NON-INVASIVE MONITORING OF RESPIRATORY RATE, HEART RATE AND APNEA

Abstract

A method and apparatus for estimating a respiratory rate of a patient. The method comprises the steps of recording respiratory sounds of the patient, deriving a plurality of respiratory rates from the recorded sounds using a plurality of respiratory rate estimating methods and applying a heuristic to the plurality of derived respiratory rates, the heuristic selecting one of the derived respiratory rates. The selected respiratory rate is the estimated respiratory rate. The apparatus comprises at least one sensor recording respiratory sounds of the patient, a plurality of respiratory rate processors, each of the processors comprising a respiratory rate calculating method, a heuristic means for selecting one of the calculated respiratory rates and a display means for displaying the selected respiratory as the estimated respiratory rate.

Description

FIELD OF THE INVENTION

The present invention relates to a method and apparatus for the non-invasive monitoring of respiratory rate, heart rate and apnea. In particular, the present invention relates to a method for determining respiratory rate by combining the results of a plurality of respiratory rate estimation methods and selecting a preferred rate using a heuristic, and an apparatus implementing the same.

BACKGROUND OF THE INVENTION

Respiratory Rate

Respiratory failure can become a life-threatening condition in a few minutes or be the result of a build up over several hours. Respiratory failure is very difficult to predict, and as a result continuous monitoring of respiratory activity is typically necessary in clinical, high-risk situations. Appropriate monitoring equipment can be life-saving (see Folke M, Cernerud L, Ekstrom M, HoK B; Critical Review of Non-invasive Respiratory Monitoring in Medical Care; Medical & Biological Engineering & Computing 2003, Vol 41, pp. 377-383).

Numerous studies have shown that Respiratory Rate (RR) provides one of the most accurate markers for indicating acute respiratory dysfunction, and thus is used to track the progress of patients in intensive care or post-operative care or anyone with potentially unstable respiration (see Krieger B, Feinerman D, Zaron A, Bizousky F; tinuous Noninvasive Monitoring of Respiratory Rate in Critically III Patients; Chest/90/5/November, 1986, pp 632-634, Browning I B, D′Alonzo G E, Tobin M J; Importance of Respiratory Rate as an Indicator of Respiratory Dysfunction in Patients with Cystic Fibrosis; Chest/97/6/June, 1990, pp 1317-1321, Gravelyn T R, Weg J G; Respiratory Rate as an Indicator of Acute Respiratory Dysfunction; JAMA, Sep. 5, 1980—Vol 244, No. 10, pp 1123-1125).

RR has also been shown to be a very accurate marker for weaning outcomes for ventilated patients (see Tobin M J, Perez W. Guenther M, Semmes B J, Mador J, Allen S J, Lodato R F, Dantzker D R; The Pattern of Breathing during Successful and Unsuccessful Trials of Weaning from Mechanical Ventilation; AM Rev Respir DIS 1986; 134:1111-1118 and EI-Khatib M, Jamaleddine G, Soubra R, Muallem M; Pattern of Spontaneous Breathing: Potential Marker for Weaning Outcome. Spontaneous Breathing Pattern and Weaning from Mechanical Ventilation; Intensive Care Med (2001) 27:52-58) as it exhibits high correlation with both the success and failure of extubations.

During sedation, monitoring of the RR has been shown to be a more rapid marker of the induction of anesthesia than any other clinical measure, such as lash reflex, loss of grip, cessation of finger tapping, and loss of arm tone (see Strickland T L, Drummond G B; Comparison of Pattern of Breathing with Other Measures of Induciton of Anesthesia, Using Propofol, Methohexital, and Servoflurane; British Journal Of Anesthesia, 2001, Vol. 86, No. 5, pp 639-644). During conscious sedation (narcotic sedation), there is always a risk of respiratory depression. However, monitoring of the respiratory pattern combined with pulse oximetry yield the most useful information about the occurrence of respiratory depression and changes in RR typically provide an earlier warning than does pulse oximetry or end-tidal CO₂ tension (see Shibutani K, Komatsu T, Ogawa T, Braatz T P, Tsuenekage T; Monitoring of Breathing Intervals in Narcotic Sedation; International Journal of Clinical Monitoring & Computing; 8: 159-162, 1991).

Respiration monitoring is also useful during non critical care, e.g. during exercise testing and different types of cardiac investigations. In the latter case there is also need to time the different phases of respiration, since the heart function is modulated by respiration. A forthcoming area of application for respiration monitoring may be that of home-care (see Hult P, et al., An improved bioacoustic method for monitoring of respiration. Technology and Health Care 2004; 12: 323-332).

Despite the obvious benefits of performing continuous respiratory monitoring, the search for an accurate, non-invasive, and non-obtrusive method to continuously monitor RR has proven to be long and unsuccessful. Several technologies have been developed in an attempt to fill this clinical gap, but none has gained sufficient physician confidence to become a standard of care. In this regard, inductive plethysmography, fiber optic humidification and capnography are among the most popular technologies. Each of these has advantages and disadvantages, but none has proven to be clearly superior. More suitable technologies are still needed to address such issues as: low signal to noise ratio, different breath sound intensities, phase duration, variable breathing patterns, interferences from non-biological sounds (electromagnetic interference, movement artifacts, environmental noise, etc.), and interference from biological sounds such as the heart beat, swallowing, coughing, vocalization, etc.

Tracheal sounds, typically heard at the suprasternal notch or at the lateral neck near the pharynx, have become of significant interest during the last decade. The tracheal sound signal is strong, covering a wider range of frequencies than lung sounds at the chest wail, has distinctly separable respiratory phases, and a close relation to airflow. Generally, the placement of a sensor over the trachea is relatively easy as there is less interference from body hair, garments, etc, as compared to chest-wall recording sites.

The generation of tracheal sounds is primarily related to turbulent air flow in upper airways, including the pharynx, glottis, and subglottic regions. Flow turbulence and jet formation at the glottis cause pressure fluctuations within the airway lumen. Sound pressure waves within the airway gas and airway wall motion are likely contributing to the vibrations that reach the neck surface and are recorded as tracheal sounds. Because the distance from the various sound sources in the upper airways to a sensor on the neck surface is relatively short and without interposition of lung tissue, tracheal sounds are often interpreted as a more pure, less filtered breath sound. Tracheal sounds have been characterized as broad spectrum noise, covering a frequency range of less than 100 Hz to more than 1500 Hz, with a sharp drop in power above a cutoff frequency of approximately 800 Hz. While the spectral shape of tracheal sounds varies widely from person to person, it is quite reproducible within the same person. This likely reflects the strong influence of individual airway anatomy.

Pulmonary clinicians are interested in tracheal sounds as early indicators of upper airway flow obstruction and as a source for quantitative as well as qualitative assessments of ventilation. Measurements of tracheal sounds provide valuable and in some cases unique information about respiratory health.

Apnea

Apnea monitoring by simple acoustical detection of tracheal sounds is an obvious application and has been successfully applied in both adults and in children. The detection of apneic events are a normal derivative from the RR estimation. A temporary cessation in breathing, typically lasting at least 10 seconds in duration, is referred to as apnea. Longer pauses may be of sufficient duration to cause a fall in the amount of oxygen in the arterial blood, and have the potential to cause permanent organ damage, or, in the extreme case, death. Adults with sleep apnea are very susceptible to exacerbation of this condition post-surgery, and therefore their respiration must be carefully monitored. Disordered breathing during sleep is a common condition with an estimated prevalence of up to 24% in men and 9% in women in North America It is associated with excessive morbidity and increased mortality from cardiovascular and cerebrovascular events and increased risk of road traffic accidents (see Young et al., The occurrence of sleep-disordered breathing among middle-aged adults, N Engl J Med 1993; 328: 1230-1235). The condition can be suspected clinically in the presence of classic symptoms such as snoring, daytime hyper-somnolence, obesity, and male gender. The diagnosis is typically confirmed by polysomnography. The most common sleep disorder is Obstructive Sleep Apnea Syndrome (OSAS), also known as Sleep Apnea Hypopnea Syndrome (SAHS). This condition is so much linked to excessive morbidity and mortality, that it is considered a public health hazard at par with smoking (see Findley et al., Automobile accidents involving patients with obstructive sleep apnea, Am Rev Respir //pis 1988; 138: 337-340).

Heart Rate

The rhythm of the heart in terms of beats per minute may be easily estimated on the tracheal site by counting the readily identifiable heart sound waves. Heart rate (FIR) is altered by cardiovascular diseases and abnormalities such as arrhythmias and conduction problems. The main cause of death in developed countries is due to cardiovascular diseases and mostly they are triggered by an arrhythmic event (ventricular tachycardia or ventricular fibrillation). The HR is controlled by specialized pacemaker cells that form the sinoatrial (SA) node located at the junction of the superior vena cava and the right atrium. The firing rate of the SA node is controlled by impulses from the autonomous and central nervous system. It is now commonly accepted that the heart sounds are not caused by valve leaflet movement per se, as earlier believed, but by vibrations of the whole cardiovascular system triggered by pressure gradients (see Rangayyan R M, Biomedica, Signal Analysis 2002, IEEE Press Series, Wiley Inter-Science). The normal (resting) HR is about 70 bpm. The HR is slower during sleep, but abnormally low HR (below 60 bpm) during activity could indicate a disorder called bradycardia. The instantaneous HR could reach values as high as 200 bpm during vigorous exercise or athletic activity; a high resting HR could be due to illness, disease, or cardiac abnormalities, and is termed tachycardia.

SUMMARY OF THE INVENTION

In order to address the above and other drawbacks, there is provided a method for estimating a respiratory rate of a patient comprising the steps of recording respiratory sounds of the patient, deriving a plurality of respiratory rates from the recorded sounds using a plurality of respiratory rate estimating methods and applying a heuristic to the plurality of derived respiratory rates, the heuristic selecting one of the derived respiratory rates. The selected respiratory rate is the estimated respiratory rate.

There is also provided a method for signaling sleep apnea comprising the steps of the above method wherein when the estimated respiratory rate exceeds a predetermined interval an alarm is raised.

Furthermore, there is provided a method for estimating a respiratory rate of a patient. The method comprises the steps of recording respiratory sounds of the patient and determining silent intervals in the recorded sounds. The estimated respiratory rate is equivalent to a frequency of the silent intervals.

Additionally, there is provided an apparatus for providing an estimated respiratory rate of a patient. The apparatus comprises at least one sensor recording respiratory sounds of the patient, a plurality of respiratory rate processors, each of the processors comprising a respiratory rate calculating method, a heuristic means for selecting one of the calculated respiratory rates and a display means for displaying the selected respiratory as the estimated respiratory rate.

Also, there is provided an apparatus for signalling sleep apnea in a patient. The apparatus comprises at least one sensor recording respiratory sounds of the patient, a plurality of respiratory rate processors where each of the processors comprising a respiratory rate calculating methods, a heuristic means for selecting one of the calculated respiratory rates and an alarm. When the selected respiratory rate is slower than a predetermined rate, the alarm is activated.

Other objects, advantages and features of the present invention will become more apparent upon reading of the following non-restrictive description of illustrative embodiments thereof, given by way of example only with reference to the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

In the appended drawings:

FIG. 1 is a front view of a patient with a sensor attached to monitor respiratory sounds according to an illustrative embodiment of the present invention;

FIG. 2 is a flow chart of a non-invasive respiratory rate, heart rate and apnea monitor according to an illustrative embodiment of the present invention;

FIG. 3 is a graph of a respiratory sound signal showing artifacts (glitches) and with glitches removed according to an illustrative embodiment of the present invention;

FIG. 4 is a graph of a respiratory sound signal, flow signal and wavelet decomposition envelope according to an illustrative embodiment of the present invention;

FIG. 5 is a graph of a respiratory sound signal divided into frequency bands according to an illustrative embodiment of the present invention;

FIG. 6 is a graph of a respiratory sound signal, flow signal and squared envelope according to an illustrative embodiment of the present invention;

FIG. 7 is a flow chart of speech processing method according to an illustrative embodiment of the present invention; and

FIG. 8 is a graph of an example of a method based on a quadratic detection function of the speech processing method according to an illustrative embodiment of the present invention.

DETAILED DESCRIPTION OF THE ILLUSTRATIVE EMBODIMENTS

Referring now to FIG. 1, a non-invasive respiratory rate, heart rate and apnea monitor, generally referred to using the reference numeral 10, will now be described. Biological sound sensors 12, illustratively identical and for example as those described in U.S. Pat. No. 6,661,161, detect the biological sounds and vibrations emanating from the throat of a patient 14 and produces an output electrical signal. Note that in a given embodiment, a single biological sound sensor as in 12 or more than one could also be used to detect biological sounds and vibrations. The signals are transferred via appropriate electrical leads 16 to a data acquisition system 18, which amplifies and filters the electrical signal prior to converting them into a digital format. Finally, the methods implemented in a computer 20 extract the physiological information from the data and display the results through a graphical user interface 22.

Acquisition System

The acquisition system comprises a Pentium based laptop computer running Windows 2000 and a multi-channel custom designed biosignal amplifier. The bandwidth of the sound channel(s) is selectable from 0 to 1500 Hz. A sampling frequency for the sound channel(s) was chosen and set at 3 kHz. The resolution of the A/D conversion of the data acquisition board was 12 bits. The graphical user interface was designed using the Labview® (National lnstrument, Austin, Tex., USA) programming language and digital signal processing methods were developed and tested in Matlab® (The MathWorks, Inc., Natick, Mass., USA).

Methods

A flow chart indicating the elements of the signal processing method used to estimate the RR from the respiratory tracheal sound signal is provided in FIG. 2 (See also, Sierra G, Telfort V, Popov B, Durand L G Agarwal R, Lanzo V; Monitoring Respiratory Rate Based on Tracheal Sounds, First Experiences; IEEE/EMBS 26th Conferences, San Francisco, Calif., 2004 which is incorporated herein by reference). These elements are described herein below.

Sound Signals and Segmentation Step. The respiratory sound collection is illustratively performed at a sampling frequency of 3000 Hz. For analysis purposes, the sound signal is illustratively segmented into 20 second blocks (although variable block lengths could also be processed) with each block containing five seconds of signal from the previous block (the overlap may also be variable). The breathing frequency is estimated from the sampled 20 seconds of data collection, averaged and displayed every minute.

Pre-processing Step. This step is aimed at ensuring a respiratory sound signal which is as free of interference from internal and external sources of sounds as possible. The following actions are performed during the preprocessing step:

• • a) A comb-filter is applied to remove the interference from 60 Hz and its harmonics;
  • b) signals with extremely low signal to noise ratio or high artifacts that saturate the amplifiers are excluded (as will apparent to persons of skill in the art, if saturation occurs, for example when a person talks or cough, signals cannot be processed because the amplifier starts clipping these strong/high amplitude signals. In this case it is preferred to exclude that data segment from analysis. Concerning the case where recordings with extremely low signal to noise ratio exist, they should also be excluded because the signal contribution is almost null due to the masking effect of noises.);
  • c) glitches (or motion artifacts) arising from rubbing clothes on the sensor, intermittent contact, etc., are removed (or attenuated);
  • d) filtering based on the multi-resolution decomposition (MRD) of a wavelet Transform; and
  • e) removal of strong biological sounds that do not saturate amplifiers but contribute to RR wrong estimation (such strong biological sounds may modify some of the statistical characteristics of the signal being processed, such as maximum amplitude, etc, that are used to detect apnea or low signal to noise ratio.).

Glitch Removal

Referring to FIG. 3, illustratively, the presence of glitches is determined by sampling the respiratory tracheal data. Samples having an amplitude in excess of three times (a value determined as sufficient to identify a larges portion of glitches while avoiding capturing other non-glitch signals) the value of the standard deviation are categorized as glitches and removed and replaced by a constant value equal to the amplitude of the sample immediately preceding the removed sample. The mean and standard deviation are illustratively calculated for every one second of signal. The net effect is clipping the signal which means that glitches are not completely removed but rather attenuated. A previous attempt using twice the standard deviation was found to be less effective as an attenuated signal was produced making the estimation process significantly more difficult to accomplish, specially in recordings with low signal to noise ratio. FIG. 3 shows in the top panel the input signal with scattered glitches and bottom panel after removing some of the glitches.

Multi-Resolution Decomposition (MRD)

Referring to FIG. 4, respiratory sounds (as well as heart sounds) are complicated multi-component non-stationary signals and lend themselves to the use of non-stationary analysis techniques for analyses. MRD allows splitting the respiratory signal into different spectral bands. This decomposition allows for extensive separation of sounds and allows the selection of the best frequency band for processing the respiratory sound signals with the least interference (see FIG. 5).

The frequency range of the above-mentioned frequency bands is determined by the sampling frequency (illustratively Fs=300 Hz). The output is the filtered signal contained in the frequency bands from 187 Hz to 750 Hz (from 200 Hz to 800 Hz is considered to contain the most important information of the tracheal signal). The MRD approach of the wavelet transform is applied to the respiratory sound signals based on a methodology known in the art for the analysis of different cardiovascular bio-signals. See Sierra G, Fetsch T, Reinhardt L. Martinez-Rubio A, Makijarvi M, Balkenhoff K, Borggrefe M, Breithardt G., Multiresolution decomposition of the signal-averaged ECG using the Mallat approach for prediction of arrhythmic events after myocardial infarction, J Electrocardiol 1995, 29:223-234, Sierra G, Reinhardt L, Fetsch T, Martinez-Rubio A, Makijarvi M, Yli-Mayry S, Montonen J, Katila T, Borggref M, Breithardt G, Risk stratification of patients after myocardial infarction based on wavelet decomposition of the signal-averaged electrocardiogram, Annals of Noninvasive Electrocardiology 1997; 2: 47-58, Sierra G, Gomez M J, Le Guyader P, Trelles F, Cardinal R, Savard P, Nadeau R, Discrimination between monomorphic and polymorphic ventricular tachycardia using cycle length variability measured by wavelet transform analysis, Electrocardiol 1998; 31: 245-255, and Sierra G, Morel P, Savard P, Le Guyader P, Benabdesselam M, Nadeau R., Multiresolution decomposition of the signal-averaged ECG of postinfarotion patients with and without bundle branch block, Proceedings of the 18th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Amsterdam, Oct. 31-Nov. 3, 1996, all incorporated herein by reference.

The MRD approach to wavelet transform allows noise to be removed from the input signals and the biological sounds to be separated into different frequency bands. As is known in art, a variety of wavelet families exist, one or more of which may be appropriate in a particular application. No established rules exist on how to evaluate the most suitable wavelet family for a specific application. Illustratively, the ‘Coifflet’ wavelet family was used although other wavelet families, such as Lenaire-Battle and Symlet may in some implementations be preferable.

Illustratively, the original signal is decomposed into ten (10) frequency bands: 750 Hz-1500 Hz, 375 Hz-750 Hz, 187 Hz-375 Hz, 93 Hz-187 Hz, 46 Hz-93 Hz, 23 Hz-46 Hz, 12 Hz-23 Hz, 6 Hz-12 Hz, 3 Hz-6 Hz and from DC to 3 Hz. As mentioned above, the range of these bands is determined by the sampling frequency (in the case at hand Fs=3000 Hz although a person of skill in the art would understand that a higher or lower sampling rate could be used). An illustrative example of the amplitudes of the samples in the first five (5) frequency bands is shown in FIG. 5. Referring to FIG. 5, frequency bands two (2) and three (3) carry the most important information to estimate respiratory rate. Frequency bands four (4) and five (5) show clearly information related to a beating heart.

The estimation of RR is hindered by several factors such as the non-stationarity and non-linear nature of the respiratory sound signal, the interference of non-biological (60 Hz, environment noise, glitches, etc) and biological signals (heart beat, swallow, cough, speech and others) and recordings with low signal to noise (S/N) ratio. Another problem arises when one of the respiratory phases is significantly stronger than the other and abnormal patterns, for example those which are the result of certain pulmonary diseases, although abnormal patterns are also present in some individuals without these diseases.

As discussed above, apnea is a temporary cessation of the respiratory function. The most widely used criterion as an indication of apnea is 10 seconds or greater of duration for the cessation. In the present proposed method, the duration, or time threshold, of cessation of respiratory function indicating apnea is a configurable parameter. Once the peaks in the envelope signal are being detected, the time interval between two consecutive peaks is estimated and compared with the configured time threshold. If it is greater than the threshold, apnea is flagged as having been detected and an alarm raised,

A special type of apnea occurs when no envelope peaks are detected. To discriminate apnea from a ‘sensor disconnected’ we use the power spectrum analysis. While the sensor is connected to the patient the low biological frequencies (frequencies with higher power values in the band from 200 Hz to 300 Hz) will prevail. If the sensor is disconnected higher frequencies (frequencies with high power values over 500 Hz) will prevail. Additionally, to be certain that no envelope peaks exist, both the root mean squared (RMS) value of the envelope and an envelope history of the previous one minute of signal are retained. If a significant drop of amplitude happens in the 20 sec segment (to less than 12% of the RMS value), then apnea is detected.

In order to overcome the difficulties found when using a single method to produce an estimation of RR with high accuracy, a multiple technique approach is used. As a result, the pre-processed signals are analyzed using both envelope-based and speech-processing related methods as described in more detail herein below.

Envelope Related Methods for Estimating Respiratory Rate

The respiratory sound signal acquired on the tracheal site can be modeled as sinusoidal signals from 200 Hz to 800 Hz modulated by a slow oscillatory signal that represents inspiratory and expiratory envelopes. Illustratively, the envelope is obtained based on a Hilbert transform and decimation of the wavelet filtered sound signal (from 187 Hz to 750 Hz) in a proportion of fifty to one. The envelope is a very low frequency signal that modulates those components of the respiratory sounds located in the band from 187 Hz to 750 Hz. in this regard, decimation means that the envelope signal (obtained with the Hilbert transform) is down sampled to have less data points to process and thus decrease the execution time targeting real-time applications. For example, a respiratory sound sampled at 3 kHz for a duration of 20 seconds corresponds to 60000 data points, which when down sampled is only 1200 data points. The low frequency envelope is detected, followed by the determination of its oscillatory period. Based on this period (time lags between consecutive inhalations or exhalations) the RR that would be accounted for after one minute has elapsed is estimated.

Estimating Respiratory Rate Based on the Fast Fourier Transform (FFT).

The power spectrum is estimated from the detrended and windowed (Harming) envelope signal based on a nonparametric fast Fourier transform (FFT). The magnitude squared of the FFT coefficients formed the power Spectrum. Once the envelope is represented in the frequency domain, typically the component with the second highest peak has the information to correctly estimate the RR result.

Estimating Respiratory Rate Based on Envelope Cotanting

In order to estimate RR based on envelope counting, all possible peaks in the envelope signal of the respiratory sound signal are identified and then the RR computed as a function of the time between consecutive inhalations or exhalations enclosed in the selected segment.

All possible peaks are detected by an analysis of samples that fulfil a criterion of local maximum plus a criterion of stability (amplitude higher than a number of samples before and after the peak, see FIG. 6). In this regard, all peaks detected within a given frame of the signal (illustratively 20 seconds but other lengths are also possible) are passed through heuristic validation method. This validation method selects only the peaks higher than 10% of the amplitude of the higher peak on the given signal. A rule of minimal possible distance between two consecutive peaks is also used. Once all peaks have been determined, the mean of the difference of consecutive acid peaks included in the data segment (illustratively of 20 seconds in length.) is calculated. The inverse of this value multiplied by 60 equals the estimation of RR.

Estimating Respiratory Rate Based on the Autocorriation Function

The autocorrelation function exploits the fact that a periodic signal, even if it is not a pure sine wave, will be similar from one period to the next. This is true even if the amplitude of the signal is changing in lime, provided those changes do not occur too rapidly. Once the autocorrelation function is obtained, the first two peaks are analyzed to select the one with the RR information. Typically, the second peak is the correct choice (but th is is not always so). Samples where one respiratory phase was more accentuated than the other and some other cases were better estimated by the first peak.

Estimating Respiratory Rate Based on Wavelet Transform

The appropriate frequency band to be selected for RR analysis changes according to the actual RR. Therefore guidance is required for the right band selection. This guidance is provided by the RR result of the FFT analysis, which allows choosing typically two, or exceptionally three, possible frequency bands. In these bands, the selection of peaks (based on maxima and minima analyses) and the estimation of RR (two or three) is performed similarly as explained in the method of envelope counting. Finally, the RR closest to the RR estimated by the FFT is taken as the RR estimated by the wavelet method.

Speech Processing Related Methods for Estimating Respiratory Rate

The speech processing approach was used to overcome some limitations of the methods that dealt directly with determining the envelope of the respiratory signal, particularly in low SIN ratio recordings. By combining methods based on the envelope and methods based on the respiratory signal, a better estimation of the RR can be achieved.

Using the speech processing approach, relevant information of the signal under analysis is acquired through the processing of small segments of signals (20 ms duration in this), case) where the statistical properties of the signal are assumed to remain stable (stationarity). The analysis is performed in the frequency domain and the main tool is a short-time FFT technique. The flowchart of FIG. 7 illustrates the approach.

The speech processing approach faces some challenges due to the fact that there are no clear spectral differences between inhalation and exhalation phases recorded at the tracheal site. To overcome these limitations, a pilot signal with a frequency (for example, 1 kHz) which is out of the frequency range of interest (200 Hz to 800 Hz), and having an amplitude at least device the minimum RMS of the respiratory signal is combined with the respiratory signal. During intervals of silence between inhalation and exhalation (i.e. where there is an absence of any respiratory sounds) the pilot signal prevails. Likewise, during inhalation/exhalation phases the respiratory signal prevails. As a result, detection of the pilot signal gives an indication of a silence interval.

An additional measure to help accentuate the difference between respiratory signal and silence is the removal (or attenuation) of biological sounds within the silence interval. This function removes or attenuates biological sounds (mainly heart sounds) found in the silence interval and that were not considered glitches in the pre-processing stage. This processing is based on an adaptive filter technique that takes the respiratory signal contaminated with the heart sounds in a first channel (from 100 to 1500 Hz) and the heart sounds from a second channel (from 1 to 30 Hz, both channels simultaneously recorded) and produces as output a respiratory signal ‘free’ of heart interferences. Respiratory signals combined with the 1 kHz pilot signal provide the input. A FFT is applied to windows of 20 ms and parameters such as power (FFT magnitude squared), centroid (frequency multiplied by power divided by power) and a quadratic detection function (squared frequency multiplied by power) are estimated. All these parameters are used as RR estimators.

As an illustration, FIG. 8 displays on the top panel a preprocessed respiratory sound signal. The middle panel shows a signal produced by the quadratic detection function with peaks indicating the position of zones of silence. The bottom panel represents the autocorrelation function of the signal in the middle panel. The second peak of the autocorrelation is used to estimate the RR.

Finally, a scoring system, comprising a heuristically-based analysis of the individual estimators, is applied to determine the final RR based on the results of the individual estimators.

Scoring System

The final respiratory rate (FR) for a particular segment is determined as a function of the RR as determined by each of the individual estimators (as discussed hereinabove) as well as the final respiratory rate of the previous segment (FR Old). In an illustrative embodiment, the individual estimators are examined and if there is one value of RR which is predominant, FR is set to the predominant RR value. If no value of RR is predominant, but two or three values have equal representation, then the value which is closest to FR Old is selected. Finally, if more than three values have equal representation FR is set to the same value as FR Old.

Heart Rate

Heart sounds are also present among the sounds captured on the trachea site. For the estimation of the respiratory rate they are considered as ‘noise’ and are removed. However, these sounds allow the possibility of estimating heart rate (one of the most important vital signs) easily. We have implemented a second hardware channel with filter settings (20-200 Hz) to enhance the detection of heart sounds and reject all other biological sounds (including respiratory). Applications involving the cardio-respiratory interactions, regulation of the autonomous nervous system on the cardiovascular system and others will be easily targeted. Once the heart sounds are filtered, sound peaks are detected. Based on the inter-peaks timing the heart rate is estimated (beats per minute).

Although the present invention has been described hereinabove by way of an illustrative embodiment thereof, this embodiment can be modified at will, within the scope of the present invention, without departing from the spirit and nature of the subject of the present invention.

Claims

1-20. (canceled)

21. A method for estimating a respiratory rate of a patient, the method comprising:

receiving respiratory sound data from at least one acoustic sensor, the respiratory sound data representing respiratory sounds of a patient;

deriving a plurality of respiratory rates from the respiratory sound data with a processor using a plurality of different respiratory rate estimating methods, wherein the plurality of different respiratory rate estimating methods comprises at least one time domain method and at least one frequency domain method; and

determining an estimated respiratory rate based at least in part on the plurality of respiratory rates; and

outputting the estimated respiratory rate.

22. The method of claim 21, wherein the at least one time domain method comprises determining a first respiratory rate of the plurality of respiratory rates using a time domain representation of the respiratory sound data, and the at least one frequency domain method comprises determining a second respiratory rate of the plurality of respiratory rates using a frequency domain representation of the respiratory sound data.

23. The method of claim 21, wherein the at least one time domain method comprises determining a first envelope of the respiratory sound data in a time domain representation of the respiratory sound data and deriving a first respiratory rate of the plurality of respiratory rates based at least on a frequency of the first envelope or a period of the first envelope.

24. The method of claim 23, wherein the at least one frequency domain method comprises determining a second envelope of the respiratory sound data in a frequency domain representation of the respiratory sound data and deriving a second respiratory rate of the plurality of respiratory rates based at least on a frequency of the second envelope or a period of the second envelope.

25. The method of claim 21, wherein the at least one frequency domain method comprises determining a second envelope of the respiratory sound data in a frequency domain representation of the respiratory sound data and deriving a second respiratory rate of the plurality of respiratory rates based at least on a frequency of the second envelope or a period of the second envelope.

26. The method of claim 21, wherein the estimated respiratory rate comprises one of the plurality of respiratory rates.

27. The method of claim 21, wherein said determining the estimated respiratory rate comprises determining the estimated respiratory rate based at least on a comparison between a previously estimated respiratory rate and the plurality of respiratory rates.

28. The method of claim 21, wherein at least some of the same set of samples of the respiratory sound data are used to derive a first respiratory rate of the plurality of respiratory rates and a second respiratory rate of the plurality of respiratory rates, the first respiratory rate derived using the at least one time domain method and the second respiratory rate derived using the at least one frequency domain method.

29. The method of claim 21, wherein the at least one acoustic sensor comprises one acoustic sensor.

30. The method of claim 21, wherein the at least one acoustic sensor is coupled to a neck of the patient.

31. An apparatus for estimating a respiratory rate of a patient, the apparatus comprising:

an input configured to receive respiratory sound data from at least one acoustic sensor, the respiratory sound data representing respiratory sounds of a patient; and

a processor configured to: derive a plurality of respiratory rates from the respiratory sound data using a plurality of different respiratory rate estimating methods, wherein the plurality of different respiratory rate estimating methods comprises at least one time domain method and at least one frequency domain method; determine an estimated respiratory rate based at least on the plurality of respiratory rates; and output the estimated respiratory rate.

32. The apparatus of claim 31, wherein the at least one time domain method comprises determining a first respiratory rate of the plurality of respiratory rates using a time domain representation of the respiratory sound data, and the at least one frequency domain method comprises determining a second respiratory rate of the plurality of respiratory rates using a frequency domain representation of the respiratory sound data.

33. The apparatus of claim 31, wherein the at least one time domain method comprises determining a first envelope of the respiratory sound data in a time domain representation of the respiratory sound data and deriving a first respiratory rate of the plurality of respiratory rates based at least on a frequency of the first envelope or a period of the first envelope.

34. The apparatus of claim 33, wherein the at least one frequency domain method comprises determining a second envelope of the respiratory sound data in a frequency domain representation of the respiratory sound data and deriving a second respiratory rate of the plurality of respiratory rates based at least on a frequency of the second envelope or a period of the second envelope.

35. The apparatus of claim 31, wherein the at least one frequency domain method comprises determining a second envelope of the respiratory sound data in a frequency domain representation of the respiratory sound data and deriving a second respiratory rate of the plurality of respiratory rates based at least on a frequency of the second envelope or a period of the second envelope.

36. The apparatus of claim 31, wherein the estimated respiratory rate comprises one of the plurality of respiratory rates.

37. The apparatus of claim 31, wherein the processor is configured to determine the estimated respiratory rate based at least on a comparison between a previously estimated respiratory rate and the plurality of respiratory rates.

38. The apparatus of claim 31, wherein the processor is configured to use at least some of the same set of samples of the respiratory sound data to derive a first respiratory rate of the plurality of respiratory rates and a second respiratory rate of the plurality of respiratory rates, the first respiratory rate derived using the at least one time domain method and the second respiratory rate derived using the at least one frequency domain method.

39. The apparatus of claim 31, wherein the at least one acoustic sensor comprises one acoustic sensor.

40. The apparatus of claim 31, wherein the at least one acoustic sensor is coupled to a neck of the patient.