Multiscale entropy based multiscale principal component analysis for multichannel ECG data reduction

In this work, multiscale principal component analysis (MSPCA) is applied to multichannel ECG signals. Multiresolution analysis of multichannel ECG data using L level Wavelet decomposition gives L + 1 subbands. Considering j^th subbands of all the channels of a standard 12 lead ECG signals, subband matrices are formed at multiscale levels. At Wavelet multiscale, principal component analysis (PCA) is applied to reduce the dimensions. For the selection of significant principal components at Wavelet subband matrices, multiscale entropy and eigenvalues are considered and a new method is proposed. The reconstructed signal fidelity is evaluated using qualitative and quantitative measures such as PRD, WWPRD & WEDD. A data reduction of 48.25% in terms of samples, is achieved with average percentage root mean square difference (PRD), Wavelet weighted PRD (WWPRD) and Wavelet energy based diagnostic distortion (WEDD) of 19.98, 31.84 & 10.07 respectively with acceptable signal quality.