Automatic atrial fibrillation detection on holter ECG signals

Problems arising from cardiological disorders are increasing in the world and in our country. Diagnosis of heart diseases is especially important. At this stage, developments in the field of engineering are effective. Thanks to the designed devices and softwares, assistant applications are made especially for physicians in biomedical field. Applications created in this way provide physicians facilities in diagnosis procedures. It also create extra time to determine treatment plans. In this study is aimed to automatically detect Atrial Fibrillation (AF), which is a type of bridging arrhythmia that is foremost of cardiologic disease. In the literature, many different methods such as Discrete or Continuous Wavelet Transform, Hadamard Transform, Wavelet Entropy are used for automatic AF detection. Phsiyobank ATM database is used for this study. Here, a total of 62 12-hour RR Interval (RRI) length series were obtained from Holter ECG signals of individuals with 31 pieces of Atrial Fibrillation Rhythm (AFR) and 31 piesces of Normal Sinus Rhythm (NSR). RRI arrays appear to be the most important determining factor for AF signals. This data is re-sampled by converting it to time axis. At this stage, signal processing techniques are used. Subsequently, the Discrete Wavelet Transform method was applied to the signals. On this page, the distinguishing characteristics of the signals have been determined. wavelet energies of RRI sequences with Wavelet Transform are then looked at. These properties are converted into graphics with Boxplot and the results are examined. To obtain statistical data after these operations, the wavelet energies of the RRA sequences are classified by the Support Vector Machine method and the AFR is decomposed from the NSR. When the results were examined, it was seen that AFR and NSR were separated by 99.60 % success rate.