Filter bank-based ECG beat classification

A multirate digital signal processing algorithm to classify heart beats in the electrocardiogram (ECG) is presented. The algorithm incorporates a Filter Bank (FB) which decomposes the ECG into subbands with uniform frequency bandwidths. The FB-based algorithm enables independent time and frequency analysis to be performed on the ECG. Various features are computed from the subbands to distinguish between between paced and non-paced beats. These features are computed in a very computationally efficient manner. A machine learning method is used to design a decision tree to be used in the classification process. The classification algorithms were tested on records from the MIT/BIH database. The paced beat classification algorithm has a sensitivity of 87.64% and a positive predictivity of 90.97%. The FB-based structure is useful for performing multiple ECG processing tasks using one set of preprocessing filters