Nanang Sulistiyanto, Raden Arief Setyawan, Made Wena Harilegawa
Direct feeding electrocardiogram (ECG) signal into a neural-based classification system may offer simplicity and low power consumption. To conserve most morphological information of a single electrocardiogram (ECG) wave, P-Q, QS, and S-T segments were windowed with different sampling periods, i.e.: 10,5, and 50 ms, respectively. Each sample was then converted to periodical spikes at a rate of 1 kHz} / mV} in a time slot of \mathbf{6 4 ~ m s}. ECG waves of modified limb lead II (MLII) from the MIT-BIH database were trained into various spiking neural networks (SNN) to classify Normal (N), Supraventricular (S), Ventricular (V), Fusion (F), and Unknown (Q) beats. As a result, SNN with two hidden layers comprising 30 neurons each attained accuracy of about \mathbf{9 4 \%} for moderate perturbations in baseline (\pm 0.1 mV) and time (\pm 48 ms). © 2025 IEEE.
(Universitas Brawijaya), DIICES Research Group, Malang, Indonesia
