Effects of statistical quality, sampling rate and temporal filtering techniques on the extraction of functional...

Several techniques for the processing of a series of curves derived from two left ventricular time-activity curves acquired at rest and during exercise with a nuclear stethoscope were evaluated. They were three and five point time smoothing, Fourier filtering preserving one to four harmonics (H), truncated curve Fourier filtering, and third degree polynomial curve fitting. Each filter's ability to recover, with fidelity, systolic and diastolic function parameters was evaluated under increasingly 'noisy' conditions and at several sampling rates. Third degree polynomial curve fittings and truncated Fourier filters exhibited very high sensitivity to noise. Three and five point time smoothing had moderate sensitivity to noise, but were highly affected by sampling rate. Fourier filtering preserving 2H or 3H produced the best compromise with high resilience to noise and independence of sampling rate as far as the recovery of these functional parameters is concerned.