Feature Selection in Medical Signal Analysis

摘要

In order to use automated learning algorithms, important parameters or features must be identified which may be potentially useful in the development of a decision model. This process is called feature extraction. The determination of possible features for traditional numerical data is fairly straightforward. More complex problems must be addressed in the determination of features for non-textual data such as images or time series. In this paper, issues which arise in the feature extraction process for medical time series are addressed, along with examples of decision models developed by the authors for different types of medical time series, including hemodynamic studies, analysis of Holter tapes, and analysis of electroencephalograms.