An adaptive learning algorithm for ECG noise and baseline drift removal

Electrical noise and power line interference may alter ECG morphology. Noise reduction in ECG is accomplished applying filtering techniques. However, such filtering may mutate the original wave making difficult the interpretation of pathologies. To overcome this problem an adaptive neural method able to filter ECGs without causing the loss of important information is proposed. The method has been tested on a set of 110 ECGs segments from the European ST-T database and compared with a recent morphological filtering technique. Results showed that morphological filters cause inversions and alterations of the original signal in 65 over 110 ECGs, while the neural method does not. In 96% of the cases the signal processed by the network is coherent with the original one within a coherence value of 0.92, whereas this values for the morphological filter is 0.70. Moreover, the adaptability of the neural method does not require estimating appropriate filter parameters for each ECG segments.