HIDDEN MARKOV MODELS IN HOLTER RECORD ANALYSIS

点击翻译按钮获取中文摘要

Although long term (24-Hour) electrocardiographic monitoring of ambulatory patients has become an integral part of medical diagnosis, analysis of recorded waveforms has advanced little since introduction of the screening procedure by Holter in 1957. Research into computerized systems using traditional pattern recognition techniques and interval analysis has not yielded a clinically reliable automated method for Holter analysis. This laboratory has employed syntactical pattern recognition in conjunction with an expert system to successfully analyze most ventricular arrhythmias. However, reliable computer detection of the P wave has limited the syntactic approach in analyzing atrial arrhythmias. P Wave detection is difficult because it is low in amplitude and easily obscured by artifact common in ambulatory records. Recently, we have tested a technique called "hidden Markov modeling" (HMM), used quite successfully in computerized identification of human speech. These preliminary tests have shown it to be surprisingly effective in computer identification of low amplitude waves of the ECG (such as the P wave) even in the presence of sufficient noise to visually obscure them. Our intent is to investigate this technique in the computerized analysis of arrhythmias in Holter records. Our eventual goal is development of a fully computerized system for arrhythmia analysis using both HMM, should our investigation verify applicability.