Information Technology Reference
In-Depth Information
The detection focusses on identifying ventricular and auricular activities on the ECG.
Many methods have been proposed for detecting the ventricular activity, i.e. QRSs.
Each one fails in specific situations and each reacts differently to the many different
QRS waveforms. This is why we retain several algorithms. However, the proposed ap-
proach is not to merge the decisions of several algorithms but to select, on line, the most
promising one according to its performance in similar to the current situation. Actually,
seven algorithms were selected [26].
In this setting, two main problems must be solved:
-
devise the activation rules associated to a detector. Decision tree were used to auto-
matically extract these rules. Three main steps were executed i) generate all possi-
ble contexts that can be found in an ECG, ii) execute all the QRS detectors in these
contexts and iii) select the best detector for a given context.
-
define an efficient technique to estimate the current context. Our proposal is based
on the observation of several real noisy contexts in order to estimate the covariance
matrix and then to apply classical decision rules based upon the covariance matrix
and the current observation vector.
Once a QRS has been detected it is classified and a symbolic event is generated. The
generation of its attribute values is based on the fact a beat can be efficiently represented
by a compactly supported wavelet base [30]. Thus, each QRS is represented by a global
extremum at each decomposed level. The QRS classification consists in labeling the
beats into two mains classes, normal or abnormal. A probabilistic neural network (PNN)
based on radial basis function has been used. A pre-processing step including a principal
component analysis was performed in order to reduce the complexity.
P wave detection is very hard because it has a weak amplitude and a variable mor-
phology. P wave detection from the surface ECG signal is generally achieved according
to two methodological approaches, window-based search or QRS-T interval cancella-
tion. We retain the QRS-T interval cancellation technique to overcome the limitations
of window-based techniques which assumes that a P wave always occurs before some
QRS. In particular, such methods have difficulties to detect the P wave in case of ar-
rhythmias with AV dissociation which happen when the AV node blocks the electrical
conduction. Our approach [31] mostly relies on: i) QRS-T interval detection and can-
cellation based on wavelet decomposition, ii) a statistical analysis of the residue for
detecting P waves not associated to a QRS, iii) an artificial neural network classifier to
reject false detection which frequently occur in P wave detection.
The temporal abstraction module ouputs a sequence of time-stamped events with a
shape attribute. For example, below is a sequence of events that could be related to the
ECG excerpt of Fig. 2:
1
event(p_wave,
5812, normal)
2
event(qrs_complex, 5924, normal)
3
event(qrs_complex, 6315, abnormal)
4
event(p_wave,
7432, normal)
5
event(qrs_complex, 7546, normal)
6
event(qrs_complex, 7882, abnormal)
7
...
Search WWH ::
Custom Search