Effects of statistical noise and digital filtering on the parameters calculated from the impulse response function

Radiotracers are widely used for the investigation of organ perfusion and function. One of the quantitative approaches to analyze radiotracer data is the calculation of the impulse response function, which is obtained by deconvolution analysis of the time-activity curves measured over the organ. Since exactness of the calculated impulse response function depends both on the counting statistics and on the deconvolution algorithm applied, computer simulated time-activity curves were used to test the least squares deconvolution program based on the matrix regularization algorithm. Criteria of clinical importance (error in the calculated organ function parameters) and criteria of mathematical importance (deconvolution and reconvolution error) were investigated. For three typical impulse response functions f(t), it was found that: 1. In cases of noncompartmental vascular-capillary f(t)'s, a high degree of smoothing is preferable during deconvolution, in this way the error becomes systematic but controllable. 2. Noncompartmental vascular-tubular f(t)'s are noise sensitive, but fortunately, noise in the data can be held to a minimum. 3. Compartmental f(t)'s need only a minimal degree of smoothing; their components can be identified in a second step using a multiexponential least squares fit.