An Algorithm for Evaluating the Performance of Adaptive Filters for the Removal of Artifacts in ECG Signals

Filtering electrocardiogram (ECG) signals calls for a filter whose impulse response can be automatically adjusted according to the varying characteristics of the signal and artifacts. In order to eliminate effectively the artifacts in ECG signals, we propose the unbiased linear artificial neural network (ULANN) as a new type of adaptive filter. This paper compares the performance of the ULANN filter with the prevailing least-mean-squares (LMS) and recursive-least-squares (RLS) adaptive filters, for the removal of artifacts in noisy ECG signals. The measures of performance include the root-mean-squared error, a normalized correlation coefficient (MX), and entropy. A template derived from each ECG signal is used as a reference to derive the measures. The NCC values for the ULANN, LMS, and RLS filter, averaged over 22 ECG signals, are 0.9956 plusmn 0.0022, 0.9948 plusmn 0.0020, and 0.9940 plusmn 0.0026, respectively. The results indicate that the ULANN filter provides filtered signals with the highest waveshape fidelity among the three filters studied.