the segments associated with each of the frequency clusters.

47. The apparatus according to claim 44, wherein the processor is adapted to assign the segments to predefined sleep stages responsively to the frequency spectrum, and to compute the statistical characteristic with respect to each of the sleep stages.

48. The apparatus according to claim 44, wherein the processor is adapted to display a plot indicative of the levels of membership of the segments in the sequence over time.

49. The apparatus according to claim 44, wherein the processor is adapted to display a plot showing a fundamental frequency of the segments in the sequence over time.

50. The apparatus according to claim 44, wherein the processor is adapted to assign the segments in the time sequence to respective frequency states, and to compute probabilities of transition among the frequency states.

51. The apparatus according to claim 44, wherein the sensor comprises at least one electrode, and wherein the physiological signal comprises an electroencephalogram (EEG) signal.

52. The apparatus according to claim 51, wherein the processor is adapted to identify transient phenomena in the EEG signal, and to compute an index quantifying a frequency of occurrence of the transient phenomena.

53. The apparatus according to claim 52, wherein the transient phenomena comprise one or more of K-complexes and spindles.

54. The apparatus according to claim 44, wherein the sensor comprises a respiration sensor, and wherein the physiological signal comprises a respiration signal.

55. The apparatus according to claim 54, wherein the processor is adapted to identify respiratory events occurring during the period of sleep, and to compute statistical characteristics of the respiratory events.

56. The apparatus according to claim 55, wherein the processor is adapted to compute and display a respiratory event histogram.

57. The apparatus according to claim 54, wherein the processor is adapted to determine a heart rate of the patient, and to compute a relative heart rate index indicative of changes in the heart rate associated with the respiratory events.

58. The apparatus according to claim 54, wherein the processor is adapted to assign respective confidence levels to the respiratory events, and to display the confidence levels as a function of respiration state.

59. Diagnostic apparatus, comprising:

a sensor, which is adapted to acquire a physiological signal from a patient during a period of sleep; and

a diagnostic processor, which is adapted to segment the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum, to compute respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum, to assign each of at least some of the segments, responsively to the respective levels of membership, to one of a waking state and a sleep state responsively to the frequency spectrum, and to display an accumulation plot showing a cumulative assignment of the segments to the waking and sleep states over time.

60. The apparatus according to claim 59, wherein the processor is adapted to apply fuzzy clustering to the segments so as to define the states.

61. The apparatus according to claim 59, and wherein the processor is adapted to determine and output respective accumulation rates of the waking and sleep states.

62. The apparatus according to claim 59, wherein the sensor comprises at least one electrode, and wherein the physiological signal comprises an electroencephalogram (EEG) signal.

63. A computer software product, comprising a computer-readable medium in which program instructions are stored, which instructions, when read by a computer, cause the computer to acquire a physiological signal from a patient during a period of sleep, to segment the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum, to compute respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum, and to display a plot indicative of the levels of membership of the segments in the sequence over time.

64. The product according to claim 63, wherein the instructions cause the computer to apply fuzzy clustering to the segments so as to define the states.

65. The product according to claim 63, wherein the plot comprises a density plot, in which the levels of membership are represented by color variations.

66. The product according to claim 63, wherein the plot comprises an accumulation plot, showing cumulative levels of membership of the segments in the plurality of frequency states over the sequence.

67. The product according to claim 63, wherein the plot comprises an accumulation plot showing cumulative durations of the segments in each of the plurality of frequency states.

68. The product according to claim 67, wherein the instructions cause the computer to determine and output respective accumulation rates of the cumulative durations in at least two of the frequency states.

69. The product according to claim 63, wherein the instructions cause the computer to assign each of at least some of the segments to one of a waking state and a sleep state responsively to the frequency spectrum, and to display an accumulation plot showing a cumulative assignment of the segments to the waking and sleep states over time.

70. The product according to claim 63, wherein the physiological signal comprises an electroencephalogram (EEG) signal.

71. A computer software product, comprising a computer-readable medium in which program instructions are stored, which instructions, when read by a computer, cause the computer to acquire a physiological signal from a patient during a period of sleep, to segment the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum, to compute a fundamental frequency of each segment in the time sequence responsively to a moment of the respective frequency spectrum of the segment, and to display a plot showing the fundamental frequency of the segments in the sequence over time.

72. The product according to claim 71, wherein the instructions cause the computer to display in the plot at least one of a trend and a variance of the fundamental frequency.

73. The product according to claim 71, wherein the instructions cause the computer to apply fuzzy clustering to the segments so as to define the states.

74. The product according to claim 71, wherein the physiological signal comprises an electroencephalogram (EEG) signal.

75. A computer software product, comprising a computer-readable medium in which program instructions are stored, which instructions, when read by a computer, cause the computer to acquire a physiological signal from a patient during a period of sleep, to segment the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum, to compute respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum, to determine, based on the respective levels of membership, a sleep quality indicator responsively to a statistical characteristic of the segments, and to display the sleep quality indicator.

76. The product according to claim 75, wherein the instructions cause the computer to apply fuzzy clustering to the segments so as to define the states.

77. The product according to claim 75, wherein the statistical characteristic comprises at least one duration measure selected from a group of duration measures consisting of:

a cumulative duration of the segments associated with each of the frequency clusters;

a relative duration of the segments associated with each of the frequency clusters;

a mean duration of the segments associated with each of the frequency clusters;

a variance of a duration of the segments associated with each of the frequency clusters;

a total number of the segments associated with each of the frequency clusters; and

a relative duration of the segments associated with each of the frequency clusters.

78. The product according to claim 75, wherein the instructions cause the computer to assign the segments to predefined sleep stages responsively to the frequency spectrum, and to compute the statistical characteristic with respect to each of the sleep stages.

79. The product according to claim 75, wherein the instructions cause the computer to display a plot indicative of the levels of membership of the segments in the sequence over time.

80. The product according to claim 75, wherein the instructions cause the computer to display a plot showing a fundamental frequency of the segments in the sequence over time.

81. The product according to claim 75, wherein the instructions cause the computer to assign the segments in the time sequence to respective frequency states, and to compute probabilities of transition among the frequency states.

82. The product according to claim 75, wherein the physiological signal comprises an electroencephalogram (EEG) signal.

83. The product according to claim 82, wherein the instructions cause the computer to identify transient phenomena in the EEG signal, and to compute an index quantifying a frequency of occurrence of the transient phenomena.

84. The product according to claim 83, wherein the transient phenomena comprise one or more of K-complexes and spindles.

85. The product according to claim 75, wherein the sensor comprises a respiration sensor, and wherein the physiological signal comprises a respiration signal.

86. The product according to claim 85, wherein the instructions cause the computer to identify respiratory events occurring during the period of sleep, and to compute statistical characteristics of the respiratory events.

87. The product according to claim 86, wherein the instructions cause the computer to compute and display a respiratory event histogram.

88. The product according to claim 85, wherein the instructions cause the computer to determine a heart rate of the patient, and to compute a relative heart rate index indicative of changes in the heart rate associated with the respiratory events.

89. The product according to claim 85, wherein the instructions cause the computer to assign respective confidence levels to the respiratory events, and to display the confidence levels as a function of respiration state.

90. A computer software product, comprising a computer-readable medium in which program instructions are stored, which instructions, when read by a computer, cause the computer to acquire a physiological signal from a patient during a period of sleep, to segment the signal to define a time sequence of quasi-stationary segments, each of the segments having a respective frequency spectrum, to compute respective levels of membership of the segments in a plurality of frequency states responsively to the respective frequency spectrum, to assign each of at least some of the segments, responsively to the respective levels of membership, to one of a waking state and a sleep state responsively to the frequency spectrum, and to display an accumulation plot showing a cumulative assignment of the segments to the waking and sleep states over time.

91. The product according to claim 90, wherein the instructions cause the computer to apply fuzzy clustering to the segments so as to define the states.

92. The product according to claim 90, and wherein the instructions cause the computer to determine and output respective accumulation rates of the waking and sleep states.

93. The product according to claim 90, wherein the physiological signal comprises an electroencephalogram (EEG) signal.