Instrumentation & Measurement Magazine 23-4 - 89

episode even for the same patient and differ broadly among
people that suffer from this condition. Nonlinear methods are
successful approach for such cases of signals. Using the exemplary decision system based on MLP proposed in [5] gives the
accuracy of a range above 87% while another epileptic seizure
detection system based on ECoG signal analysis moves the accuracy level up to 98% [4].
Another example of brain activity is related to a person's
ability to remain focused for prolonged times. Monitoring,
measurement and computing human concentration are available thanks to BCI-based methods. Brain emotional conditions
can also be evaluated based on vision signals e.g., responses to
stimuli such as pictures or movies.
An example of vision research is similar to [7] where 3D
face model features were calculated and neural classifiers
were applied to automatic detection of emotions from the visual facial images with the accuracy ranging from 73% up to
96% (Fig. 4).
The human heart is also the organ of fundamental importance and awaits a wide spectrum of computer aided methods
for observation and physician decision support. Cardiovascular system mechanical functionality can be described, in a
strongly simplified form, by a set of tens of differential equations based on Navier-Stokes equations.
These equations remain with a strict connection to the electric equivalent simplified circuit presented in Fig. 5, which at
first glance reveals to be nonlinear. Electrical activity of a heart
can be observed during a simple and noninvasive examination
using ElectroCardioGraphy (ECG). In this area of research,
computer aided processing and diagnosis support systems
also involve nonlinear methods. One classic meaningful nonlinear operation is an elementary squaring following bandpass
extraction filtering stage. It is performed to emphasize the outcome to extend the S/N ratio of a final detection function.
Nonlinear operations can be applied to the primary ECG signal analysis stage that involves the QRS complex detection
[8], [9]. The ECG signal processing field where the nonlinear

Fig. 4. Visual human face characteristics for different emotions.
June 2020

Fig. 5. Nonlinear electrical model of cardiovascular system.

operations are also widely used is a problem of a arrhythmia classification. The classification problem is illustrated in
Fig. 6 where signals were taken from the MIT-BIH standard
Arrhythmia Database. Fig. 6a presents different arrhythmia
dysfunctions of the heart. They have direct impact on the most
important element of the ECG signal, the QRS complex. Generally, these different QRS shapes can constitute the input data
for neural classifiers. In practice, the data are often preprocessed in order to emphasize wanted properties or to narrow
down the amount of input data set.
An example of a preprocessed input data set is presented
in Fig. 6b. It can serve as the input applicable to the ANN.
The picture presents an example of scale Discrete Wavelet
Transform (DWT) coefficients for the normal and abnormal
QRS complex detection and classification process. The use
of a DWT tool brings additional gain in the concentration of
different arrhythmias properties. That makes it more suitable to be classified by a neural classifier (ANN or SVM) or
a Bayes classifier [10], [11]. There are concentrations of the
parameters connected to different states or classes that are
commonly processed by ANN. There is also a common approach based on a combination of several processing tools
(i.e., neural classifiers and Bayes based methods) into a single expert or ensemble system [3], [11]. It was empirically
proven that the combination gives promising results even
if a single classifier alone acts rather poorly. The confusion
matrix describing the performance of the classification system presented in [11] is characterized by the total accuracy
as high as 98.6%.

IEEE Instrumentation & Measurement Magazine 89

Instrumentation & Measurement Magazine 23-4

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 23-4