Systematic random sampling for selective interactive nuclear morphometry in breast cancer sections. Refinement and multiobserver evaluation.

Interactive nuclear morphometry in histologic tumor sections is widely employed in quantitative pathology because it can add to objectivity with regard to the diagnosis and prognosis of cancer patients. Among many sources of variation, it is mainly the sampling method chosen that has far-reaching consequences for the reproducibility of nuclear measurements. Previous studies have suggested that systematic random sampling gives the best reproducibility, but as yet, extensively tested and practicable implementation of interactive nuclear morphometry has been described only rarely. With this study we tried to fill this gap. First, different systematic random sampling schemes were explored with computer simulation. That sampling scheme that resulted in the smallest measurement error was translated into a workable method. By this method nuclear measurements were conducted within small measurement frames that were regularly spread across the measurement field according to a certain distribution rule. Second, this technique was evaluated in a multiobserver study measuring breast cancer sections with an interactive digitizing video overlay system. In addition, it was compared to the more traditional zone method, in which sampling is confined to a rather small portion of the measurement field. The computer simulations showed that for a given sample size, the fewer nuclei that are assessed per measurement frame and the better their spread, the higher the measurement precision becomes. The multiobserver study demonstrated the practicability of such a sampling approach. Assessing one nucleus per measurement frame only, the measurement precision was much better than with zone sampling.(ABSTRACT TRUNCATED AT 250 WORDS)