[Back]

M. Bachler, C. Mayer, B. Hametner, S. Wassertheurer:
"Automatic Detection of the QRS Complex, P and T Wave in Electrocardiography in Real Time";
Poster: ASIM-TCSE Workshop 2012, TU Wien; 2012-02-13 - 2012-02-14; in: "ARGESIM Report", S. Tauböck, F. Breitenecker (ed.); Argesim / Asim, 37 (2012), 1 - 2.

Motivation.
Cardiovascular diseases are amongst the most common dis-
eases and the leading cause of death in developed countries. The earlier these
diseases are diagnosed, the better is the success rate of the treatment and the
prognosis. Dynamic changes in the duration of certain intervals (shown in
the figure to the right) in the electrocardiograph (ECG) are well established
indicators in the diagnosis of cardiac diseases. Automating the measurement
process of the ECG and combining it with its analysis yields numerous ad-
vantages over manual methods, therefore an embedded system for data acqui-
sition, processing and analysis was developed.
Development process.
The first phase of the development process con-
sisted of the creation of an algorithm using MATLAB
®
and its verification
against ECG signals manually annotated by medical experts using PhysioNet
databases [1]. During the second stage, the algorithm was ported to the em-
bedded System using Embedded MATLAB
®
, followed by a hardware-in-the-
loop simulation coupling the measurement hardware with the MATLAB
®
model in order to validate the developed
system.
The algorithm.
The algorithm presented in this work detects R peaks based on the signals amplitude and first
derivative [2]. False positive detections due to artefacts are prevented by analysing the signal´s local statistic
characteristics. These intermediate results are classified in real-time to distinguish normal heartbeats from potential premature ventricular contractions. Templates are built from each class of heartbeats to reduce noise. These are
used for the detection of the QRS complex, P and T wave. These characteristic features of the ECG are detected by
the first derivatives of their class-templates [3] and a geometric method, respectively. Finally, they are separately
refined for each detected heartbeat.
Results.
In the verification phase as well as in the hardware-in-the-loop simulation, the algorithm achieved a
sensitivity of 98.2% and a positive predictive value of 98.7%, respectively. Its accuracy, shown in the figure below,
matches those of human medical experts, so one can conclude that its performance is adequate for the calculation of certain relevant intervals of the ECG.
-60.00
-40.00
-20.00
0.00
20.00
40.00
60.00
80.00
Time difference in ms
Mea