Observer variation in interactive computerized morphometry.

Abstract

The validity of a test system in morphometric histopathology depends on the variation sources involved. Biological variation is one of the variation sources, but is also the object to be studied with morphometry. We studied the variation sources in interactive computerized morphometry in two test systems involving two different commercially available image analyzers. The measurements were made on nuclei in a microscopic field of 5 transitional cell tumors (papilloma and WHO grade I-III carcinomas) of the urinary bladder. It turned out that different observers selected a variable number of nuclei for measurements, the coefficient of variation (CV) in the number of nuclei being 11-11.5%. Mean CV in nuclear perimeter measurements was 4.4-4.8%, and in nuclear area measurements 8.2-8.6%. The variation in measurements by 1 observer was smaller, the CV values being 2.8% for the number of nuclei, 1.2% for nuclear perimeter, and 2.4% for nuclear area. The results showed that interobserver variation can be considerable in these systems. It is suggested that special sampling rules should be tested with the idea of finding the relevant approach with the smallest interobserver variation.