.TH NGUESS 1 "11 June 2008" "WFDB 10.4.7" "WFDB Applications Guide"
.SH NAME
nguess \- guess the times of missing normal beats in an annotation file
.SH SYNOPSIS
\fBnguess -r\fR \fIrecord\fR \fB-a\fR \fIinput-annotator\fR [ \fIoptions\fR ... ]
.SH DESCRIPTION
.PP
This program copies its input (a WFDB annotation file containing beat
annotations), removing annotations of events other than sinus beats, and
interpolating additional Q (unknown beat) annotations at times when sinus beats
are expected. Intervals between sinus beats are predicted using a predictor
array as described by Paul Schluter ("The design and evaluation of a bedside
cardiac arrhythmia monitor"; Ph.D. thesis, MIT Dept. of Electrical
Engineering, 1981). When the predictions are inconsistent with the known sinus
beats, as may occur in extreme noise or in highly irregular rhythms such as
atrial fibrillation, no interpolations are made.
.PP
Options for \fBnguess\fR include:
.TP
\fB-f\fR \fItime\fR
Begin at the specified \fItime\fR. By default, \fBnguess\fR starts at the
beginning of the record.
.TP
\fB-h\fR
Print a usage summary.
.TP
\fB-m\fR \fIM\fR
Insert Q annotations in the output at the inferred locations of sinus beats
only when the input RR interval exceeds \fIM\fR times the predicted RR interval
(default: \fIM\fR = 1.75). \fIM\fR must be greater than 1; its useful range
is roughly 1.5 to 2.
.TP
\fB-o\fR \fIoutput-annotator\fR
Write output to the annotation file specified by \fIoutput-annotator\fR
(default: \fBnguess\fR).
.TP
\fB-t\fR \fItime\fR
Stop at the specified \fItime\fR.
.PP
It should be understood that, as the name of this program implies, the Q labels
it generates represent, at best, good guesses about the times at which sinus
beats may be expected. Ideally, one should avoid having to make such guesses,
but some commonly-used techniques for study of heart rate variability (for
example, conventional methods for power spectral density estimation in the
frequency domain) require a uniformly sampled instantaneous heart rate signal,
such as can be obtained using \fBtach\fR(1) to process the output of
\fBnguess\fR. Other techniques, such as the Lomb periodogram method
implemented by \fBlomb\fR(1), can obtain frequency spectra from time series
with missing and irregularly spaced values, such as can be produced from a beat
annotation file using \fBihr\fR(1) without the need to use \fBnguess\fR. Use
\fBnguess\fR only when necessary and do not expect it to perform miracles; as
a rule of thumb, if the number of guesses (Q annotations) exceeds one or two
percent of the number of known sinus beats (N annotations), be exceedingly wary
of the guesses and consider using techniques such as \fBlomb\fR(1) that do not
require the use of \fBnguess\fR. Also as a general rule, \fBnguess\fR works
best when it is guessing the locations of sinus beats obscured by noise, or
those of sinus beats that were inhibited by isolated premature ventricular
beats; the underlying hypothesis of a quasi-continuous sinus rhythm, the basis
not only of \fBnguess\fR but also of all other algorithms for reconstructing
NN interval time series, is most suspect in the context of supraventricular
ectopic beats (which may reset the SA node, thus interrupting the sinus rhythm)
and consecutive ventricular ectopic beats.
.PP
The predictor array method is based on the observation that most of the
short-term variability in normal sinus inter-beat (NN) intervals is due to
respiratory sinus arrhythmia (RSA, the quasi-periodic modulation of heart rate
by respiration, which is most notable in young, healthy subjects and decreases
with age). Since respiration rate is (in humans and smaller mammals)
substantially slower than heart rate, it is possible to estimate the length of
the respiratory cycle in terms of some number of NN intervals. If, for
example, heart rate is around 60 beats per minute and respiration rate is
around 10 breaths per minute, one might expect that 6 NN intervals would
correspond to one breath, and that the current interval might be particularly
well-approximated by the sixth previous interval. Since we don't know the
ratio between heart and respiration rate a priori, we can observe how well each
of the previous \fBPBLEN\fR (a constant defined in \fBnguess.c\fR, see below)
intervals predicts the current interval on average. Thus we have \fBPBLEN\fR
predictors for each interval, some of which may be much better on average than
others. At any time, we know which predictor is (locally) the best, and we can
use that predictor to make a best guess of the location of the next sinus beat.
In subjects with significant RSA, the best predictor may be determined by the
length of the respiratory cycle; in others, the previous beat may be a better
predictor. For our purposes, it really doesn't matter which predictor is best,
only that the mean error of the best predictor is small. If the next known
sinus beat is at least 1.75 times as distant as the prediction, and if the
predictions are reasonably good on average, 'nguess' asserts that a gap exists
and fills it in with a Q annotation (or more than one, if the gap is
sufficiently long).
.SH ENVIRONMENT
.PP
It may be necessary to set and export the shell variable \fBWFDB\fR (see
\fBsetwfdb\fR(1)).
.SH FILES
.TP 22
\fIrecord\fR.hea
header file
.TP 22
\fIrecord\fR.\fIinput-annotator\fR
input annotation file (may contain any annotations)
.TP 22
\fIrecord\fR.\fIoutput-annotator\fR
output annotation file (contains N and Q annotations only)
.SH SEE ALSO
\fBihr\fR(1), \fBlomb\fR(1), \fBsetwfdb\fR(1), \fBtach\fR(1)
.SH AUTHOR
George B. Moody (george@mit.edu)
.SH SOURCE
http://www.physionet.org/physiotools/wfdb/app/nguess.c