Spectral vs. compartmental averaging of VA/Q distributions: confidence limits

We have investigated the method of statistical averaging as a nonparametric approach to obtain a representative ventilation-perfusion (VA/Q) distribution that exemplifies the family of compatible solutions for multiple inert gas elimination data. The variability of the compatible solutions was examined by determining the standard deviation of the statistical average. For six inert gases, it can be predicted that a distribution with up to seven contiguous nonzero VA/Q compartments can be uniquely recovered, whereas the compatible family becomes more diverse, the broader the distribution. For a given compatible family consisting of multimodal distributions with various phase relationships, the average distribution was found to display an uncharacteristically unimodal shape as a result of modal smoothing. To avoid this possible artifact, an alternative approach was adopted in which statistical averaging was performed in the frequency domain. For both deterministic and empirical data, the energy spectra of all feasible VA/Q distributions displayed a well-defined low-frequency band that was invariant within the compatible family and with a bandwidth that approximated the predicted sampling cutoff frequency. The nonuniqueness of the result was ascribable to a variable high-frequency band that was due to an aliasing effect. For a wide range of clinical data, the representative distributions resulting from compartmental and spectral averaging were indistinguishable from each other and had little variability both in the VA/Q and frequency domains. For these cases, therefore, the resolving power of the recovery algorithm was not critical. Finally, an efficient method of finding the average distribution was proposed.