VOL. 10, NO. 20, NOVEMBER 2015 ISSN 1819-6608 ARPN Journal of Engineering and Applied Sciences © 2006-2015 Asian Research Publishing Network (ARPN). All rights reserved. www.arpnjournals.com 9839 ARRHYTHMIA DETECTION BASED ON HERMITE POLYNOMIAL EXPANSION AND MULTILAYER PERCEPTRON ON SYSTEM-ON-CHIP IMPLEMENTATION Amin Hashim 1 , Rabia Bakhteri 2 and Yuan Wen Hau 3 1,2 Faculty of Electrical Engineering, Universiti Teknologi Malaysia, Skudai, Johor, Malaysia 3 IJN-UTM Cardiovascular Engineering Centre, Faculty of Biosciences and Medical Engineering, Universiti Teknologi Malaysia, Skudai, Johor, Malaysia E-Mail:cik.amin@gmail.com ABSTRACT As the number of health issues caused by heart problems is on the rise worldwide, the need for an efficient and portable device for detecting heart arrhythmia is needed. This work proposes a Premature Ventricular Contraction detection system, which is one of the most common arrhythmia, based on Hermite Polynomial Expansion and Artificial Neural Network Algorithm. The algorithm is implemented as a System-On-Chip on Altera DE2-115 FPGA board to form a portable, lightweight and cost effective biomedical embedded system to serve for arrhythmia screening and monitoring purposes. The complete Premature Ventricular Contraction classification computation includes pre-processing, segmentation, morphological information extraction based on Hermite Polynomial Expansion and classification based on artificial Neural Network algorithm. The MIT-BIH Database containing 48 patients’ ECG records was used for training and testing purposes and Multilayer Perceptron training is performed using back propagation algorithm. Results show that the algorithm can detect the PVC arrhythmia for 48 different patients with 92.1% accuracy. Keywords: artificial neural network, electrocardiography, FPGA, hermite polynomial expansion, multilayer perceptron. INTRODUCTION The heart is the most important organ in the human body and it beats non-stop throughout the human lifespan. Its main function is to supply oxygen and nutrients throughout the body. Cardiovascular diseases (CVD) are heart related diseases and they are the leading cause of death worldwide. According to World Health Organization estimate, in the year 2012, 17.5 million people died from CVD, representing 31% of all global deaths (Dinc, 2013). Electrocardiography is the process of monitoring the electrical activity that is primarily used to diagnose heart disease. A study done in (De Backquer, 1998) shows that the CVDs that caused mortality have high correlation with arrhythmia found in the patient’s electrocardiogram (ECG) record. Therefore, a highly accurate screening device is important to detect arrhythmia for early treatment, thus reducing the risk of death due to CVD. Premature Ventricular (PVC) heartbeat is the most common arrhythmia. Framingham Heart Study reported that risk of mortality, cardiac death and myocardial infarction is double when associated with PVC (Ng, 2006), (Bikkina, 1992). PVC can be recognized by its irregular rhythm, absence of P wave and wide QRS complex shape. If PVC happens more than three times, it is named as Ventricular Tachycardia that later might evolve to Ventricular Fibrillation that is a fatal rhythm (Reid, 1924). Thus it is important to monitor and detect PVC at early stage. The electrocardiogram is a widely accepted tool used to detect abnormal heart rhythms and as the first indicator to investigate cardiac disorder, such as the cause of chest pain. It records the electrical activity of heart by attaching electrodes to the patient limbs. However, the ECG signals are irregular in nature and occur randomly at different time intervals during the day. Thus the need of continuous monitoring of the ECG signal is desired, which by nature is complex to comprehend. Moreover, there is a possibility of the analyst missing vital information through manual analysis which can be crucial in determining the nature of the disease. To reduce the medical practitioner’s workload, a highly accurate algorithm for computerized arrhythmia detection proposed by researchers. To design a portable, lightweight heart monitoring device that can detect arrhythmia accurately, the most suitable implementation is as a System-On-Chip (SOC). The prototype of SOC can be designed on Field Programmable Gate Array (FPGA) where the more compute intensive computation modules can be designed in hardware while other process blocks are implemented on software. This would greatly reduce the computational clock cycle needed to perform the task, thus it might reduce the computational time tremendously compared to general purpose embedded processor (Nambiar, 2012). RELATED WORK There are many general computations involved in detecting the arrhythmia in ECG. The first is preprocessing stage to remove noise, normalize the signal and prepare the data so that it contains only useful information. The second stage is feature extraction, which is to extract the ECG feature so that it can be classified accordingly. Feature extraction algorithm such as Hermite Polynomial Expansion (Lagerholm, 2000), Wavelet transform (Pachauri, 2009), Peak Valley Detection (Al- Aloui, 1986), Statistical analysis based on heart rate variability (Mohammadzadeh, 2006) and many more can