Recognition of the Life-Threatening Cardiac Arrhythmias in the Frequency Domain

Abstract

During the clinical monitoring of the human heart activity the main goal is to detect heart arrhythmias and capture their precursors as early as possible. We decided to divide ECG fragments into six classes depending on their danger to the human life. As a first step in solving the problem we have grouped these classes into two parts: threatening humans’ life and others. For maintaining low response time of arrhythmia detection during long-term monitoring, we are using a 2 seconds long ECG fragments. As a classification features Fourier transform with spectrum up to 15 Hz were picked. In this paper we describe the formed dataset of ECG fragments and compare efficiency of different simple classification algorithms for this two-class problem. The following algorithms were tested: k-nearest neighbors, nearest convex hull algorithm, nearest mean and SVMs with different kernels. The results appeared to be sufficiently appropriate.