Page 139 - Control Theory in Biomedical Engineering

Genetic fuzzy logic based system for arrhythmia classification 123

Table 7 Performances of the FLC before genetic optimization.
S TP TN FP FN ACC (%) Se (%) Sp (%)
NSR 31 10 1 6 85.416 83.783 90.909
LBBB 3 43 1 1 95.833 75.000 97.729
RBBB 3 42 1 2 93.750 60.000 97.674
PVC 1 42 3 2 89.583 33.333 93.333
P 0 40 4 4 83.333 0.000 90.909
Average ACC 89.583

Table 8 Performances of the FLC after the genetic optimization.
S TP TN FP FN ACC (%) Se (%) Sp (%)
NSR 32 16 0 0 99.999 99.999 99.999
LBBB 4 31 0 1 97.058 66.666 99.999
RBBB 4 32 0 0 99.999 99.999 99.999
PVC 2 31 2 1 94.117 50.000 96.875
P 2 31 2 1 94.117 50.000 96.875
Average ACC 97.054

Tables 7 and 8 summarize the FLC performances achieved before and after
the genetic optimization, respectively. In each table, bold values indicate the
highest and lowest class performances, as well as the average ACC.
On one hand, by referring to Table 7, we deduce that the configured
FLC reaches an acceptable average accuracy (ACC¼89.583%). Globally,
it achieves good results for classifying four heartbeats types (NSR, PVC,
LBBB and RBBB). In addition, we found that the maximum accuracy
(ACC¼95.833%) is obtained by classifying (LBBB) heartbeats with signif-
icant sensitivity (Se¼75%) and specificity (Sp¼97.72%). However, the
performances are less efficient (ACC¼83.333%, Se¼0%, Sp¼90.90%)
by evaluating the (P) heartbeats. This indicates that the FLC has not classified
any examples with arrhythmia (P). We also noticed that only 33.33% of the
examples with arrhythmia (PVC) are correctly classified.
On the other hand, by referring to Table 8, we deduce that after applying
the genetic optimization the FLC reaches a more accurate average accuracy
(ACC¼97.054%). In fact, the optimized FLC achieves good results (ACC,
Se, Sp¼99.999%) for classifying (NSR and RBBB) heartbeats. However,
the obtained performances for the classification of the (P and PVC) arrhyth-
mias are improved by using the optimized parameters (ACC¼94.117,
Se¼50% and Sp¼96.875). However, they are still less accurate than the
other heartbeat classes (NSR, LBBB and RBBB).