The morphological classification of red cells using an image analysing computer.

Abstract

Image analysis provides a potential method for automation in the examination of blood films. In order to assess the applicability of this technique to red cell morphology, a scheme has been defined for the classification of red cells into six categories, namely, round cells, elongated cells, tear drop poikilocytes, helmet cells, irregular cells and spherocytes. A study was carried out to determine the consistency which observers classified red cells according to this scheme. Measurements of area, perimeter, maximum diameter and integrated optical denisty of fixed, Romanowsky-stained red cells were made using an image analysing computer. A multivariate classifier based on parameters derived from these measurements, namely cell area and the quotients of integrated density over area, area over the square of perimeter and perimeter over maximum diameter was also used to classify red cells into the above categories. It was found that the inherent error of the machine-based classifier was of the same order as the degree of inconsistency encoutered between trained observers. The significance of these findings is discussed.