Reproducibility and efficiency testing of two sampling methods for image analysis in cytology

Quantitative nuclear DNA analyses and nuclear morphometry have become a widely used tool in pathology. Nevertheless, studies done by different observers utilizing the same methods very often have shown different results. One of the reasons could be that different sampling methods were used. In this study two selection methods for image analyses (the consecutive selection method and the random selection method) were tested for reproducibility and efficiency.

Using an interactive imaging analysis system, 10 cases (5 breast adenocarcinomas, 3 prostatic cancers, 1 sarcoma, 2 prostatic hyperplasias) were measured twice by using each sampling method. Median nuclear area and entropy (i.e., a statistical descriptor of DNA histograms) were the parameters calculated.

The results showed that using the consecutive selection method, the number of individual nuclei in a cytologic sample that should be measured to obtain reproducible results was 150 for nuclear area and 110 for entropy. The corresponding figures using random systematic selection method were 80 for assessment of nuclear area and 100 for assessment of entropy.

Both methods showed high reproducibility and a good correlation with each other. The mathematical idea of "plateau values," which was used to estimate the number of measurements needed, was considered a good contribution to increasing the efficiency of both methods.