Information Technology Reference
In-Depth Information
The other main signal used for cardiac monitoring is the arterial blood pressure
(ABP). It depends on the contraction of the cardiac muscle (and therefore indirectly,
also on the propagation of the electrical wave). The heart pumps the blood through the
vessels such as the arteries. ABP measures the blood pressure in the largest arteries.
The most informative points in an ABP curve are the instants when the pressure is the
lowest (the diastole) and when the pressure is the highest (the systole).
3.2
Extracting Observation Information from the ECG
The main task of cardiac monitoring is to process the observed signal data and diag-
nose pathological states. A monitoring system has generally several stages [24]: signal
processing, diagnosis and alarm handling, therapy advising. To tackle these three tasks,
intelligent supervision systems appeared in the 90's. Their aim was to integrate sev-
eral sources of observation (numeric or symbolic) and several types of medical knowl-
edge e.g. surface and deep knowledge. Surface knowledge is related to so-called expert
knowledge while deep knowledge corresponds to a complete theory about a particular
subject from which all valid statements can be derived.
ECG signals are generally more or less noisy. This is a particular challenge for signal
processing, particularly to detect and analize P waves. It turns out that a global approach
to data abstraction, i.e. taking into account the particular signal features as well as the
context of their occurrence, is preferable to a standalone signal processing approach.
This is the way we have followed to design the Calicot cardiac monitoring prototype,
the architecture of which is described in the following section.
4
Architecture of the Cardiac Monitoring System Calicot
Calicot has two execution modes: offline and online. The online mode, depicted in
Fig. 3, is devoted to monitoring and adopts a pattern-matching approach: multivariate
signals are first abstracted in series of symbolic timestamped events and then a matcher
attempts to recognize, on the fly, instances of temporal patterns called chronicles in the
symbolic event series. A chronicle is associated to some cardiac disease and represent
a temporal signature of this disease. More details are given in Section 6 about chronicle
representation and chronicle recognition. The offline mode is dedicated to learning and
updating the decision rule base and the chronicle base.
Contextual information is of great importance for monitoring. On the one hand, by
taking the signal quality (noise) into account one can decide more accurately which
is the most relevant signal processing algorithm to use in the current situation. On the
other hand, by taking the current state of the patient, as well as his past states, into
account one can decide more accurately which are the most expected disorders and,
consequently, which are the most relevant chronicles. Moreover, these two decisions
are not independent: there are situations in which input signals are so noisy that it is
useless to try to detect P-waves, for example. Consequently, chronicles that contain P-
wave events cannot be recognized and should be removed from the set of candidates.
Also, in the context of a particular disease, some types of event could be absent from
the set of candidate chronicles and so it is useless to execute costly signal processing
Search WWH ::
Custom Search