An analysis of red blood cell volume populations has been attempted as an aid to diagnose hematologic diseases, but its discriminability is unsatisfactory, because it has been proved useful only if the histogram indicates an abnormal distribution curve such as double peaks-curve derived from blood transfusion. In this paper, a quantitative analysis for normal distribution curve was examined by the following method.
The histograms measured by a SYSMEX PDA-400 were treated by the polynominal smoothing method, and these distribution curves were regarded as a function of frequency versus volume, and were fitted by an unweighted non-linear least square method in ten distribution models which consisted of from one to five normal distribution components or from one to five lograithmic normal distribution components, and the fitnes for models was judged by Akaike's Information Criterion (AIC).
A tri-Gaussian model was decided on an appropriate model from statistical information which showed the lowest AIC in any histograms on both subjects and osmolality of suspensions. The results made clear that the difference of distribution curve was mainly dependent on the difference of mean corpusuclar volume belonging to the largest volume distribution of three Gaussian components, and the relative area (expressed as per cent of the total count) of minor-volume distribution of them decreased and the largest increased under hypotonic osmotic stress.