PCA-based noise reduction in ambulatory ECGs
Iñaki Romero · Computing in cardiology · 2010
PCA can be used for cleaning noisy ECGs. With this aim, ECG with artificial motion artifacts were generated by combining clean 8-channel ECG with noise signals. 8-channel PCA was applied and then inverted after selecting a subset of principal components (PC). Input and output of PCA filtering was compared by calculating the correlation coefficient and estimating the SNR. Above 0dB, the PC corresponding to highest variance gave best performance, below 0dB the best PC was the second highest or lower variance. When SNR decreased, PCA performed better when retaining more number of PCs (3 PCs for a SNR=10dB down to 6 out of 8 PC for SNR=−10dB). Reducing the number of input ECG channels did not yield to a significant difference when it was reduced from eight down to two. A method for identifying the optimal subset of PC as a function of input SNR and number of channels was proposed. This method achieved an SNR improvement of 0.95dB–1.92dB.