Database Preprocessing H. Boon

TELKOMNIKA ISSN: 1693-6930  Ventricular Tachyarrhythmia Onset Prediction Based on HRV and Genetic… K. H. Boon 1001 stage respectively. Pre-processing stage includes ectopic beat correction, resampling of RR interval and conversion to HRV sequences by computing the reciprocal of RR interval. In HRV feature extraction stage, 5 time-domain, 6 frequency-domain, 5 non-linear and 37 bispectrum features are then extracted from quantified HRV. During the feature selection stage, genetic algorithm GA is used to optimize the feature set. Finally, prediction performance of the proposed method is evaluated in term of sensitivity, specificity and accuracy. Detail of each block is described in the remaining sub-sections.

2.1. Database

RR intervals recording are obtained from the Spontaneous Ventricular Tachyarrhythmia Database Version 1.0 from Medtronic, Inc. from the Physionet [16]. The database was collected from records of 78 patients with ICDs 63 males and 15 females, aged from 20.7 to 75.3 and consisted of the following RR intervals: 106 pre-VT records, 29 pre-VF records, and 135 control data sets. Each data included 1024 RR intervals corresponding to around 15 minutes. Short- term HRV analysis is performed on the 5 minutes RR intervals prior to each VTAVTVF event.

2.2. Preprocessing

RR intervals are obtained from the database. HRV is then formed by computing the reciprocal of the intervals between successive R peaks. After that, the signal is resampled to 4 Hz signal by using the cubic spline interpolation as it has been reported that this technique is better than linear interpolation [17]. Before the HRV is resampled, it is evaluated and corrected based on McNames’s algorithm [18]. This algorithm detects the abnormal heart rate signal in the series which may be caused by ectopic beat, artifact noise or miss peak detection. The algorithm evaluates heart rate using test statistic, Dn calculated with following equation: �� = | ���−���| 1.483���{|���−���|} 1 Where HR m is the average heart rate of the series, HRn is instantaneous heart rate and med{ } is median filter. If Dn value excess certain threshold, τ, the instantaneous heart rate is considered as abnormal heart rate. The heart rate is corrected with following equation: �� � � = ��� ��� + �: |�| ��−1 2 � 2 Where w m is the window length of the medium filter. In this work, τ and w m are set to 4 and 11 respectively according to the literature review [18]. Figure 2 shows 5 minutes RR intervals of a patient prior to VTA event. RR intervals before and after correction by McNames’s algorithm are shown. Figure 2. Five minutes RR intervals prior to VTA event before and after ectopic beat correction 1 2 3 4 5 0.7 0.72 0.74 0.76 0.78 Timesminute R R I nt er v al s ec ond RR intervals prior to VTA eventsafter correction 1 2 3 4 5 0.4 0.6 0.8 1 Timesminute R R I nt er v al s ec ond RR intervals prior to VTA eventsbefore correction a b  ISSN: 1693-6930 TELKOMNIKA Vol. 14, No. 3, September 2016 : 999 – 1008 1002

2.3. HRV feature extraction