WO2005096170A1

WO2005096170A1 - Ecg-signal analysis for shock outcome prediction

Info

Publication number: WO2005096170A1
Application number: PCT/GB2005/000477
Authority: WO
Inventors: Paul Stanley Addison; James Nicholas Watson
Original assignee: Cardiodigital Limited
Priority date: 2004-03-31
Filing date: 2005-02-10
Publication date: 2005-10-13
Also published as: GB0407293D0

Abstract

The presented invention predicts the outcome of defibrillation attempts for out-of-hospital cardiac arrest patients. In the presented method, the continuous wavelet transform, modulus maxima techniques and temporal statistical analysis are applied to preshock segments of ECG trace for patients in ventricular fibrillation (VF). Time-frequency markers are extracted from the transform and a linear threshold derived from a training set to provide high sensitivity prediction of successful defibrillation. The use of the wavelet marker is associated with a high specificity value at high sensitivities in comparison to previously reported methods. It therefore provides an improved index for the identification of patients for whom shocking would be futile, and for whom an alternative therapy could be considered.

Description

ECG-SIGNAL ANALYSIS FOR SHOCK OUTCOME PREDICTION

The present invention relates to a method of signal analysis, and in particular to a method for the prediction of the likelihood of successful defibrillation for subjects exhibiting ventricular fibrillation; and an associated signal analysis apparatus and defibrillator .

BACKGROUND ART

Despite improvements in the rapidity with which shocks are delivered and the shock characteristics themselves, results achieved from the treatment of ventricular fibrillation remain sub-optimal. For the past two decades, therefore, efforts have been made to characterise the VF waveform in an attempt to optimise shock delivery and outcome. Recent experimental [i,ii,iii] and clinical [iv] studies have indicated that Cardiopulmonary Resuscitation (CPR) can increase the likelihood of successful defibrillation for established VF. Further studies have suggested that delaying CPR for a defibrillation attempt may cause a dramatic decrease in the likelihood of defibrillation success [v] . These results are consistent with recent clinical studies [vi,vii] which indicate that pre-shock CPR can improve the rates of return of spontaneous circulation (ROSC) and survival to hospital discharge when emergency medical services (EMS) response times exceeded 4-5 minutes. With this in mind, the need for an effective predictor of shock outcome becomes apparent . Such a technology would enable the emergency responder to provide a therapy tailored to the patient's needs. The ^λshock or CPR' question becomes answerable for an individual rather than being a statistically-based heuristic within a standard protocol. The electrocardiogram (ECG) derived from a patient's skin surface during ventricular fibrillation (VF) was, until recently, considered to represent the unstructured electrical activity of the heart. It has been variously described in literature as random, noisy, and chaotic. Techniques used to analyse the VF waveform include: amplitude [viii,ix] which has not proved to be a reproducible marker of defibrillation success [x,xi], Fourier- based spectral analysis [xii, iii, xiv, xv, xvi] [US Patent 5,683,424: Brown et al., US Patent 6,171257 Bl: Weil et al] and techniques from non-linear dynamics such as fractal [xvii,xviii] [US Patent 6,438,419: Callaway et al] and phase-delay [xix] which in practice can often be shown to be related to previously investigated methods [xx] . Wavelet transform analysis is especially valuable because of its ability to elucidate simultaneously local spectral and temporal information from a signal [xxi] . An improvement in the predictability of shock outcome is observed using wavelet techniques due to this fundamental difference between the wavelet transform and alternative prior art previously mentioned. While these alternatives, in some form or another, characterize an aspect of the behaviour of the signal over a period of time, however short, the wavelet method can identify pertinent information in time. Hence, temporal partitioning of salient aspects of the transformed ECG becomes achievable prior to any classification step. The identification of coherent structure in VF signals using wavelet transform-based techniques has been previously reported [xxii, xxiii, xxiv] [International Patent Application WO 01/82099 Al : Addison and Watson] . In WO 01/82099, modulus maxima techniques are used after CPR is initiated in order to distinguish portions of a wavelet scalogram that are affected by the CPR from those that are unaffected. In all this prior art, modulus maxima techniques are not used at all for the analysis of the VF scalogram when no CPR is applied. SUMMARY OF THE INVENTION According to the present invention, there is provided a method of signal analysis comprising the steps of deriving an electrocardiogram (ECG) signal; deriving a wavelet scalogram of the ECG; analysing the scalogram using ridge following techniques; and deriving an output from the analysis, said output representing a temporal population statistic of the ECG signal.

Preferably, the ECG signal is derived from a subject whose heart is in Ventricular Fibrillation (VF) or Ventricular Tachycardia (VT) .

Preferably, said ridge following techniques include modulus maxima techniques.

Preferably, atypical coefficient values pertaining to artefact are identified and removed.

Preferably, said analysis includes temporal statistical analysis, wherein temporal features are derived from a temporal population statistic calculated over one or more of the scalogram scales.

Optionally, the temporal statistical analysis includes an entropy measure of the form

Preferably, said entropy measure is computed from coefficients associated with wavelets with characteristic frequencies above that of the typical fibrillating frequency of the heart.

Optionally, said population statistics includes the median coefficient value of a number of the highest coefficient values from one or more of the scalogram scales.

Preferably, said population statistics includes the median coefficient value of the three highest values.

Preferably, said population statistics are computed from coefficients associated with wavelets with characteristic frequencies above that of the typical fibrillating frequency of the heart.

Preferably, said temporal population statistic is used as an indication on the likelihood of the future success of a defibrillation event.

According to a second aspect of the present invention, there is provided a method of selectively delivering a defibrillation shock to a subject whose heart is in Ventricular Fibrillation (VF) or Ventricular Tachycardia (VT) comprising the steps of a. connecting electrodes to a patient; b. deriving analogue input signals from said electrodes to derive the electrocardiogram (ECG) ; c. sampling said ECG to derive a digitised signal; d . deriving the wavelet scalogram of said ECG; e. analysing said wavelet scalogram using ridge following techniques to determine the likely outcome of defibrillation; and f . guiding the resuscitation procedure accordingly.

Preferably, the step (e) of analysing said wavelet scalogram is performed in accordance with the method of the first aspect.

According to a third aspect of the present invention, there is provided apparatus for signal analysis comprising: sensor means suitable to derive an electrocardiogram (ECG) signal from a subject; signal processing means suitable for deriving a wavelet scalogram from the ECG; and an outcome prediction unit suitable for performing ridge following techniques on the scalogram and to derive an output from the analysis, said output representing a temporal population statistic of the ECG signal.

According to a fourth aspect of the present invention, there is provided a defibrillator for selectively delivering a defibrillation shock to a subject whose heart is in Ventricular Fibrillation • (VF) or Ventricular Tachycardia (VT) comprising the apparatus of the third aspect. Preferably, the defibrillator further comprises a user interface which includes a display means.

Preferably, the outcome prediction unit is arranged to perform the method of the first aspect.

Preferably, the defibrillator further comprises decision means arranged to selectively apply either a first method according to the first or second aspects of the invention in a case where the defibrillator' s computational power exceeds a predetermined threshold, or a second method according to the first or second aspects of the invention in a case where the defibrillator' s computational power is less than or equal to a predetermined threshold. The current invention provides an improved method for predicting the immediate success of a defibrillation attempt. Here, novel markers obtained from the wavelet transform scalogram for use in the prediction of shock outcome are defined and the methodology for their derivation specified. The wavelet transform of a signal x (t) is defined as

where ψ (t) is the complex conjugate of the wavelet function ψ (t) , a is the dilation parameter of the wavelet and b is the location parameter of the wavelet. The scalogram is the time-scale half-space generated by plotting \ T (a ,b) \ , the modulus of wavelet transform coefficient value, for varying scales and. locations. The scalogram may also be said to mean any suitably scaled power of \ T (a ,b) | .

A key advantage of wavelet techniques is the variety of wavelet functions available thus allowing the most appropriate to be chosen for the signal under investigation. This is in contrast to Fourier analysis which is restricted to one feature morphology: the sinusoid. In some embodiments the Morlet wavelet is used. This is defined as:

ψit) = π^'λ' (e¹²*' - _e-ⁱ²*^Ϋ/2)e-^2/2 _[2]

where f₀ is the characteristic, or central, frequency of the mother wavelet and is chosen to best accentuate the features under investigation.

Using wavelet transforms temporal behaviour of signal features can be quantified from the scalogram. These temporal features can be derived from any intermittency measure calculated over one or more of the scalogram scales. The efficacies of these measures are enhanced through the reduction of the scalogram to its turning points in b for each scale a: the modulus maxima of the scalogram. In the preferred embodiment a novel wavelet-entropy marker is used as a metric of the temporal behaviour of the signal . The wavelet-entropy at a scale a ' is defined as :

Where, |T(a^F,b) | are the wavelet transform modulus values at scale a ' .

The scales over which the marker is calculated and the intermittency metric derived is dependent upon • the design characteristics of the defibrillator, such as analogue ECG signal conditioning (e.g. band-pass filtering, comb filtering) ; digital sampling rate; electrode size; electrode location; skin/electrode interface resistance. In the preferred embodiment a priori knowledge of these characteristics are used to identify optimal processing paths for calculating the marker. Typically the scales from which the metric is extracted will be of the order of that associated with a central frequency of around 45Hz.

From the marker value either: a probability of successful defibrillation can be derived using standard techniques (e.g. Bayesian) or a simple threshold rule can be applied to identify whether defibrillation should be attempted. Both techniques require empirical data to identify the best course of action. In the preferred embodiment a linear threshold is derived from historical data to identify those patients that would benefit from defibrillation, with 95% certainty, while the remainder receive an alternative therapy (e.g. CPR).

BRIEF DESCRIPTION OF THE DRAWINGS

The present invention will now be described, by way of example only, with reference to the accompanying drawings, in which:

Figure 1 is a schematic diagram of a defibrillator with reference to the current invention. The COP analysis block returns the likelihood of successful defibrillation to the embedded controller having previously been passed a digitised ECG to analyse;

Figure 2 is a flow diagram outlining the therapeutic decision process with respect to the additional information supplied by the invention;

Figure 3 shows raw ECG data with its associated wavelet scalogram beneath (a) and the modulus maximal plot of this scalogram (b) with its filtered equivalent beneath (c) ;

Figure 4 is a flow diagram of the invention's methodology for identifying the most effective therapy to be applied; Figure 5 shows the Receiver Operator Characteristic (ROC) curve resulting from the high computation route of the invention's method. (Specificity 62%±2% at sensitivity 97%±3%) ; and

Figure 6 shows the Receiver Operator Characteristic (ROC) curve resulting from the low computation route of the invention's method. (Specificity 61%+4% at sensitivity 97%±3%)

Figure 1 shows a schematic block diagram indicating the usage of an outcome prediction unit (5) within a defibrillator (2) . The analogue electrocardiogram (ECG) of a patient in ventricular fibrillation (1) is detected through sensors (3) and monitoring circuit (9) where it is digitised. The embedded controller (4) passes selected regions of the ECG to the outcome prediction unit (5) for analysis. In the preferred embodiment these regions will be of around 5 seconds in length with a sample rate of at least 100Hz and be from an artefact free region of trace. The outcome prediction unit (5) passes back an indication of the likelihood of successful defibrillation either as a probability of successful defibrillation or as a direct command to defibrillate.

Additional information provided by the outcome prediction decision process (11) can be introduced into the resuscitation protocol as shown in Figure 2. Here, instead of all shockable rhythms being defibrillated, only those patients likely to benefit from this therapy are shocked. Hence the outcome prediction decision process (11) provides an additional conditional (14) allowing a route to CPR (12) rather than defibrillation (15) .

In one embodiment the continuous wavelet transform of equation 1 is applied employing the Morlet wavelet of equation 2. Figure 3(a) shows an example of a section of digitised ECG with its associated wavelet scalogram beneath. The dark islands in the plot (a selection of which are shown at (28) ) indicate the location and morphology of high energy features within the ECG. Temporal measures such as that described by equation 3 can be applied to the scalogram at this stage. In a preferred embodiment the step of reducing the scalogram to its coefficient turning points in scales across time, as defined by equation 4 below, is carried out. An example of this so called Modulus Maxima is shown in Figure 3(b) .

[4] = 0 otherwise Using modulus maxima reduces the degree of redundancy in the raw data making its analysis less computationally complex and hence less time consuming. These turning points can either in a first outcome prediction method be used directly in the derivation of markers for the prediction of defibrillation success or, in a second outcome prediction method, be parameterised into continuous ridges across scales for advanced feature identification and artefact filtering prior to marker derivation. The second method is a more computationally complex task and may prove prohibitively complex for a given de ibrillator . Hence, an outcome prediction unit can be arranged to carry out one or other of the first and second methods, depending on the computing power of the defibrillator that carries out the method. Alternatively, an outcome prediction unit can be arranged to selectively operate the first or second method, as shown in figure . The path taken is dependent upon the available computing power; conditional (16) in figure 4. Where computing power (17) allows, a full scalogram (18) is generated from the passed ECG (20) over 150 scales ranging from those wavelets with a central frequency of 1Hz to those wi-th a central frequency of 45Hz (19) . It will be appreciated that a different number of scales over a different frequency range could also be considered. The modulus maxima for each scale are derived and the turning points characterised as continuous ridges across scales (21) . There exist many standard techniques for ridge following. In a preferred embodiment non-zero coefficient points in successive scales are defined as belonging to "the same continuous ridge where their amplitude values are of the same order (for example within 10%) and when the higher dilation point is located within the temporal support of the lower dilation point. Where two or more points satisfy these criteria the points on successive scales that are closest to each other in location are deemed to be of the same continuous ridge. Once the ridges have been parameterised in this way ridges associated with individual features in time may be analysed. By way of example figure 3(c) shows the five longest ridges of the modulus maxima scalogram of figure 3 (b) . Ridges containing coefficient amplitudes atypical to the remainder of the scalogram may be assumed to be associated with artefact and removed. Shorter, low amplitude, ridges with components only in the low dilation (i.e. top) region of the scalogram may be assumed to be electrical noise and removed. The remainder of the scalogram may then be analysed without the loss in performance associated with such noise. There are many alternative statistical measures that may be applied to predict outcome from the filtered ridge scalogram either individually or in combination. These may include: entropy values across scales as equation 3, or peak, median and mean dilation values, and energy values of the filtered ridge scalogram. The chosen measure will reflect the characteristics of the defibrillator within which the method is embodied. The derived wavelet marker (s) may then be used for the prediction of the likelihood of successful defibrillation. They may be applied directly or indirectly following a statistical optimisation procedure such as Linear Discriminant Analysis (LDA) or Principal Component Analysis (PCA) . Any probabilistic classifier or heteroassociative function approximation method may be employed at this stage to generate a system capable of predicting a probability of defibrillation outcome. In a preferred embodiment the median coefficient value of a scale above that of the typical fibrillating frequency of the heart for the 3 longest ridges is taken as a marker to indicate the likelihood of successful defibrillation (25) . Typically this scale is associated with wavelets with a central frequency of 45Hz. It will be appreciated that the median value of a different number of ridges, for example five, may be chosen. Further, in a preferred embodiment the method of characterisation is that of a linear threshold classifier trained on previously collected data. When the marker value exceeds a threshold derived empirically from historic data (27) a decision (26) is taken to defibrillate. Where the power required to execute the high computational complexity path of the method is prohibitive to the defibrillator in which the method is embodied a low computational complexity path may be followed (16) . Here, a limited scalogram (22) is generated from the passed ECG (20) over a single scale above that of the typical fibrillating frequency of the heart (23). Typically this scale is associated with wavelets with a central frequency of 45Hz. The modulus maxima for this scale is then derived (24) . In a preferred embodiment no filtering of the scale coefficients occurs in this case although those skilled in the art will appreciate that atypical coefficient values may be removed at this stage if assumptions are made concerning their association with noise. Furthermore, the entropy measure of equation 3 can be applied on the turning points to indicate the likelihood of successful defibrillation. There are many alternative statistical measures that may be applied to predict outcome from scale turning points either individually or in combination. These may include: peak, median, mean or sum of all the coefficient values of the scale or percentile thereof. The chosen measure will reflect the characteristics of the defibrillator within which the method is embodied. As before, when the marker value exceeds a threshold derived empirically from historic data (27) a decision (26) is taken to defibrillate .

EXAMPLE OF METHOD PERFORMANCE

This example of the invention's efficacy uses a human out-of-hospital data set containing 878 pre- shock ECG traces all of at least 10 second duration from 110 patients with cardiac arrest of cardiac etiology. The data was recorded from the Medical Device Module of a Laerdal Heartstart 3000 defibrillator. A full review of the data acquisition procedure and statistics can be found in [v] . For this example we define ^successful defibrillation' as those attempts which result in a pulse and co-ordinated electrical activity sustained for a period greater than thirty seconds and originating within a minute of the applied shock. For this example we have used all data as provided whilst limiting our system to hands-free periods of trace (i.e. the period immediately preceding the shock during which no CPR is applied) . We limit the number of shocks from each patient to the first six so reducing the chance of the system efficacy statistics becoming loaded due to intra-patient correlation. Parametric investigation shows the optimal length of preshock trace for this defibrillator is around 5 seconds with a position immediately prior to shock. Each ROC curve in figures 5 and 6 contains standard error bars. Each study has been cross validated by repetition using a different subset of the data to parameterize and subsequently test the system each time. That is, the classifying boundary for a given required sensitivity is derived using training data (to parameterize) and evaluated using the withheld test data iteratively. Any patient with data in a training set, does not have data in the test set for that cross-validation iteration because of the likelihood of intra-patient correlation. It has also been ensured that the distribution of patients per cross-validation set, shocks per set, and shocks per patient in the sets remain broadly the same. Figure 5 shows the receiver operator characteristic (ROC) curve indicating system performance when using the high computational complexity path of the method. Here the system performance was obtained though the median coefficient value from the 3 longest filtered ridges at the scale associated with a central frequency of around 45Hz. The central frequency of the mother wavelet (the previously described f₀) is 0.40 in this case. The system performance achieved in this example is a specificity of 62%±2% at sensitivity 97%±3%. Thus for this example 490 of the 791 unsuccessful defibrillation attempts may have been avoided and an alternative therapy such as CPR applied with 97% of those patients which would benefit from defibrillation still receiving this therapy. Figure 6 shows the ROC curve indicating system performance when using the low computational complexity path of the method. Here the system performance was obtained though the wavelet entropy value of the scale associated with a central frequency of around 45Hz. The central frequency of the mother wavelet is 0.87 in this case. The system performance achieved in this example is: specificity of 61%±4% at sensitivity 97%±3%. Thus for this example 483 of the 791 unsuccessful defibrillation attempts may have been avoided and an alternative therapy such as CPR applied with 97% of those patients which would benefit from defibrillation still receiving this therapy.

Various improvements and modifications can be made to the above without departing from the scope of the invention.

OTHER REFERENCES CITED

[i] Berg R A, Hilwig R W, Kern K B et al. Precountershock cardiopulmonary resuscitation improves ventricular fibrillation median frequency and myocardial readiness for successful defibrillation from prolonged ventricular fibrillation: a randomised, controlled swine study. Ann E erg Med 2002; 40: 563-571. [ii] Achleitner U, Wenzel V, Strohmenger HU et al. The beneficial effect of basic life support on ventricular fibrillation mean frequency and coronary perfusion pressure. Resuscitation. 2001: 51:151-158. [iii] Kolarova J, Ayoub IM, Yi Z et al . Optimal timing for electrical defibrillation after prolonged untreated ventricular fibrillation. Crit Care Med. 2003; 31: 2022-2028. [iv] Eftestol T, Wik L, Sunde K, Steen PA. Effects of CPR on predictors of VF defibrillation success during out-of-hospital cardiac arrest. Circulation 2004; 110: 10-15. [v] Sunde K, Eftestol T, Askenberg C, Steen P A. Quality assessment of defibrillation and advanced life support using data from the medical control module of the defibrillator. Resuscitation 1999; 41: 237-247. [vi] Wik L, Hansen TB, Fylling F, Steen T, Vaagenes P, Auestad B H, Steen PA. Delaying defibrillation to give basic cardiopulmonary resuscitation to patients with out-of-hospital ventricular fibrillation - A randomized trial. JAMA 2003; 289 (11): 1389-1395. [vii] Cobb LA, Fahrenbruch CE, Walsh TR et al . Influence of cardiopulmonary resuscitation prior to defibrillation in patients with out-of-hospital ventricular fibrillation. JAMA H999; 1182-1188. [viii] Weaver W D, Cobb L A, Dennis D et al. Amplitude Of Ventricular Fibrillation Waveform And Outcome After Cardiac Arrest. Ann Intern Med 1985; 102: 53-55.

[ix] Stutts K R, Brown D D, Kert>er R E. Ventricular Fibrillation Amplitude Predicts Ability To Defibrillate Am J Emerg Med 1986; 4: 423. [x] Ewy G A, Dahl C F, Zimmerman et al : VF Masquerading As Ventricular Standstill. Crit Care Med 1981; 9: 392 [xi] Jones D L, Klein G J. Ventricular Fibrillation: The Importance Of Being Coarse. J Electrocardiol 1984; 17: 393-400. [xii] Brown C G, Griffith, R F, ^"Van Lighten P, Hoekstra et al. Median Frequenc y—A New Parameter For Predicting Defibrillation Success Rate. Ann Emerg Med 1991; 10: 787-89. [xiii] Strohmenger H-U, Lindner K H, Keller A, et al. Effects Of Graded Doses Of Vasopressin On Median Fibrillation Frequency In A Porrcine Model Of Cardiopulmonary Resuscitation: Results Of A Prospective, Randomized, Controlled Trial. Crit Care Med 1996; 24: 1360-65. [xiv] Strohmenger H-U, Lindner IK H, Prengel A W, et al. Effects Of Epinephrine And Vasopressin On Median Fibrillation Frequency And Defi-brillation Success In A Porcine Model Of Cardiopulmonary Resuscitation. Resuscitation 1996; 31: 65-73.

[xv] Frossard M, Schδrkhuber W, Sterz F, et al. Ventricular Fibrillation Waveform Characteristics And Outcome After Cardiac Arrest: Preliminary Results. 1994 SAEM Annual Meeting, May 9-12, Washington, DC, USA Academic Emergency Medicine 1:A92, 1994. [xvi] Povoas H P, Weil M H, Tang W, Bisera J, Klouche K, and Barbatsis A. Predicting the success of defibrillation by electrocardiographic analysis. Resuscitation 2002; 53(1): 77-82. [xvii] Mann A, Achleitner U, Antretter H, Bonatti J O, Krismer A C, Lindner K H, Rieder J, Wenzel V, Voelckel W G and Strohmenger H-U. Analysing ventricular fibrillation ECG-signals and predicting defibrillation success during cardiopulmonary resuscitation employing (α) -histograms . Resuscitation 2001; 50() : 77-85. [xviii] Callaway C W, Sherman L D, Mosesso V N, Dietrich T J, Holt E, Clarkson C. Scaling Exponent Predicts Defibrillation Success for Out-of-Hospital Ventricular Fibrillation Cardiac Arrest. Circulation, 2001; 103: 1656-1661. [xix] Sherman L D, Flagg A, Callaway C W, Menegazzi J J and Hsieh M. Angular velocity: a new method to improve prediction of ventricular fibrillation duration. Resuscitation 2004; 60(1): 79-90. [xx] Watson J N, Addison P S, Steen P A, Robertson C E and Clegg G R. Angular velocity: a new method to improve prediction of ventricular fibrillation duration - CORRESPONDENCE. Resuscitation 2004; 62(1) : 122-123.

[xxi] Addison P S. The Illustrated Wavelet Transform Handbook: Introductory Theory and Applications in Science, Engineering, Medicine and Finance. Institute of Physics Publishing 2000; Bristol, UK. [xxii] Addison P S, Watson J N, Clegg G R, Holzer M, Sterz F and Robertson C E. A novel wavelet based analysis reveals hidden structure in ventricular fibrillation, IEEE Engineering_/ in Medicine and Biology 2000;19(4): 383-392.

[xxiii] Watson J N, Addison P S,- Clegg G R, Holzer M, Sterz F and Robertson C E. Evaluation of arrhythmic ECG signals using a novel wavelet transform method. Resuscitation 2000; 43(2): 121--127.

[xxiv] Addison P S, Watson J N_r Clegg G R, Steen P A, Robertson C E. Finding Coordinated atrial activity during ventricular fibrillation using wavelet decomposition. IEEE Engineering in Medicine and Biology 2002; 21: 58-65.

Claims

1. A method of signal analysis comprising the steps of deriving an electrocardiogram (ECG) signal; deriving a wavelet scalogram of the ECG; analysing the scalogram using ridge following techniques; and deriving an output from the analysis, said output representing a temporal population statistic of the ECG signal.

2. The method of claim 1, wherein the ECG signal is derived from a subject whose heart is in Ventricular Fibrillation (VF) or Ventricular Tachycardia (VT) .

3. The method of claim 1 or claim 2, wherein said ridge following techniques include modulus maxima techniques.

4. The method of claim 3, wherein atypical coefficient values pertaining to artefact are identified and removed.

5. The method of any preceding claim, wherein said analysis includes temporal statistical analysis, wherein temporal features are derived from a temporal population statistic calculated over one or more of the scalogram scales.

6. The method of claim 5, wherein the temporal statistical analysis includes an entropy measure of

7. The method of claim 6 wherein said entropy measure is computed from coefficients associated with wavelets with characteristic frequencies above that of the typical fibrillating frequency of the heart .

8. The method of claim 5, wherein said population statistics includes the median coefficient value of a number of the highest coefficient values from one or more of the scalogram scales

9. The method of claim 8, wherein said population statistics includes the median coefficient value of the three highest values .

10. The method of claim 8 or claim 9, wherein said population statistics are computed from coefficients associated with wavelets with characteristic frequencies above that of the typical fibrillating frequency of the heart.

11. The method of any preceding claim, wherein said temporal population statistic is used as an indication on the likelihood of the future success of a defibrillation event.

12. A method of selectively delivering a defibrillation shock to a subject whose heart is in Ventricular Fibrillation (VF) or Ventricular Tachycardia (VT) comprising the steps of a. connecting electrodes to a patient; b. deriving analogue input signals from said electrodes to derive the electrocardiogram (ECG) ; c. sampling said ECG to derive a digitised signal; d. deriving the wavelet scalogram of said ECG; e . analysing said wavelet scalogram using ridge following techniques to determine the likely outcome of defibrillation; and f . guiding the resuscitation procedure accordingly .

13. The method of claim 12, wherein the step (e) of analysing said wavelet scalogram is performed in accordance with any of claims 1 to 11.

14. Apparatus for signal analysis comprising: sensor means suitable to derive an electrocardiogram (ECG) signal from a subject; signal processing means suitable for deriving a wavelet scalogram from the ECG; and an outcome prediction unit suitable for performing ridge following techniques on the scalogram and to derive an output from the analysis, said output representing an intermittency measure of the ECG signal.

15. A defibrillator for selectively delivering a defibrillation shock to a subject whose heart is in Ventricular Fibrillation (VF) or Ventricular Tachycardia (VT) comprising the apparatus of claim 14.

16. The defibrillator of claim 15, further comprising a user interface which includes a display means .

17. The defibrillator of claim 15 or claim 16, wherein the outcome prediction unit is arranged to perform the method of any of claims 1 to 13.

18. The defibrillator of claim 15 or claim 16, further comprising decision means arranged to selectively apply either the method of any of claims 6 to 7 in a case where the defibrillator' s computational power exceeds a predetermined threshold, or the method of any of claims 8 to 10 in a case where the defibrillator' s computational power is less than or equal to a predetermined threshold.