Computer assisted efficiency testing of different sampling methods for selective nuclear graphic tablet morphometry.

Abstract

Selective nuclear graphic tablet morphometry is widely employed as a useful tool in quantitative pathological assessments. The value of such a measuring system is strongly determined by the sampling rule that has to ensure high measurement precision with minimal effort. Therefore, the efficiency of four different random sampling methods (i.e., the pure "random," "zone," "at convenience," and "raster" methods) was tested by means of computer simulated sampling in nuclear study populations derived from histological sections of three endometrial hyperplasias and nine endometrial carcinomas. The 12 nuclear study populations each consisted of 1000 intact nuclei that were systematically measured within demarcated measurement fields by an experienced morphometrist. To attain an arbitrarily chosen measurement precision of 2.5% the computer simulations showed that the random and the raster methods required much smaller sample sizes than the zone and at convenience methods. By using the random and raster methods for assessment of the nuclear area, a maximal sample size of 200 nuclei was necessary, whereas the zone and at convenience methods mostly required more than 200 nuclei. For features such as the perimeter, longest axis, shortest axis, and two shape factors (form-PE and roundness), and the same level of precision, the sample size could be considerably smaller. The differences in efficiency can be explained by the existence of clustering and gradients in the value distribution of a nuclear feature within the measurement field. In contrast to the zone and at convenience methods, the random and raster methods sample nuclei from the entire measurement field and thus do not limit sampling to only a few larger areas within that field. Consequently, the latter two sampling methods cope better with an uneven spatial distribution in the magnitude of a nuclear feature and can thereby help to keep the measurement reproducibility high.