Decomposition of tensors of cardio-vascular signals using CANDECOMP / PARAFAC algorithms

Abstract

In the processing of biological signals of the electrocardiogram (ECG) and Arterial Diameter Variation (ADV) there are several methods for the extraction of cardiovascular event characteristics. In this case, the canonical polyadic decomposition of CANDECOMP/PARAFAC (CP) tensors is used in the processing of the mixed signals of ECG-ADV; the ECG-ADV signals are extracted from each patient simultaneously and the detection quality of the ECG complex is verified. To do this, the ECG complex and the systolic wave of the ADV wave are aligned with Gaussian noise, and then the tensors were constructed for both signals. Five algorithms of CP were applied and the quality of the factorization of each one of the algorithms was checked with four indices. Both signals were shown to be non-collinear, and the algorithms that have the minimum number of iterations at the convergence were determined.