Observer reproducibility during computer-assisted planimetric measurements of nuclear features.

To evaluate observer variance during microprocessor-assisted planimetry, nuclear features (area, perimeter, and form factor) were studied in a series of mammary ductal carcinomas. Fragments of formalin-fixed, paraffin-embedded tissue were re-embedded in plastic, sectioned at 1 micron, stained with methylene blue, and studied with a Zeiss-Kontron MOP-3 microprocessor-assisted planimeter. Both interobserver and intraobserver reproducibility were evaluated, the former among three different observers and the latter by two observers repeating their own measurements. Reproducibility was good for measurement of area, but deteriorated progressively for measurement of perimeter and form factor. Not only was observer correlation poor (identified via linear regression), but paired t-tests also showed consistent variation among observers. The major difficulty was in following the irregular nuclear contours that are characteristic of cancer cells. It is concluded that adequate demonstration of observer reproducibility remains an essential part of tissue investigation, even when the objective methods typical of morphometry are used.