The filtering of multilead ECGs by means of the singular value decomposition

By applying the singular value decomposition (SVD) to the multilead electrocardiogram (ECG) two sets of eigenvectors (eigen-maps and eigen-ECGs) result which are connected through the singular values. When applied to the data of individual patients the spectrum of ordered (non-negative) singular values falls off rapidly beyond a certain number k, reflecting the effective rank of the signal matrix. This implies that the recorded data may be represented very well by k singular values and the corresponding set of eigenvectors. This procedure can be interpreted as filtering, with the smaller singular values reflecting noise. The authors studied the performance of this SVD filter, in particular various methods of establishing the value of k, as well as the influence of the non-white noise