 |
Description  |
|
|
BACKGROUND OF THE INVENTION
I. Field of the Invention:
This invention relates generally to biomedical apparatus; and more
particularly to a system for detecting the onset of tachyarrhythmias, such
as ventricular fibrillation whereby effective intervention can be
initiated to restore the patient to normal sinus rhythm.
II. Discussion of the Prior Art:
To better comprehend the prior art, it is deemed necessary to explain the
physiologic characteristics of the heart which might be used for
distinguishing between normal sinus rhythm (SR), ventricular tachycardia
(VT), supra-ventricular tachycardia (SVT) and ventricular fibrillation
(VF). In SR, there is a synchronized depolarization of the cardiac cells
resulting in the conventional QRS electrogram waveform. In VF, however,
the cardiac cells generally lose their synchrony and the depolarization of
one cell or group of cells no longer bears any particular relationship to
the depolarization of other cell groupings resulting in a loss of distinct
rhythm. Not only does the rhythm become indistinct but morphological
changes also occur such that the beats are no longer of uniform shape due
to the fact that the number of cells participating in each independent and
nonsynchronized depolarization also changes.
Known VT and VF sensing techniques have involved the use of a bipolar or
unipolar catheter disposed proximate the right ventricular apex for
detecting signals occasioned by the cell depolarization and the use of a
counting algorithm in an attempt to distinguish between SR, VT and VF on
the basis of pulse rate. In connection with that algorithm, VF is
considered to be a heart rate that is greater than some certain value
typically exceeding rates associated with VT.
It is also found that the transition from SR to either VT or VF is
generally accompanied by a fairly severe change in morphology or pulse
shape because the overall pattern of depolarization and timing between
cell changes when passing from SR through the onset of VT. A number of
pulse counting techniques described in the literature depend not only upon
determining the rate of the tachycardia, but also on the so-called "rate
of onset". That is to say, they not only assess the absolute pulse rate
but how rapidly the tachycardia begins relative to sinus rhythm.
Investigators have determined that absolute rate alone is not a reliable
indicator of VT or SVT in that high SR may be physiologically appropriate
due to physical activity or even sudden fright resulting in autonomic
stimulus to the heart. The "rate-of-onset measure" is generally based upon
a threshold criteria determined by how rapidly the interval between
successful QRS complexes is changing. Even when rate-of-onset
considerations are relied upon in conjunction with absolute rate, known
prior art detection apparatus still may mistake physiologically
appropriate tachycardias from the non-appropriate tachycardias, i.e.,
there is an overlap in the domain which could contribute to inappropriate
intervention. Moreover, pulse morphology changes have frequency
extensions, as well, which may affect detection circuit performance. This
may occur precisely at the critical point where rate-of-change information
is needed for rate determination. Thus rhythm discrimination techniques
which depend on rate and rate-of-change measurements may fail on two
counts: rate overlap and misidentified rate changes.
Another known approach for identifying and discriminating the states and
transitions between SR, SVT, VT and VF is based upon measurements on the
cardiac electrogram and utilizes a frequency analysis technique which is
used in determining the rhythm from the power spectrum. This technique is
limited to the surface ECG by computational requirements. Still other
techniques attempted for identifying and discriminating between normal SR
and various tachycardias and fibrillation involve morphologic measures,
such as rise time or polarity of the ECG waveform and the so-called
"probability density function (PDF)" measurements in which the cardiac
waveform is characterized by the relative percentage of time that it
spends at various amplitude levels.
Each of these techniques has significant shortcomings. The outcome of the
evaluation involving the PDF, for example, is found to be quite dependent
upon electrode placement and the evaluation function requires an
inordinate amount of time. Implantable tachyarrhythmia control devices
have a variety of responses keyed to specific arrhythmias. The appropriate
response is selected and triggered by the arrhythmia detection circuit.
Inappropriate or missed detections carry significant penalties.
While maintaining accuracy, arrhythmia recognition, especially for VF, must
be prompt. Shortening the interval from VF onset to intervention results
generally in a higher proportion of successes due to halting the decay of
the cardiac substrate before that process becomes irreversible. While it
is not yet clear that this benefit extends to the first 30 seconds
following onset, it is still vital that recognition, which controls the
entire process leading to intervention, be as accurate and timely as
possible. In the presence of SR and SVT, however, intervention is
inappropriate. False identification of SR or SVT as VT or VF can lead to
potentially dangerous intervention, again confirming the need for
accuracy.
In contrast to the prior art, the present invention relies upon a spacial
coherence detection approach to tachyarrhythmia discrimination. It was
observed that the morphologic and probability density function techniques
referred to above depended on a determination of the temporal departure
from synchronization on a beat-to-beat basis. This, however, results from
and is secondary to spacial decorrelation over the entire cardiac domain.
It was felt that the temporal detector, based upon signals picked up by a
single lead, would take an inordinate amount of time from the onset of VF
or VT due to the degree of spacial desynchronization that must occur in
order to appear locally on such a single lead. In the case of the present
invention, a measurement is taken of the spacial decorrelation directly.
Instead of using a single sensing electrode, one or more leads having
plural electrodes are implanted within the heart but at somewhat remote
locations with respect to one another, assuring that substantially
distinct segments of cardiac tissue dominate the electrical field
surrounding each electrode. As a result, the dominant electrical activity
influencing one lead tends to be somewhat independent from the electrical
activity influencing the other lead. During SR, each electrode receives
some contribution from every electrically active cardiac cell. While the
resulting QRS complex may appear quite different on the two leads, these
signals are substantially coherent in the sense that there is a linear
relationship between them. That is, QRS wave on one lead may be derived
from that on the other via a linear transfer function. Assuming that the
leads are mature and stable in the sense that they do not move, the
transfer function also will be stable and will consistently reproduce the
signal on one lead from that on the other.
During VT, the QRS morphology on both leads will, in general, appear
different than during SR or SVT. This difference reflects the change in
the depolarization sequence, as the depolarization wave travels through
the muscle tissue, rather than the Purkinje fibers. In stable, monomorphic
VT, the signals on the two leads will again be substantially coherent, but
with different linear relationship than that which describes SR. Stated
otherwise, a different transfer function is required to convert the signal
on one lead to that of the other.
During VF, the depolarization sequence is constantly modifying from
beat-to-beat, with dispersal of the depolarization sequence occurring to
the point where cardiac activity can no longer be characterized as a
sequence of beats, but rather as continuous, fragmented electrical
activity.
The concept of coherence and the implied existence of linear transfer
function depend on the existence of a persistent linear relationship.
Taking the ratio of the complex spectra of the electrode signals over a
specific time interval produces a result which resembles a transfer
function, but does not carry with it the weight of persistence. Thus, such
a function could be derived for every interval during the course of
fibrillation, but the result would be inconsistent and incoherent. A
stable transfer function cannot be said to exist under these
circumstances. Thus, during VF, not only the signal morphology is
changing, but the channel characteristics between leads, as reflected in
the transfer function, is likewise changing.
SUMMARY OF THE INVENTION
In accordance with the present invention, a method and apparatus is
provided for detecting and discriminating between sinus rhythm,
ventricular tachycardia, supra-ventricular tachycardia and ventricular
fibrillation. The technique involves applying the least-mean-squares (LMS)
algorithm first described by B. Widrow and M.E. Hoff in "Adaptive
Switching Circuits", IRE WESCON Convention Record, Part 4, pp. 96-104,
Sept. 1960, and elaborated subsequently by Professor Bernard Widrow in a
book captioned "Adaptive Signal Processing", Prentice-Hall Publishing
Company, 1985, which estimates the transfer function or channel
characteristics between a signal (or noise) source and the perceived
signal plus noise. This signal processing technique is used to implement
coherence analysis between two sources of cardiac electrical activity, one
of which is sensed on a local basis using an endocardial bi-polar lead and
one sensing more global activity such as may be obtained between two
spaced surface ring electrodes on an endocardial lead. During sinus
rhythm, the vector difference is essentially constant and is estimated as
a transfer function by the LMS algorithm. This transfer function is then
used to estimate the global electrogram from the local. The estimate is
then substracted from the actual, leaving a null output. In the presence
of VT, the vector difference changes and results in non-zero output from
the null, until the channel characteristics are re-estimated. In VF, the
vector difference is constantly changing and the channel characteristics
can no longer be estimated. Therefore, during VF, the "null" output is
constantly changing and does not fall to zero. Thus, by monitoring the
"error" signal emanating from the LMS adaptive filter, one can readily
detect a shift from SR to VF. Similarly, in tachyarrhythmia
discrimination, several LMS adaptive filters may be coupled to the dual
leads and by appropriately causing each to adapt and, therefore, null a
different rhythm and by noting the rate range, VT can be distinguished
from SR and VF.
OBJECTS
It is accordingly a principal object of the present invention to provide an
improved system and method for detecting and identifying diverse cardiac
rhythms.
Another object of the invention is to provide a system and apparatus for
sensing ventricular fibrillation and for automatically activating
appropriate intervention.
Yet another object of the invention is to provide an apparatus and method
for distinguishing between sinus rhythm, ventricular tachycardia and
ventricular fibrillation in a more reliable fashion than has been possible
using prior art techniques.
A yet further object of the invention is to provide a system in which dual
leads or plural electrodes on a single lead are appropriately located
within a ventricular chamber of the heart and coupled to an LMS adaptive
filter for detecting a particular cardiac rhythm by noting the magnitude
of the error output of that filter.
DESCRIPTION OF THE DRAWINGS
These and other objects and advantages of the invention will become
apparent from the following detailed description of the preferred
embodiment, especially when considered in conjunction with the
accompanying drawings in which:
FIG. 1 is a schematic of electrical activity of a heart in sinus rhythm;
FIG. 2 is a schematic illustration of a heart in ventricular tachycardia;
FIG. 3 is a schematic illustration of a heart in a state of ventricular
fibrillation;
FIG. 4 depicts the manner in which signals derived from a multi-electrode
lead located near the right ventricular apex in the right ventricle are
applied to a LMS filter along with waveforms showing the input and output
relationship;
FIG. 5 is a block diagram representation of the LMS adaptive filter;
FIG.6 is a block diagram of the update algorithm of the LMS adaptive filter
of FIG. 5;
FIG. 7 is a block diagram of a system employing more than one adaptive LMS
filter and pulse rate detecting means for identifying/classifying the type
of tachyarrhythmia which may be in progress; and
FIGS. 8(a) and 8(b) comprise a flow chart of the software executed by a
general purpose digital computer for carrying out the method of the
present invention.
DESCRIPTION OF THE PREFERRED EMBODIMENT
Referring to FIGS. 1 through 3, there is shown in schematic form a model of
how the heart behaves during sinus rhythm, tachycardias and during
ventricular fibrillation, respectively.
In FIG. 1, a ventricular section of the heart is identified by the numeral
10 in which or on which two distinct catheters 12 and 14 have been
implanted. Catheter 12 has an electrode E.sub.1 in contact with the heart
tissue while catheter 14 has a corresponding electrode E.sub.2.
Schematically representing the sinus node or the supra-ventricular source
is a master clock 16 which provides regularly occurring timing pulses
which are distributed to various points in the heart, via the cardiac
conduction system, represented by the delay paths 18, 20 and 22. The
dotted line arcs as at 24 represent the propagation of the cardiac cell
depolarization fields, finally impinging on the sensing electrodes. Each
of the electrodes E.sub.1 and E.sub.2 senses the composite depolarization
activity from all over the heart, but the effect of each specific cellular
depolarization field depends on the distance to each electrode and its
relative timing.
Because the position of the two electrodes is different and because the
orientation of the electrode relative to its indifferent electrode may
also be different, the resulting signal waveforms 26 and 28 picked up by
the electrodes generally will have differing morphologies during sinus
rhythm. The QRS signals picked up by the electrodes in FIG. 1 are a
composite projection of a consistent pattern of cellular electrical
activity, schematically represented as being triggered through a system of
fixed delays by each pulse of a common clock. While these signals may
appear quite different on the two leads, they are substantially coherent
in the sense that there is a persistent linear relationship between them.
The QRS on the lead 12 may, therefore, be deduced from that on the lead
14, and vice-versa, by a linear process or transfer function. Convolving
one signal with a filter derived from the transfer function will reproduce
the other signal.
With respect to FIG. 2, there is a similar depiction of the heart 10 during
an episode of ventricular tachycardia. In comparing FIGS. 1 and 2, the
chief distinction is that a different pattern of delays times the cellular
depolarizations, depicted by a shift in the master clock 16 position. (The
positions shown have no specific significance other than to indicate a
change in timing.) Thus, in stable monomorphic VT, all cells are driven by
a master clock through a different pattern of delays than those which
persist in SR. The QRS morphology on each of the leads 12 and 14, in
general, appears different from their observed counterparts during sinus
rhythm, which is a result of the change in the depolarization sequence.
However, the two signals are, again, substantially coherent, but with a
different linear relationship reflecting the altered cellular
depolarization order, or delay pattern. Therefore, a different transfer
function is required to convert the signal on one lead to that on the
other. The similarity between sinus rhythm and ventricular tachycardia is
that each cardiac cell depolarizes once, and only once, during a cardiac
cycle. However, the timing is no longer controlled by the Purkinje system,
but by the relative distance from the ectopic focus that is stimulating
the ventricular tachycardia. As with sinus rhythm, in ventricular
tachycardia, a consistent pattern of delays and attenuations prevail, but
a different timing pulse period go into the master clock 16 and a
different system of delays and attenuation exit the master clock. The
important thing to note is that with ventricular tachycardia a consistent
linear process exists between the two electrodes and a transfer function
relating the two signals to one another can be estimated.
FIG. 3 is intended to depict the gradual breakdown which occurs when
progressing into a ventricular fibrillation. The schematic illustration
depicts the heart as being controlled by a number of subclocks, each
controlling a small domain within the heart and these domains may, in
effect, run at different rates as illustrated by the different timing
pulse patterns 30, 32, 34 shown as entering the subclocks 36, 38 and 40.
It is found that these rates may also change in an inconsistent and
continuous fashion. Because of relative tissue proximity, each lead is
dominated by electrical activity under control of a different clock. As
such, the net result of electrodes E.sub.1 and E.sub.2 in FIG. 3 is that
they sense cardiac activity over the entire heart, but the spacial
decorrelation manifests itself as a temporal decorrelation on leads 12 and
14. A consistent transfer function no longer exists between the leads due
to the face that the system of subclocks are clocking rates as well as
attenuations and delays is changing in a continuous and inconsistent
fashion.
Thus it can be seen that estimating the transfer function offers a means
for discriminating between tachyarrhythmias. The transfer function, once
determined, generates an accurate replica (estimate) of the signal on one
lead from that on the other, which can be subtracted from the reference to
form a null for stable rhythm. A filter tuned to produce a null to sinus
rhythm will also null SVT, since the depolarization sequence has little
dependence on rate in a healthy heart. Monomorphic VT is also sufficiently
stable to produce a null by this means. However, a filter tuned to null
sinus rhythm or SVT will not, in general, null VT. The converse of the
latter statement is also true. Since the relationship is constantly
changing, it is not possible to null VF consistently. Failure to produce a
stable transfer function is a hallmark of VF. Moreover, filters tuned to
null SR or VT will both produce an error output in the presence of VF.
It has been stated that the complex ratio of Fourier transforms of the
signals on the two leads produces a result which resembles a transfer
function, but does not carry with it the idea of persistence. There are,
however, techniques for estimating a transfer function in which such a
ratio is calculated for each of many adjacent intervals. The separate
coefficients are then allowed to average, or adapt, to their final result
by averaging over all the intervals. While implementable on a general
purpose computer, this is a formidable task for an implantable device. The
LMS estimation technique was chosen because it is simple computationally
and, therefore, more suited to an implantable device. While it has
shortcomings relative to the more general technique, these shortcomings do
not prevail in this application.
In accordance with the present invention, ventricular activity is sensed on
two separate endocardial leads, e.g., a right ventricular apex bipolar
lead and a right ventricular multi-ring catheter. The signals are applied
to a variable transfer function, LMS filter of the type described by
Bernard Widrow et al in a paper entitled "Adaptive Noise Cancelling:
Principles and Applications", Proc. IEEE, Vol. 63, Vec. 1975. More
particularly, and with reference to FIG. 4, numeral 42 again refers to a
schematic representation of the heart with a multi-electrode lead 44
positioned in the right ventricle chamber and having a tip electrode
E.sub.1 at the right ventricular apex and ring electrodes E.sub.2 and
E.sub.3 positioned proximally to effect primarily local near field
activity between E.sub.1 and E.sub.2 and primarily global far field
activity between E.sub.1 and E.sub.3 spaced-apart locations. Conductive
wires running through the body of the lead 44 connect electrodes E.sub.1
and E.sub.2 to a first differential amplifier 48 while conductors couple
electrodes E.sub.1 and E.sub.3 to a differential amplifier 50. Adopting
the designations utilized by Professor Widrow in the above-cited
references, the output from the differential amplifier 48 comprises the
input signal vector =[x.sub.0k, x.sub.1k ... x.sub.(n-1)k ], to the LMS
adaptive filter 52, wherein k is the time index. The signal emanating from
differential amplifier 50 is defined as the desired signal d.sub.k and is
applied to a summing node 54 along with the output from the LMS adaptive
filter 52. The output from the adaptive filter 52 is the "estimate" signal
Y.sub.k. The output from the summation node 54 then comprises the error
signal E.sub.k that is the difference between the desired or reference
signal d.sub.k and the estimated siganl Y.sub.k. Also depicted in the
block diagram of FIG. 4 is a correlator 56 which receives as its inputs
the desired signal d.sub.k and the estimate signal Y.sub.k and yields the
output (1-R), where R is the correlation coefficient. When the leads are
detecting sinus rhythm and the filter has locked onto that, the
correlation coefficient R is high, and the output from the correlation
channel, (1-R), is low (zero). During VF, however, the estimate Y.sub.k
poorly approximates d.sub.k. Therefore, R is low and the (1-R) output will
go high.
To better understand the fibrillation detection system of FIG. 4, that
figure also includes signal traces corresponding to data taken from a
human patient undergoing cardiac electrophysiology laboratory procedures.
Because of the particular lead arrangement, the two channels have
substantially parallel electrical vectors with common local components
from electrode E.sub.1 and with the desired channel incorporating
differential amplifier 50 receiving a stronger global contribution. An
examination of the waveforms of the outputs from the differential
amplifiers 48 and 50 reveal that the two channels exhibit substantial
morphologic similarity particularly during sinus rhythm. The low residual
error signal E.sub.k is indicative of sinus rhythm and indicates a stable
transfer function.
The LMS adaptive filter algorithm can be implemented using analog to
digital converters and a programmed microprocessor for computing the
various parameters. Using state-of-the-art technology, it is thought that
the technique can be adapted to implantable devices.
Having described in general terms with the aid of FIG. 4 the adaptation of
the LMS adaptive filter to the discrimination between SR, VT and VF, it is
believed helpful to a complete understanding of the present invention to
set out in greater detail the implementation of the LMS adaptive filter.
In this regard, reference will be made to the block diagrams of FIGS. 5
and 6. FIG. 5 is an aid to understanding the LMS algorithm when
functioning in its filtering mode while FIG. 6 pertains to operation in
the up-date mode.
When functioning in the filtering mode (FIG. 5) it behaves as a regular
finite impulse response (FIR) linear filter and the vector equation for
that can be represented as:
E.sub.k =d.sub.k -x.sub.k.sup.T W.sub.k
In the above expression, k is a time index, E.sub.k is the error signal at
time k, d.sub.k is the instantaneous value of the desired signal at time
k, X.sub.k is the input signal vector which is convolved with a weight
vector W.sub.k and which indicates that the weights adapt over time to
form an output signal Y.sub.k which is an estimate of the value of d.sub.k
at the time k. Following the calculation of the vector quantity d.sub.k
and W.sub.k, the weight vector is updated from W.sub.k to W.sub.k+1 by the
up-date algorithm represented as "U" in FIG. 6. The up-date algorithm can
be represented by the equation:
W.sub.k+1 =W.sub.k +2.mu.E.sub.k X.sub.k
where .mu. is the adaptation time constant. By way of further explanation,
if there are, for example, 32 weights which comprise W.sub.k, they may be
identified as w.sub.1k, w.sub.2k . . . w.sub.nk on out to 32, implying
that these are the values of the 32 weights at time k. Similarly, the
input vector signal X.sub.k, which is the input signal at time k, will
also have 32 values, i.e., the values at time k, the immediate moment,
plus the value at time k-1, k-2, etc., going back 32 sample spaces. The
summation of those products is the filter output, Y.sub.k, which is
subtracted from the immediate signal value d.sub.k to form the immediate
error response E.sub.k. This error response is both the signal output from
the adaptive filter system and the feedback to the updating equation for
the weight vector (FIG. 6).
With reference to FIG. 5, the boxes labeled z.sup.-1 represent unit delays
between the elements of the input signal while the weights w.sub.0k,
w.sub.1k are represented by the circles immediately below the unit delays.
The individual products are summed at 58 to form the output signal
Y.sub.k. This signal is, in turn, summed with the desired signal d.sub.k
at 60 to form the error signal E.sub.k at its output. The error signal is
fed back via path 62 for updating the weight vector.
Referring now to FIG. 6, there is shown the means by which the weight
elements of the filter of FIG. 5 are updated in response to the value of
the error signal. Understanding that a new error output occurs each time a
new element of the input signal vector is sampled, the weight up-date
operation also may occur as often as each new sample arrives, but it takes
place between the calculations for the filter output. Thus, a new filter
output is first calculated to form the error signal and tht is followed by
using that error signal to calculate a new set of increments to add to the
weight vector to up-date the weight elements of the filter with the
process continually repeating until a point of maximal adaptation has been
reached. The updated weight elements w.sub.0k+1, w.sub.1k+1, etc.,
represent the summation of the prior weight values w.sub.0k, w.sub.1k with
the product of the current value of the error signal, E.sub.k and the
input signal vector samples. The convergence value, .mu., is a scaler
quantity in the range of from 10.sup.-4 to 10.sup.-12 depending upon the
specific application for the adaptive filter. The value of .mu. affects
the rate at which the system will converge.
Having explained in detail the computational algorithm for implementing the
LMS adaptive filter, consideration will next be given to the manner in
which the LMS algorithm can be utilized with other conventional techniques
in tachyarrhythmia discrimination. In this regard, reference is made to
FIG. 7. It shows the outputs of several LMS filters and rate classifiers
as inputs to a "truth table" which determines the rhythm. the output of an
LMS filter is shown as "0" when the rhythm for which it is tuned is
present, or, for continuous adaptation, when the rhythm is sufficiently
stable to permit convergence. As an example, a "1001" pattern shows "1"
for sinus rhythm filter (rhythm is not SR), a "0" for the primary VT
filter (specific to primary VT), a "0" for the adaptive filter (rhythm is
stable enough to permit convergence) and a "1", indicating rate is in the
tachycardia range. Taken together, these responses indicate primary
monomorphic VT. Similarly, if LMS.sub.1 produces a null output while
LMS.sub.2 does not, and the rhythm is stable such that the LMS.sub.3
filter converges, and the rate is high, supraventricular tachycardia is
indicated. The other truth table entries shown in FIG. 7 define other
modalities.
FIGS. 8(a) and 8(b) together comprise a flow chart of the program
executable in a general purpose digital computer to implement the adaptive
filtering and control whereby fibrillation and tachyarrhythmias
discrimination can be achieved. For ease of explanation, the software is
partitioned into an inner loop and an outer loop, the inner loop being
shown in FIG. 8(a) and the outer loop in FIG. 8(b). The operations
performed in the inner loop are initiated by the generation of an
interrupt which is occasioned by the outputs from a sample clock or timer
in the microprocessor hardware. This sample clock oversees the
digitization process by the A-to-D converter, which may typically comprise
a sample-and-hold circuit used to periodically capture the magnitude of
these analog signals on the endocardial leads positioned within the right
ventricle. When the sample-and-hold circuit is triggered to capture a new
value, an interrupt is generated to the microprocessor causing the main
control program to shift to the inner loop which then executes as an
interrupt driven subroutine. Block 60 represents a vectored interrupt
stimulated by the sampling clock. In accordance with standard interrupt
procedure, as indicated by block 62, when the interrupt is true, the
values contained in various registers in the microprocessor are stored so
that when the interrupt subroutine is completed, control can return to the
point in the program where it left off at the time that the vectored
interrupt occurred. For example, the contents of the program counter and
other operational registers are stored temporarily. Once this operation
has been performed, the operations, identified in block 64 are executed.
Specifically, a new sample of the input signal following its conversion
from analog-to-digital form is brought into the microprocessor and stored
away in a memory so as to become accessible to the arithmetic/logic
portion of the microprocessor.
In that in the present application, many samples of input data are
simultaneously stored, it has been found convenient to utilize a so-called
circular buffer, again, in accord with standard software techniques. It is
to be recognized that the buffer needs to hold as many samples of the
digitized input data as there are weights on the adaptive filter and these
weights must periodically be updated. Using a circular buffer, the address
pointer is used to address the part of the circular buffer where the next
sample is to be stored or where the next sample is to be fetched for
performing an arithmetic operation.
As represented by block 66, the next operation in the inner loop is to
calculate the lastest filter output, Y.sub.est where Y.sub.est, at time t,
is equal to the summation over the range of i=1 to k of the products of
the current weights w.sub.i (t) and x.sub.i (t), the current data input
value. When it is considered that the index, i, can be the address in the
circular buffer referred to by the pointer, it can be seen that if the
weight values are stored at sequential locations in the buffer, they may
be read out in synchrony with the input samples contained in a like buffer
similarly addressed but with an address offset and presented to the
arithmetic logic unit where the sum of the products operation set out in
box 66 of FIG. 8(a) is carried out. The computed result, y.sub.est (t) is
the "estimate" output from the LMS filter 52 (FIG. 4). Following the
calculation of the latest filter output as represented by block 66 in the
flow diagram, a test is made at 68 to determine whether the weights are to
be updated. In performing this decision, a test is made of a flag which is
either in a set or a cleared condition depending upon the control
exercised by the external program. For example, if the flag always remains
set, the weights will be updated each time the filtering operation is
executed. If the signals X.sub.k and d.sub.k are sufficiently stable,
computation time may be saved by setting the "update weights" flag only
periodically, say every tenth sample or, alternatively, it may be set in
response to some external condition, such as a rapid transition in heart
rate from SR to VT.
Assuming that the flag is set and that the weights are to be updated, the
operation, for which the Fortran "Do" statement in box 70 is illustrative,
is carried out by the microprocessor. Following that, the storage location
pointers for the input circular buffers are updated and the sample flag
for the outer loop is set (block 72). Finally, as represented by block 74,
the program exits the interrupt routine once the program counter and other
registers whose contents had been stored in the operation of the block 62
have been restored their appropriate registers. This completes the inner
loop subroutine.
The outer loop depicted in FIG. 8(b) depicts the software operations which
are primarily "housekeeping" in nature. At the time that the system of the
present invention is put into operation, a step referred to as "full
initialization" (block 76) is carried out. It is at this time that certain
parameters are initially loaded into memory or holding registers
associated with the microprocessor. For example, such things as adaptation
time, bandwidth and the number of weights in the filter are established.
Following that, once the system is brought into operation, the "sample
flag" is evaluated to determine whether a new sample is available from the
A-to-D converter. This test is represented in the flow diagram by block
78. It will be recalled that it was during the operations represented by
block 72 in the inner loop subroutine of FIG. 8(a) that the "sample flag"
for the outer loop was set. If no new test sample is available, the outer
loop goes into a wait mode until such time as a new sample is available.
At that time, the operation represented by block 80 in FIG. 8(b ) is
carried out. Specifically, a test is made to determine whether the
absolute value of the error plus the absolute value of the estimate is
much greater than the absolute value of the desired signals (FIG. 4). If
so, it is an indication that the filter is no longer stable indicative of
a problem requiring reinitialization. When it has been determined that the
system has lost stability, further interrupts to the inner loop are locked
out and the system is reinitialized (blocks 82 and 84). In partial
initialization, typically the tap-weights for the filter will be reloaded
from memory in the pattern indicating convergence to sinus rhythm.
Where it is determined that the filter is stable, the sequence progresses
to block 86 which is the operation of "automatic gain control". Since in
accordance with the LMS adaptive filter algorithm, the "error" is equal to
"desired" minus "estimate", it necessarily follows that the "error" plus
the "estimate" is equal to the "desired" signal. Now, by normalizing,
i.e., dividing through by the "desired" signal for every sample point, two
fractional values are obtained, one for the "error" and one for the
"estimate", and their sum is always equal to unity. As previously
described, where there is a perfect accommodation to the input signal,
substantially all of the energy will be in the "estimated" signal such
that when the "estimated" signal is subtracted from the "desired" signal,
the "error" signal has essentially zero energy. If then it is later
determined that there is substantial energy in the "error" signal relative
to the "estimate" signal, one of three situations are indicated. First,
the heart may be in a transition between rhythms. Secondly, it may have
happened that the filter has diverged and is no longer operating properly.
The third and most probable event is that ventricular fibrillation is
involved precluding the LMS adaptive filter from accommodating the
continuously changing vector characteristic of ventricular fibrillation.
The stabilized outputs (desired, estimate, error) are processed for pulse
detection, beat-to-beat interval and variation and interchannel pulse
coincidence. During stable rhythms, the "estimate" will coincide with the
"desired", and the "error" output will be below some predetermined
threshold. During transistions between stable rhythms, the error output
will be pulsatile and above that threshold. During unstable rhythms, e.g.,
VF, the "error" output will be above threshold and there will be
asynchrony between the "desired", the "estimate" and the "error" channels.
The operation indicated by block 88 indicates the necessity of keeping
track of the duration of the period in which the "error" signal approaches
or exceeds the threshold and, in this regard, a clock register in the
microprocessor is utilized. A counter is also used to keep track of how
many pulse may occur during a given length of time. The thresholds, counts
and clocks thus need to be periodically updated when it is considered that
the system of the present invention is utilized for detecting shifts in
rhythms over extended time periods.
Finally, once a determination as to the nature of the rhythm involved has
been determined, the microprocessor sets one or more "state" flag (block
90) so that this information becomes available to an external device to be
controlled, such as, a defibrillator or a pacemaker.
This invention has been described herein in considerable detail in order to
comply with the Patent Statutes and to provide those skilled in the art
with the information needed to apply the novel principles and to construct
and use such specialized components as are required. However, it is to be
understood that the invention can be carried out by specifically different
equipment and devices, and that various modification, both as to equipment
details and operating procedures, can be accomplished without departing
from the scope of the invention itself. In particular, those skilled in
the art will realize that the invention can not only be practiced using a
programmed digital computer, but also can be implemented using special
purpose digital logic devices (hardware).
* * * * *
|
|
 |
|
 |
Description |
|
