ECG Signal Analysis - ECG Acquisition and Automated Remote Processing - page 22

TP
TP þ FN

Sensitivity ð S e Þ¼

ð 2 : 5 Þ

TP
TP þ FP

Positive Predictivity ð P þÞ ¼

where the first letter describes the correctness of the detection algorithm (i.e.,
right or wrong) in terms of TRUE or FALSE. The second letter indicates the
output of the algorithm (i.e., found or not found) in terms of POSITIVE or
NEGATIVE. For example, TP (true positive) stands for correctly detected peaks,
FN (false negative) for missed peaks and FP (false positive) for misdetections.
Sometimes, a third parameter named, specificity (S p ), is used, defined as

TN
TN þ FP

Specificity ð S p Þ¼ ð 2 : 6 Þ

where TN (true negative) stands for a correctly detected non-QRS region. The four
conditions can be described by the chart shown in Table 2.1 .
The QRS detection algorithms can be classified into the following classes:

(a) Principles based on digital filters and derivatives;
(b) Template matching techniques;
(c) Wavelet and other non-linear transform-based approaches;
(d) Neural network and GA techniques.

A typical software-based QRS detector can be represented by Fig. 2.3 . The first
stage, called 'preprocessing stage,' eliminates the noises and enhances the QRS
features. The following stage applies the peak detection logic based on some rules
and characterizes it.
A comprehensive review and comparison of QRS detection algorithms are
provided in [ 39 ]. The early approaches for QRS detection (up to 1985) used
discrete-derivative-based methods [ 40 ] based on digital filters, implemented

Table 2.1 Interpretation of sensitivity parameters

True False

Positive Wave peak (R) is correctly detected at

its position

Wave peak is detected at other positions
(misdetected)

Negative Wave peak is not found in non-QRS

region

Wave peak is missed at correct position

Pre-processing
(Denoising, QRS
enhancement)

Detection logic
based on rules and
QRS characterization

R-peak indexes /
locations

ECG samples x(n)

Fig. 2.3 Generalized software-based QRS detection representation