Page 121 - Control Theory in Biomedical Engineering

Genetic fuzzy logic based system for arrhythmia classification 107

Radosav, 2018). The results indicate that convolutional neural networks are
the most widely represented (Parvaneh et al., 2019).
In addition, the FLC that mimics experts to generate a suitable control
action for a particular system has been largely integrated in DSS process. By
referring to the expert knowledge, the FLC is mainly used to overcome the
uncertainty and vagueness and to involve approximate reasoning for
decision-making tasks. In the medical sector, and more particularly for clas-
sifying ECG signals, numerous works based on the FLC approach have been
developed and evaluated (Krishnaiah et al., 2016). However, its configura-
tion is still a challenging task. Indeed, the FLC performances depend on the
adjustment of the membership functions parameters and its rule base. To
overcome this problem, a number of hybrid approaches were offered in
several research studies, such as fuzzy clustering techniques and fuzzy evo-
lutionary algorithms (Rathi and Narasimhan, 2017).
Thus, in this study, a fuzzy genetic-based system for cardiac arrhythmia
classification is investigated. It is applied to classify the MIT-BIH Arrhyth-
mia Database recordings into five arrhythmia types: (1) Normal Sinus
Rhythm (NSR), (2) Premature Ventricular Contraction (PVC), (3) Left
Bundle Branch Block (LBBB), (4) Right Bundle Branch Block (RBBB)
and (5) Paced beats (P) (Silva and Moody, 2014). Previously, we have used
the neural networks for this task (Lassoued and Ketata, 2018c). However, we
concluded that it is usually difficult to select the more appropriate neural net-
work structure and to interpret its results. Accordingly, the proposed based
system consists of a FLC whose membership parameters and rules number
are adjusted by a GA. This system has to define the correct arrhythmia type
for a previously unknown sample. Then, a comparison study between the
obtained accuracies before and after the genetic optimization will be ana-
lyzed. Simultaneously, in order to access the efficiency of this study, a second
comparison study between neural network approaches and the proposed
FLC will be done.
This chapter is organized as follows. Section 2 describes the proposed
arrhythmia classification methodology. Section 3 details the experiments
and results, and the final section provides a conclusion.

2 Methodology

In this section, we describe the suggested arrhythmia classification method-
ology, which consists of a hybrid system between a FLC and GA. As it is
described in Fig. 1, the FLC uses a pre-processing block to filter the original