18-02-2013, 09:44 AM
Principal Component Analysis as a Tool for Analyzing Beat-to-Beat Changes in ECG Features: Application to ECG-Derived Respiration
Principal Component.pdf (Size: 495.14 KB / Downloads: 66)
Abstract
Analgorithm for analyzing changes in ECGmorphology
based on principal component analysis (PCA) is presented
and applied to the derivation of surrogate respiratory signals from
single-lead ECGs. The respiratory-induced variability of ECG features,
P waves, QRS complexes, and T waves are described by the
PCA.We assessedwhichECGfeatures and which principal components
yielded the best surrogate for the respiratory signal. Twenty
subjects performed controlled breathing for 180 s at 4, 6, 8, 10,
12, and 14 breaths per minute and normal breathing. ECG and
breathing signals were recorded. Respiration was derived from the
ECG by three algorithms: the PCA-based algorithm and two established
algorithms, based on RR intervals and QRS amplitudes.
ECG-derived respiration was compared to the recorded breathing
signal by magnitude squared coherence and cross-correlation. The
top ranking algorithm for both coherence and correlation was the
PCA algorithm applied to QRS complexes. Coherence and correlation
were significantly larger for this algorithm than the RR
algorithm (p < 0.05 andp < 0.0001, respectively) but were not
significantly different from the amplitude algorithm. PCA provides
a novel algorithm for analysis of both respiratory and nonrespiratory
related beat-to-beat changes in different ECG features.
INTRODUCTION
WE PRESENT an algorithm for analyzing beat-to-beat
changes in features such as P waves, QRS complexes,
and T waves from single-lead ECGs. The algorithm has potential
use in identifying abnormal changes in these features, for
example, T wave alternans, but here we seek to quantify the
sensitivity of the algorithm to respiratory-induced variability of
ECG features. We aim to show that the algorithm is able to
track beat-to-beat changes in different ECG features and in doing
so provides a surrogate respiratory signal comparable with
other algorithms for ECG-derived respiration (EDR).
Respiratory-Induced Modulation of ECG
Respiratory-induced changes in the ECG arise due to several
mechanisms. First, the electrical impedance of the thorax
changes due to changes in lung volume [4]. Second, the heart
vector changes due to changes in the displacement and orientation
of the heart with respect to the ECG electrodes [5]. Third,
heart rate changes due to respiratory-induced changes to the
autonomic nervous system [6]. These factors give rise to morphological
variation in ECG features related to the breathing
cycle that we hope to capture in our algorithm.
ECG-Derived Respiration Algorithms
A number of algorithms for deriving respiration from the
ECG have been described in the literature. These algorithms
exploit the respiratory-induced changes of the ECG to provide
a surrogate respiratory signal. By this we mean a signal with
varying amplitude corresponding to the different phases of respiration
[1]. From this, the respiratory rate and temporal pattern
of breathing can be estimated. Many of these algorithms employ
a multilead approach, which provide robust estimates of
respiration, but can be implemented only where multilead ECG
recordings are available [1]. The simplest single-lead EDR algorithms
measure the beat-to-beat amplitude variation of the
QRS complex [7], [8] or T wave [9] that are associated with
the respiratory-induced variation in thoracic impedance [10].
However, such algorithms are susceptible to errors due to the
inherent uncertainty in measuring amplitude when the ECG
contains noise. A development to overcome the susceptibility
of the amplitude-based algorithms to noise was the EDR
algorithm based on the area under the QRS complex.
ECG-Derived Respiration
Three algorithms for deriving the respiratory signal from the
ECG were implemented: the PCA-based algorithm and two established
algorithms, one based on RR intervals and one based
on QRS amplitude. The established algorithms, alongside the
direct respiratory measurements, provided an objective assessment
of the accuracy of the respiratory signals obtained from the
PCA algorithm and facilitated comparisons of this algorithm’s
performance against the other algorithms.
1) Preprocessing of ECG: Each algorithm required the ECG
beats to be detected. An approximate time marker for each beat
was obtained as the maximum rate of change of the ECG by
identifying maxima in the differential of the ECG signal.
Comparisons of ECG-Derived Respiration With
Respiratory Signal
For each subject recording, we computed 14 EDR signals.
These comprised one from the RR algorithm, one from the QRS
amplitude algorithm, and 12 from the PCA algorithm, which
was applied separately to four ECG features (whole beat, P
wave, QRS complex, and T wave), and the eigenvectors from
the first three PCs were analyzed for each feature. Before
comparing the EDR with the respiratory signal, the beat-wise
samples of the EDR were resampled to the same sample rate
as for the respiratory signal (500 Hz) using linear interpolation.
Similarity in the time domain between EDR and the respiratory
signal was quantified using cross-correlation. The maximum
absolute correlation was determined for each recording; hence,
correlation was unaffected by differences in phase between the
ECG-derived signal and the respiratory signal.
DISCUSSION
PCA is a novel approach to EDR. We have demonstrated its
application using the different features of the ECG and obtained
significantly better results than the RR-based algorithm. The
PCA algorithm gave the best surrogate to the respiratory signal
when applied to QRS complexes, and good results were also obtained
when applied to other ECG features such as T waves. The
eigenvector of the first PC was sensitive to respiratory changes,
and in most recordings, these were associated with amplitude
changes. However, unlike other algorithms, the PCA algorithm
also extracts the subtle morphological changes to ECG features,
which were expressed in the second and third PCs. The
PCA-based algorithm also has the advantage that dominant PCs
are relatively noise free because ECG noise is uncorrelated to
the ECG features and is separated into the lower PCs. This is
demonstrated by the reconstructed beats shown in Fig. 6 where
the first two PCs were used to reconstruct the beats at the different
phases of the respiratory cycle.