WO2010040973A2

WO2010040973A2 - Method and apparatus for analysing biomedical signals

Info

Publication number: WO2010040973A2
Application number: PCT/GB2009/001239
Authority: WO
Inventors: Raúl Jiménez CASTRO; Felipe Ignacio Donoso Urrutia; Gustavo Andrés Soto MUSTER; Claudia Alejandra Graciela Jiménez QUINTANA
Original assignee: Universidad De Concepcion; Butler, Michael, John
Priority date: 2008-10-10
Filing date: 2009-05-18
Publication date: 2010-04-15

Description

Method and Apparatus for Analysing Biomedical Signals

The present invention relates to the analysis of biomedical electrical signals, such as may be obtained in electrocardiography, electromyography or electroencephalography, for example.

One aspect of the invention is particularly concerned with the analysis of signals that have been obtained over a long period of time, whether from a patient in a clinic over a period of a few hours, for example or from a patient over a more extended period such as 24 hours whilst performing normally during the day, such as through the use of ambulatory electrocardiography device, sometimes referred to as a "Holter monitor".

The spectral analysis of a biomedical signal frequently consists of a detailed study of the behaviour of its harmonic components in the frequency domain. To obtain a simple spectrum of a function a Fourier Transform can be used. However, most biological signals are non-stationary, so that their spectrum varies over time. For example, electrocardiogram ("ECG") signals are signals in the time domain that can be analyzed in the frequency domain by using the Fourier Transform. However, an ECG is often measured over a long period of time in relation to the intrinsic fluctuations in the cardiovascular system. That is to say that heart beats rarely possess exactly the same frequency, and as time passes it is possible to detect changes in both the frequency of the beats as well as in their form. In addition, these variations are clearer in the case of certain cardiac alterations. Because of this, if using Fourier analysis for a complete ECG signal in which there are variations over time, the amount of information included in this spectrum will be excessive and therefore very difficult to interpret.

One object of this aspect of the invention is to enable easier interpretation of signals that have been obtained over an extended period of time. Thus, viewed from one aspect of the present invention, there is provided a method of analysing biomedical signals that have been received over a period of time, using data processing equipment, comprising the steps of storing data representing the amplitude of the signals as a function of time; allocating the data into a plurality of consecutive segments of time; in respect of each segment carrying out a transformation into a frequency domain representation and storing data representative of energy as a function of frequency within that segment, and displaying graphically a representation of segment identifiers and, for each segment, data representative of energy as a function of frequency.

Preferably, the segments of time overlap.

In accordance with embodiments of this aspect of the invention, a method of processing signals in the time / frequency domain makes it possible to synthesize or concentrate long term recordings by separating the signal data into short periods or segments and then applying a Fourier Transform, for example. In the specific case an ECG signal, it is possible to synthesize data to obtain more precise information, so as to observe the evolution of the patient during the period of study and to identify visually a number of different deviations from standard behaviour of the heart, which might indicate auricular fibrillation, ventricular tachycardia, supraventricular tachycardia, and extrasystoles, as well as observing normal sinus rhythm. This makes it easier to carry out diagnosis, to verify previous diagnoses and to monitor of patients over time with or without treatment indicated by a doctor.

In an embodiment of this aspect of the invention, there is provided output in the form of a diagram in which the spectral energy density of an ECG signal is shown. Typically there may be provided a pseudo two dimensional graph, in which the x and y axes correspond to segment identifiers and frequency, and the energy spectrum is represented by different visual artefacts. For example, different densities of spots could be used, or different shapes of markings, but preferably different energies are represented by different colours. For example there could be a background colour such as white which indicates that the energy value is practically zero, and a range of colours for increasing energy values. This could be, for example, the visible spectrum with red at one extreme and blue at the other, and other colours such as green, yellow, orange and intermediate colours in between.. It is a matter choice whether blue or red indicates the highest energy level, but in a preferred embodiment blue indicates low-to-medium energy values and red indicates high energy values. A colour bar may be provided alongside the graph to assist in identifying energy levels.

In a preferred embodiment the method segments the signal into slices of time of equal length that can be analyzed separately, so there is performed a dynamic analysis in the frequency domain. The signal can be separated into segments, or time intervals, of for example about 10 to about 15 seconds, to which a Fast Fourier Transform algorithm (FFT) can be applied.

In general, the time extent of a segment may be, for example, at least about 1 second; at least about 5 seconds; at least about 10 seconds; no greater than about 60 seconds; no greater than about 30 seconds; between about 1 and about 60 seconds; between about 1 and about 30 seconds; between about 5 and about 60 seconds; between about 5 and about 30 seconds; between about 10 and about 60 seconds; between about 10 and about 30 seconds; between about 5 and about 20 seconds; between about 10 and about 20 seconds; between about 5 and about 15 seconds; between about 10 and about 15 seconds; or about 15 seconds. The time extent of a segment is preferably constant, although varying length segments could be used.

The period of time over which the biomedical signals have been received may be, for example, at least about 15 minutes; at least about 20 minutes; at least about 30 minutes; at least about 45 minutes; at least about 60 minutes; at least about 75 minutes; at least about 90 minutes; at least about 2 hours; at least about 6 hours; at least about 9 hours; at least about 12 hours; at least about 15 hours; at least about 18 hours; at least about 21 hours; or at least about 24 hours. - A -

There may be displayed graphically at one time data in respect of a period of time of at least about 15 minutes; at least about 20 minutes; at least about 30 minutes; at least about 45 minutes; at least about 60 minutes; at least about 75 minutes; at least about 90 minutes; or at least about 2 hours. In general, embodiments of the invention are intended to provide a long range visualisation of the data received from, for example, an electrocardiogram device over an extended period.

If the ECG or other signal to be analyzed is very long, for example spanning up to several hours, it is possible to separate it into shorter periods - such as from 30 to 60 minutes, obtaining a set of results for the long term signal. For example, a 6-hour signal can be represented in twelve representations of 30 minutes each.

The graphical display of information may, for example, display data for frequencies up to no more than about 50 Hz; no more than about 40 Hz; no more than about 30 Hz; or no more than about 25 Hz.

The sampling rate may be, for example, less than about 500 Hz; no more than about 475 Hz; no more than about 450 Hz; no more than about 425 Hz; no more than about 400 Hz; or no more than about 375 Hz. The sampling rate may be, for example, no less than about 100 Hz; no less than about 125 Hz; no less than about 150 Hz; or no less than about 175 Hz.

The representation of segment identifiers may be direct, for example being an ID for a segment such as its number, and depending on the scale at which the representation is viewed indications may be given for individual segments, or for intervals of segments, such as every five, ten and so forth. The representation of segment identifiers may be indirect, for example by giving an indication of time. Again, depending on the scale at which the representation is viewed may be given for individual instances of time, or for intervals of time, such as every minute and so forth. To carry out the analysis of each segment, the segment is preferably windowed to avoid the introduction of undesirable frequency components in the FFT result. By using overlapping segments, information is not lost from the beginning and/or the end of each segment. In preferred embodiments, the result obtained from the application of FFT to each segment is shown on a graph using outlines or level curves, thus allowing a compact, continuous visualization with less noise for the whole signal.

One embodiment of this aspect of the invention provides a method of processing biomedical electrical signals in the time-frequency domain in order to synthesize or concentrate long term recordings, by means of the separation of a digital input signal into short periods or segments. The method may include providing a digital input signal and converting the digital input signal to obtain a signal with a standardized format. The signal, after standardisation if appropriate, may be pre-processed to obtain a noise-free signal with normalized amplitude values, for example to reduce the effects of patient breathing or other movements, for example. There may be the facility to configure the system, for example to define the size of the segment in fixed time units and the degree of overlap of segments, for example expressed as a percentage. The signal, after pre-processing if desired, is divided into predefined fixed time segments. The signal separated into segments may be stored in a matrix S of size M x N, with M being the number of segments and N the number of points per segment. The type of window to be applied in each segment maybe incorporated. A type of window may be applied over each segment or line of the matrix S. There is obtained the spectral power or energy density / distribution within each segment based on the FFT (Fast Fourier Transform) algorithm. These values may be stored in a matrix X, in which each line is the spectral power / energy density of the corresponding line in the matrix S. The number of outline levels and the frequency interval for visualization are incorporated. There are generated outlines or level curves, for example, from the matrix X in the defined frequency interval. There is obtained an "LRV" (Long-Range Visualization) diagram for the level curves generated, one axis of which describes the index i of each segment, and the other the frequency within the defined interval. A third "dimension" includes the intensity of the spectral power / energy density represented by a colour bar that goes, for example, from blue in the lower part to red in the upper part, passing through light blue, green, yellow and orange, in which the colour bar represents a percentage value corresponding to the portion of energy of each point with respect to the total energy of each segment.

The most commonly used tool to obtain the spectrum of a signal is the Fourier Transform (FT). To additionally achieve a time localization of the spectral components, other transforms are used, which provide a time-frequency representation of the signal. The STFT (Short Time Fourier Transform) and the WT (Wavelet Transform) belong to this special group of transforms. Other transforms are used to produce a temporal localization of the spectral components, providing a time- frequency representation of the signal. The STFT (Short Time Fourier Transform) and the CWT belong to this special group of transforms. A preferred embodiment of the invention can be understood as a particular example of STFT, with, for example, a Hann window and overlap. The preferred embodiment, at least, makes it possible to study the frequency behaviour of a biomedical signal of long duration, distinguishing different types of cardiac alterations such as atrial fibrillation, ventricular tachycardia, extra systoles, and artefacts, among others. In addition, it makes it possible to compare the frequency behaviour of different patients with the same diagnosis and short recordings , such as 1 minute.

This aspect of the invention may be expressed in a number of forms. For example, this aspect of the invention also provides data processing apparatus for analysing biomedical signals that have been received over a period of time, the data processing apparatus being configured to store data representing the amplitude of the signals as a function of time; to allocate the data into a plurality of segments of time which overlap; in respect of each segment to carry out a transformation and to store data representative of energy as a function of frequency within that segment, and to display graphically a representation of segment identifiers and, for each segment, data representative of energy as a function of frequency. This aspect of the invention also provides a computer program product containing instructions which when carried out on data processing apparatus will configure the data processing apparatus so that the apparatus is adapted for analysing biomedical signals that have been received over a period of time, the data processing apparatus being configured to store data representing the amplitude of the signals as a function of time; to allocate the data into a plurality of segments of time which overlap; in respect of each segment to carry out a transformation and to store data representative of energy as a function of frequency within that segment, and to display graphically a representation of segment identifiers and, for each segment, data representative of energy as a function of frequency. This aspect of the invention also provides a method of diagnosing abnormal heart behaviour by analysing signals that have been received over a period of time by an electrocardiograph machine, using data processing equipment, comprising the steps of storing data representing the amplitude of the signals as a function of time; allocating the data into a plurality of consecutive segments of time; in respect of each segment carrying out a transformation and storing data representative of energy as a function of frequency within that segment, displaying graphically a representation of segment identifiers and, for each segment, data representative of energy as a function of frequency, and analysing the displayed information to identify abnormal heart behaviour.

A further aspect of the invention concerns the analysis of an electrocardiographic signal in space: the time, signal and first derivative of the signal. More specifically, it is concerned with a method of visualization of the changes in the form of the QRS wave complexes, and, depending on the electrical alteration, changes in the form of the P and T waves can also be visualized. In addition, this aspect makes it possible to know the derivative in each point of the electrocardiogram (ECG). As these changes, both in form and in the values of the derivatives, can be associated with heart disease, the preferred field of application of this aspect is cardiology.

In the mathematical discipline of nonlinear population dynamics, a bi-dimensional graphic representation called a phase profile is known, in which one axis is the function and the other axis is the first derivative of this function. In general, these studies have focused on, for example, determining the behaviour of two competitive species (for example, foxes and rabbits) with regard to calculating the number of individuals of each species which will enable both species to survive or cause one of them to disappear. This phase profile coincides with one of the three projections of the spatial curve over one of the planes: the signal versus its derivative. Currently, the study of electrocardiographic signals only considers the maximum derivative of the ECG zone, which goes from the minimum of the Q wave to the maximum of the R wave, and in other cases the maximum positive and negative dV/dt of the QRS are considered. These values are of clinical importance, but they are difficult to obtain, and very often their calculation is very approximate. In accordance with this aspect of the invention, these values can be obtained readily, as well as the value of the slopes in each point of the ECG. In addition, the evolution of the changes can be visualized in the forms of the different waves that compose the ECG, above all that of the QRS complex. These changes are associated with electrocardiographic alterations and thus this aspect of the invention offers medical specialists a simple graphical method that makes it possible to visualize these changes and additionally quantify them into mV/s units (millivolts divided by seconds).

This aspect of the invention provides a method of simple spatial visualization of the changes in the forms of ECG waves, especially the QRS complex, and additionally makes it possible to know the derivative in each point of the ECG. The three spatial dimensions are: a time axis, for example measured in seconds; an amplitude (voltage) axis, for example measured in millivolts (mV); and finally, the slope axis, measured in millivolts divided by seconds.

In accordance with this aspect of the invention this spatial curve is projected over the three Cartesian planes, bi-dimensional images are obtained corresponding to the following curves: the original ECG, the curve derived from the ECG and the bi- dimensional orbits. In this orbit projection more and better information can be obtained from the ECG. The curves that form the QRS have fluctuations that are not visible in the ECG. This is contrary to the conventional belief that the QRS complex is formed by straight lines, and therefore the maximum dV/dt values that are used clinically are not constant, as has been thought previously thought. Additionally, in the orbital plane the maximum positive and negative derivatives can be calculated (increasing and decreasing part of the QRS), as well as in any other point of the ECG. It is also possible to calculate the variation in voltage of the maxima of R and the minima of Q or S.

Viewed from another aspect, the invention thus provides a method of analysing biomedical signals using data processing equipment, comprising the steps of storing data representing the amplitude of the signals as a function of time, establishing data representing the value of dV/dt where V is the amplitude and t is the time, and generating a spatial curve to represent the relationship between amplitude V, time t, and dV/dt.

This aspect of the invention may be used alone, or in combination with other aspects of the invention as described above. This aspect of the invention may also be expressed as a data processing system configured to carry out the method, a computer program product adapted to configure the data processing method to carry out the method, and a method of diagnosis using the method. As with the other aspect of the invention, this aspect is of particular, but not exclusive benefit in the context of electrocardiography.

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

Figure 1 shows a general schematic view of a system in accordance with the invention;

Figure 2 is a diagram showing steps in a method according to the invention;

Figure 3 A shows an example of an ECG signal without filtering;

Figure 3B shows the example of the ECG signal in Figure 2A with filtering; Figure 4 shows an example of the generation of segments with an overlap.

Figure 5 shows a window for three consecutive segments of 15 seconds with an overlap of 50%.

Figure 6 is an example of an output display;

Figure 7 A shows an example from Figure 5;

Figure 7B shows another example from Figure 5;

Figure 7C shows a third example from Figure 5;

Figure 8 shows a general schematic of a process in accordance with another aspect of the invention;

Figure 9 A shows a portion of an ECG recording used in accordance with this aspect of the invention;

Figure 9 B shows the transformation of the 5 pulses of Figure 9 A as a continuous curve in space;

Figure 10 A presents the projection of the spatial orbits in the XY plane;

Figure 10 B shows the projection of the spatial orbits in the XZ plane;

Figure 11 A shows a pulse indicating some points where the dV/dt derivative is zero;

Figure 11 B shows the orbit of the pulse of Figure 11 A;

Figure 12 A shows a normal ECG; Figure 12 B shows the 2D orbits of the ECG of Figure 12 A;

Figure 13 A shows an ECG in which an arbitrary point is taken as a reference;

Figure 13 B shows this point in the ECG derivative;

Figure 13 C, shows this point in a two dimensional orbit; and

Figure 13 D, shows this point in a three dimensional orbit.

Referring now to Figure 1 there is shown a first patient 1 on a bed 2 in a clinic or hospital. Electrodes 3 are connected to an electrocardiograph machine which monitors the signals for a period of, say, a few hours and stores them. The data is subsequently sent via a data link 5 to a data processing system 6 which includes a main processing unit 7 containing components such as a processor and memory, a display monitor 8, and a keyboard / mouse 9 as input devices. A second patent 10 is shown wearing a portable Holter monitor 11 , which can monitor the patients hear for a period of say 24 hours, whilst at home, work and so forth. At the end of the monitoring period the patent goes to a clinic or hospital where there is an interface 12 that reads the data from the monitor 11 and transmits the data over a data link 13 to the data processing system 6. hi both cases the data is manipulated by the data processing unit in the manner discussed bellow. The data processing system is configured to operate in the required manner by software. This is stored on non- volatile memory in the main processing unit and can have been supplied originally on physical media 14 such as a DVD or CD which is loaded into a drive 15, or for example as signals transmitted over the Internet or another network from a remote server 16.

The general mode of operation using the data processing unit 6 is summarized in Figure 2. The signal to be analyzed corresponds to the ECG signal obtained from the measuring instrument i.e., the electrocardiograph machine 4 or the Holter unit 11. The signal is generally delivered as a digital signal, but it is often in a format chosen by the manufacturer of the instrument, with different calibrations and voltage levels. For this reason it is necessary to convert the signal delivered by the instrument to obtain a standardized digital signal at step 17. The format chosen for the signal may be an ASCII code archive, in which the data correspond to strings of numbers separated by a suitable character such as a space, punctuation mark or the return code.

Each string corresponds to a signal value that is interpreted as an element in the vector of data corresponding to the ECG signal. Each element of this vector is given by the readings taken by the instrument 4 or 11, at a predetermined sample frequency that allows the calculation of the time associated with the signal under study. These signals possess a variable noise content that depends on the different measuring conditions and on the quality of the instrument, so that a pre-processing step 18 is necessary to achieve a noise-free signal with normalized amplitude values, making it possible to obtain the LRV diagram with reliable and useful results.

Filtering the signal should eliminate low frequency components caused by patient breathing and movements, noise components due to the electric grid, and high frequency components that are of no interest for the purpose of this invention. Figure 3A shows an ECG signal segment without filtering and Figure 3B shows the signal segment after filtering. Normalization is carried out by extracting the continuous component of the signal; that is, by establishing a zero average and fixing the variance to one.

The signal is separated into fixed time segments at step 19.. The time is defined by the user, using the input components such as the keyboard and mouse 9 and a configuration facility. There is a default value of 15 seconds in this embodiment. The segmentation is carried out in relation to the duration of each segment and the overlap percentage, which is also defined by the user but which has a 50% default value. This means that between two consecutive segments there is an overlap of 50% of the segment size. In the case of 15_rsecond segments, that means that a segment begins in the 7.5 seconds of the previous segment, as shown in Figure 4.

The result of the separation of the signal into segments (7) is stored in a matrix S, in which each line corresponds to a segment, and the number of columns corresponds to the length of each segment in points. This number of points is defined according to the sample frequency and duration of each segment. In the case of 15 second segments: N, number of points per segment = T x f_s, where T corresponds to segment duration time and f_s to the sample frequency of the ECG signal. Therefore the resulting matrix S has the dimension M x N, with M = number of segments and N = number of points per segment.

The algorithm used for the separation of the signal into segments taking overlap into consideration is the following:

1. Calculation of the number of overlap points:

T = P x N / 100

where P is the overlap percentage.

2. Generation of segments:

Si = [ ECG(S_-1) X (N-T)+! ECG(J-J) _x (N_-τ)₊N ]

where ECG is the signal to be analyzed and the subscripts indicate the number of the initial point and final point, respectively, for the i ^th segment. The value of i varies from 1 to M, where M is given by: M = Integer { n/(N-τ) - 1 }, with n the total points of the ECG signal.

The overlap allows the application of a window to each segment that makes it possible to reduce the injection of high frequencies onto the spectrum of the signal analyzed. Various windows are known in the art, among which are; the Hamming window, Hann window, Kaiser window, rectangular window, and triangular window. The Hann window is chosen for the present embodiment, because of its positive response with regard to the attenuation of undesirable components. Figure 5 shows an example of the Hann window form and its distribution over the time axis in the case of 15 second segments with 50% overlap. It can be observed that when curves (similar to Gaussian distribution curves) are superimposed on the ECG signal, the latter is attenuated at the edges of each segment. However, with 50% overlap, the gain is not lower than 0.5, which ensures signal analysis without losing the information at the edges of the segments.

The window chosen must be applied on each segment or line of the previously defined matrix S, so that then the FFT algorithm can be applied, and the spectral power density of each segment can finally be obtained (step 24). This operation is as follows:

1. Generation of window segments :

S_wi = Si * W

where * is multiplication point by point, Sj is the i^th segment (i^th line of S), W is the chosen window and S_wj is the i^th window segment.

2. Fourier spectrum calculation:

Fi = FFT { S_wi }

where FFT is the Fast Fourier Transform and F, is the Fourier spectrum of the i^th window segment.

3. Calculation of the spectral energy (power) density: X₁ = I Fi I ² =;?_<?{ F₁ }² + /m{Fi }²

where Xj is the spectral power density of the i^th segment, Re indicates the real part and Im indicates the imaginary part of Fj.

Each Xj vector is stored as the i^th line in a matrix X, producing in this way a matrix in which the lines contain the spectral power densities of the corresponding lines of the matrix S. Therefore, both the matrix S, containing the ECG signal separated into segments (step 23) and the matrix X, containing the spectral power density by segment, (step 24) are delivered as output variables.

From the X matrix the outline contour graph is generated (step 21) or a level curve graph that corresponds to the LRV (Long Range Visualization) diagram (22). This graph will have the index / of each segment along one axis and the frequency corresponding to each point of the vector X; along the other axis, which is determined by the sampling frequency. These outlines have been graphed in the frequency interval [0, 30] Hz, but these values can be modified by the user and another range of interest can be chosen.

Figure 6 shows an example of an output display for a 10 minute-signal. The colour bar (25) on the right, which goes from blue in the lower part to red in the upper part, passing through light blue, green, yellow and orange, indicates the intensity of each point on the graph. These values are in percentages and correspond to the energy portion of each point with respect to the total energy of each segment. For example, if a point is light blue in colour, corresponding to a value of 5 in the colour bar (25), this means that 5% of the total energy of the segment is concentrated in this point.

Figure 6 highlights an ECG signal with auricular fibrillation from segment 1 to segment 35, where it passes to a sinus rhythm until segment 43, and finally passes to a sinus rhythm with auricular extrasystoles. The frequency range chosen was between 0 and 25 Hz. As above, this value has practically no harmonic content. Figure 7 A shows segment 20, in which the fibrillation can be clearly observed in the tracing. Figure 7B shows the tracing for segment 40 and Figure 7C shows the tracing for segment 44. In these figures, the spectral power density can be seen, which is noisy and irregular in the case of auricular fibrillation, while for the sinus rhythm it is a discrete spectrum with little noise. It is possible to see this clearly in the LRV diagram in Figure 6, where a drastic change in the representation is observed, beginning in segment 35. Small changes in cardiac frequency can also be seen, so that the evolution of the patient throughout the whole duration of the signal is visible.

It will be appreciated that the features of the invention and embodiments thereof are not limited to signals from an electrocardiograph, nor to biomedical signals.

Some embodiments of the invention can be considered as providing a method of processing signals in the time frequency domain to synthesize or concentrate long term recordings by means of the separation into short periods or segments of a digital input signal that includes: providing a digital input signal; converting the digital input signal to obtain a signal with a standardized format; pre-processing the signal with a standardized format to obtain a noise-free signal with normalized amplitude values; defining the size of the segment in fixed time units and incorporating the overlap percentage; separating the pre-processed signal into predefined fixed time segments; storing the signal separated into segments in a matrix S of size M x N, with M number of segments and N number of points per segment; applying a type of window on each segment or line of the matrix S; obtaining the spectral power density of each segment based on the FFT (Fast Fourier Transform) algorithm; storing the spectral power density values in a matrix X; incorporating the number of outline levels and the frequency interval for visualization; generating outlines or level curves from the matrix X in the defined frequency interval; and obtaining an LRV (Long-Range Visualization) diagram for the level curves generated. Some embodiments of the invention can be considered as providing a method of processing signals in the time-frequency domain to synthesize or concentrate long term recordings, by means of the separation into short periods or segments of a digital input signal. The method is characterised in that it includes: a) providing a digital input signal; b) converting the digital input signal to obtain a signal with a standardized format; c) pre-processing the signal with a standardized format to obtain a noise-free signal with normalized amplitude values; d) defining the size of the segment in fixed time units and the overlap percentage; e) separating the pre- processed signal into predefined fixed time segments; f) storing the signal separated into segments in a matrix S of size M x N, with M number of segments and N number of points per segment; g) incorporating the type of window to be applied in each segment; h) applying a type of window on each segment or line of the matrix S; i) obtaining the spectral power density of each segment based on the FFT (Fast Fourier Transform) algorithm; j) storing the spectral power density values in a matrix X, in which each line is the spectral power density of the corresponding line in the matrix S; k) incorporating the number of outline levels and the frequency interval for visualization; 1) generating outlines or level curves from the matrix X in the frequency interval defined; and m) obtaining a LRV (Long-Range Visualization) diagram for the level curves generated, one axis of which, e. g. the horizontal, describes the index i of each segment and the other axis of which, e. g. the vertical, the frequency within the defined interval, and a third dimension is provided that includes the intensity of the spectral power density represented by a colour, with a colour bar optionally being provided which represents a percentage value corresponding to the energy portion of each point with respect to the total energy of each segment.

Such a method may be characterised in that it provides a digital signal that includes an ECG (electrocardiogram) signal obtained from a measuring instrument for this type of signal, for example, an electrocardiograph or a Holter.

The method may be further characterised in that converting the digital signal includes choosing an archive in ASCII code, in which the data correspond to strings of numbers separated by a character such as that representing a "space", "comma" or "enter".

The method may be further characterised in that each string corresponds to a signal value, which is interpreted as a vector data element corresponding to the ECG signal, where each element of this vector is given by the readings taken by the instrument, at a predetermined sample frequency that makes possible to calculate the time associated with the signal under study.

The method may be further characterised in that pre-processing the signal with a standardized format to obtain a signal free of noise includes filtering the signal to eliminate low frequency components due to patient breathing and movements, noise components from the electric grid, and high frequency components.

The method may be further characterised in that pre-processing the signal with a standardized format with normalized amplitude values includes normalization extracting the continuous component of the signal, that is, establishing zero average and fixing the variance to one.

The method may be further characterised in that segmenting the pre-processed signal into predefined fixed-time segments includes a predefined time for a user or a default value of, for example, 15 seconds.

The method may be further characterised in that incorporating the overlap percentage includes a predefined value for the user or a default value of 50%, that is, that between two consecutive segments there is an overlap of 50% of segment size.

The method may be further characterised in that separating the signal into segments includes segmenting the signal into slices of time of equal size that can be analyzed separately in order to carry out a dynamic analysis in the frequency domain. The method may be further characterised in that the calculation of the number of overlapping points is given by τ = P x N / 100

where P is the overlap percentage and N is obtained in the following way: N = Number of points per segment = T x f_s Where T corresponds to the duration time of the segments and f_s to the sample frequency of the digital input signal.

The method may be further characterised in that generation of segments is given by:

Si = [ ECG(M) x (N-τ)+l ECG(H) x (N-τ)+N _

The method may be further characterised in that defining a type of window includes selecting a Hamming window or a Harm window or a Kaiser window or a rectangular window or a triangular window.

The method may be further characterised in that applying a type of window on each segment or line of the matrix S is given by: Sw, *= Si * W

where * is multiplication point by point, Sj is the i^th segment (i^th line of S), W is the chosen window and S_Wi is the i^th window segment.

The method may be further characterised in that calculation of the spectral energy (power) density is given by:

Xi = | Fi | ² =Λ2{ Fi }² + /m{Fi }² where Xj is the spectral power density of the i^th segment, Re indicates the real part and Im indicates the imaginary part of Fj.

The method may be further characterised in that incorporating the number of outline levels includes fixing the number of level curves that are displayed in the LRV diagram.

The method may be further characterised in that incorporating the frequency interval for visualization includes focalizing the display of a range of frequencies selected by the user or a default range of [0, 30] Hz, in the LRV diagram.

It will be appreciated that the further optional characteristic features of the method given in the preceding paragraphs may apply not only to the statement of method preceding them, but also to there other aspects of the invention given above.

A method in accordance with the second aspect of the invention will now be described, in which there is a Fast Orbital Transformation ("FOT").

Figure 8 shows a general schema of the generation process of the FOT 3D, highlighting the input parameters and the output variables in addition to the input signal.

Figure 9 A shows a little more than 10 seconds of an ECG recording with five QRS complexes, indicated by R₁, R₂, R₃, R₄ and R₅, during a coronary angioplasty. In Figure 9 B can be seen the transformation of these five pulses as a continuous curve in space, in which the previous QRS complexes are now "spatial orbits", indicated by the same symbols as in Figure 9 A. Three axes can be distinguished in the spatial representation: time, measured in seconds corresponding to the X axis; voltage, measured in mV corresponding to the Y axis; and the voltage derivative, measured in mV/s corresponding to the Z axis.

Figure 10 A presents the projection of the spatial orbits in the XY plane, that is, as in the original ECG. Figure 10 B shows the projection of the spatial orbits in the XZ plane, that is, the curve representing the ECG derivative is visible. Figure 10 C shows the projection of the spatial orbits in the YZ plane, that is, the "bi- dimensional orbits" of the original ECG can be seen. As the derivative is zero in the maxima and in the minima, on the right side, marked by the symbols R₁, R₂, R₃, R₄ and R₅. are shown the maximum positive dV/dt in the straight line dV/dt = 0. On the left, the maximum negative dV/dt are shown.

Figure 11 A presents a pulse indicating some points where the dV/dt derivative = 0, where A corresponds to the beginning of the P wave, B to the maximum value of the P wave, C to the maximum value of the R wave and D to the maximum value of the T wave. Figure 11 B shows the 2 D orbit of this pulse. Given that there are two axes, the voltage and its derivative, time appears as a parameter and is related to the direction of the path of the curve, indicated by arrows. In this case the changes in QRS and in P and T waves, respectively, are clearly observed.

Figure 12 A visualizes a normal ECG for a period of 10 seconds. Figure 12 B shows the 2D orbits of this ECG. In these 10 seconds can be seen the variation, indicated by numbers, of the maximum positive (and negative) dV/dt [1], the variation in the values of R_max [2], the variation in the Q_mj_n [3], and in this case the variations in the P waves [4] and T waves[5].

Figure 13 A shows an ECG in which an arbitrary point is taken as a reference (t = 2.008 s and V = 0.902 mV) . The system of this aspect of the invention makes it possible to see this same point in the ECG derivative, Figure 13 B, in the 2D orbit, Figure 13 C, and in the 3D orbit, Figure 13 D. In addition, the software allows the calculation of the time corresponding to this point to a thousandth of a second (t=2.080 s), the value in.mV of the ECG in this point (0.902 mV), and the value of the ECG derivative in this point (131.010 mV/s).

An embodiment of this aspect of the present invention provides a method of simple spatial visualization of the changes in the forms of ECG waves, especially the QRS complex, and additionally makes it possible to know the derivative in each point of the ECG. The three spatial dimensions are: the time axis, the voltage axis, and the slope axis as shown in Figure 9B. If this spatial curve is projected over the three Cartesian planes, bi-dimensional images are obtained corresponding to the following curves: the original ECG (see Figure 10 A), the curve derived from the ECG (see Figure 10 B) and the bi-dimensional orbits, which may be referred to as

"FOT 2D" (see Figure 10 C). In this FOT 2D projection more and better information can be obtained from the ECG. In effect it is possible to visualize the fact that the curves that form the QRS have fluctuations that are not visible in the ECG. This simple fact invalidates the belief that the QRS complex is formed by straight lines and therefore the maximum dV/dt values that are used clinically are not constant, as was previously thought (see Figure 12 B). Additionally, in the FOT 2D plane the maximum positive and negative derivatives can be calculated (increasing and decreasing part of the QRS), as well as in any other point of the ECG. It is also possible to calculate the variation in voltage of the maxima of R and the minima of Q or S.

A proposed FOT 3D embodiment is summarized in Figure 1. The signal to be analyzed is an ECG digital signal obtained from any specific instrument for obtaining an ECG, whether an electrocardiograph or another instrument of the Holter type, for example, and the general layout can be as described with reference to Figure 1. Generally the signal is delivered as a digital signal, many of which are in the format determined by the manufacturer of the instrument, with different calibrations and voltage levels. For this reason it is necessary to adjust the signal delivered by the instrument to obtain a digital signal with a standardized format (101). The format chosen for the signal is an archive in ASCII code, in which the data correspond to strings of numbers separated by "space", "commas", or "carriage return". Each string corresponds to a value of the signal, which is interpreted as an element of the data vector corresponding to the ECG signal. Each element of this vector is given by the readings taken by the instrument in increasing order, at a predetermined sampling frequency that permits the calculation of the time associated with the signal under study. Additionally, these signals possess a variable noise content that depends on the different measuring conditions and on the quality of the instrument, which means that it is necessary to carry out pre-processing (102) in order to obtain a signal free of noise, so that the process for obtaining the FOT 3D can be applied with reliable and useful results.

Signal filtering is designed to eliminate low frequency components due to breathing, muscular movement (EMG), unexpected movements from the patient, noise components due to the electric grid, and high frequency components of no interest for the purpose of the invention.

To display the FOT 3D on the screen, it is necessary to previously obtain information on the data of the signal (S) that has already been filtered, among which are the sampling frequency at which it was acquired (Fs) and the number of points in the signal (Ix) . Once this information has been established, the signal derivative can be calculated (103). To calculate the derivative, in this embodiment the following algorithm is used:

dS = S(2 \ h) -s(\ : lx -\)

where S is the signal obtained from the pre-processing in (102) and Ix is the length of the signal (the number of points). The derived signal is stored in a column vector (dS) .

In order to improve the visualization in three dimensions and make it more user- friendly, 10 points are then added between consecutive samples by means of a cubic interpolation (104) for the signal and its derivative. The cubic interpolation of the signal (S) is stored in a new vector called spSl and the cubic interpolation of the derived signal (dS) is stored in a new vector called (spdSl) . Because of the cubic interpolation, the point density of each (S) and (dS) signal increases significantly, the sampling frequency is constant, and therefore the sampling time too. This is the reason why it is necessary to re-calculate the vector, so that for each point added there is a corresponding correct time. This is done with the following formula: time = [1 : lengtk(spSV) ] ■ - —

I* s 10 where length(spSΪ) is the number of points of the interpolation of the signal (S) derivative, not forgetting that the derived signal (dS) has fewer points than the original signal because of the algorithm used. The factor — gives the time in

Fs seconds and the factor — allows the time vector to be maintained within the same 10 values, even when additional points have been added.

Finally, once the number of time points (time), the interpolated signal (spSΪ) and its interpolated derivative (spdSΪ) have been adjusted, this new technology is displayed graphically in the following way:

X axis = lime Y axis — spSl Z axis — spάSl • Fs

In the X axis the time vector is shown in seconds, re-calculated in function of the number of points obtained after interpolation. In the Y axis the interpolated signal is shown in millivolts (mV), and finally in the Z axis the interpolation of the signal derivative multiplied by the sampling frequency is shown in millivolts divided by seconds (mV/s).

In addition to the process by which the FOT 3D is obtained, the signal derivative (dS) , its interpolation (spdS) , and the FOT 2D are obtained as output variables (106).

It will be appreciated that the features of this embodiment may be applicable to a range of biomedical signals, and indeed to the analysis of other signals.

Claims

CLAMS

1. A method of analysing biomedical signals that have been received over a period of time, using data processing equipment, comprising the steps of storing data representing the amplitude of the signals as a function of time; allocating the data into a plurality of consecutive segments of time; in respect of each segment carrying out a transformation into a frequency domain representation and storing data representative of energy as a function of frequency within that segment, and displaying graphically a representation of segment identifiers and,' for each segment, data representative of energy as a function of frequency.

2. A method as claimed in claim 1, wherein the transformation is a Fourier transformation.

3. A method as claimed in claim 1 or 2, wherein the segments overlap.

4. A method as claimed in claim 1, 2 or 3, wherein a window is applied to each segment.

5. A method as claimed in any preceding claim, wherein the time covered by a segment is predetermined and constant for each segment.

6. A method as claimed in any preceding claim, wherein there is displayed a graph with segment identifiers along one axis and frequency along the other axis, and different visual artefacts to indicate the energy at any point on the graph.

7. A method as claimed in claim 6, wherein the visual artefacts are colours.

8. A method as claimed in any preceding claim, wherein the biomedical signals are signals from an electrocardiograph device.

9. A method as claimed in any preceding claim, comprising the further step of establishing data representing the value of dV/dt where V is the amplitude and t is the time, and generating and displaying a spatial curve to represent the relationship between amplitude V, time t, and dV/dt.

10. A method as claimed in any preceding claim, wherein the time extent of a segment is at least 1 second.

11. A method as claimed in any preceding claim, wherein the time extent of a segment is no greater than 60 seconds.

12. A method as claimed in any preceding claim, wherein the time extent of a segment is selected from one of the following values:

at least 5 seconds; at least 10 seconds; no greater than 30 seconds; between 1 and 60 seconds; between 1 and 30 seconds; between 5 and 60 seconds; between 5 and about seconds; between 10 and 60 seconds; between 10 and 30 seconds; between 5 and 20 seconds; between 10 and 20 seconds; between 5 and 15 seconds; between 10 and 15 seconds.

13. A method as claimed in any preceding claim, wherein the biomedical signals have been received over a period of time of at least 15 minutes.

14. A method as claimed in any preceding claim, wherein the biomedical signals have been received over a period of time selected from one of the following values:

at least 20 minutes; at least 30 minutes; at least 45 minutes; at least 60 minutes; at least 75 minutes; at least 90 minutes; at least 2 hours; at least 6 hours; at least 9 hours; at least 12 hours; at least 15 hours; at least 18 hours; at least 21 hours; at least 24 hours.

15. A method as claimed in claim 13 or 14, wherein there is displayed graphically at one time data in respect of a period of time of at least 15 minutes.

16. A method as claimed in claim 15, wherein there is displayed graphically at one time data in respect of a period of time selected from one of the following:

at least 20 minutes; at least 30 minutes; at least 45 minutes; at least 60 minutes; at least 75 minutes; at least 90 minutes; or at least 2 hours.

17. A method as claimed in any preceding claim, wherein the graphical display of information may, for example, display data for frequencies up to no more than 50 Hz; no more than 40 Hz; no more than 30 Hz; or no more than 25 Hz.

18. A method as claimed in any preceding claim, wherein the biomedical signals are obtained using a sampling rate which is: less than 500 Hz; no more than 475 Hz; no more than 450 Hz; no more than 425 Hz; no more than 400 Hz; or no more than 375 Hz.

19. A method as claimed in any preceding claim, wherein the biomedical signals are obtained using a sampling rate which is: no less than 100 Hz; no less than 125

Hz; no less than 150 Hz; or no less than 175 Hz.

20. Data processing apparatus for analysing biomedical signals that have been received over a period of time, the data processing apparatus being configured to store data representing the amplitude of the signals as a function of time; to allocate the data into a plurality of segments of time which overlap; in respect of each segment to carry out a transformation into a frequency domain representation and to store data representative of energy as a function of frequency within that segment, and to display graphically a representation of segment identifiers and, for each segment, data representative of energy as a function of frequency.

21. A computer program product containing instructions which when carried out on data processing apparatus will configure the data processing apparatus so that the apparatus is adapted for analysing biomedical signals that have been received over a period of time, the data processing apparatus being configured to store data representing the amplitude of the signals as a function of time; to allocate the data into a plurality of segments of time which overlap; in respect of each segment to carry out a transformation into a frequency domain representation and to store data representative of energy as a function of frequency within that segment, and to display graphically a representation of segment identifiers and, for each segment, data representative of energy as a function of frequency.

22. A method of diagnosing abnormal heart behaviour by analysing signals that have been received over a period of time by an electrocardiograph machine, using data processing equipment, comprising the steps of storing data representing the amplitude of the signals as a function of time; allocating the data into a plurality of consecutive segments of time; in respect of each segment carrying out a transformation into a frequency domain representation and storing data representative of energy as a function of frequency within that segment, displaying graphically a representation of segment identifiers and, for each segment, data representative of energy as a function of frequency, and analysing the displayed information to identify abnormal heart behaviour.

23. A method of analysing biomedical signals using data processing equipment, comprising the steps of storing data representing the amplitude of the signals as a function of time, establishing data representing the value of dV/dt where V is the amplitude and t is the time, and generating and displaying a spatial curve to represent the relationship between amplitude V, time t, and dV/dt.

24. A method as claimed in claim 23 wherein the biomedical signals are signals from an electrocardiograph device.

25. Data processing equipment configured to carry out the method of claim 23 or 24.

26. A computer program product carrying instructions adapted to program data processing equipment to carry out the method of claim 25.

27. A method of diagnosing abnormal heart behaviour using a method as claimed in claim 24.