Reduced parameter space for reliable clinical identification of the ECG using neural networks and a skewed Gaussian template

In this paper, we present an improved method for the filtering, compression and clinical parameter identification of the ECG. The method involves the use of an improved instrumentation setup with high resolution and sample rate, as well as DSP monitored automatic gain and offset control (AGOC) which maximizes the range of the signal. It also involves the fitting of the signal with a nonlinear model consisting of six generalized Gaussians (with skew) and the identification of the pathologies using a neural network fed with the Gaussian coefficients. The method further includes the use of sine-fitting techniques to selectively remove electromagnetic interference such as the power-line (50 Hz or 60 Hz component). We will also describe the experimental setup developed, which is a DSP based portable 3-lead device, and show experimental and clinical results