On the classification of nucleated red blood cells

Abstract

Bone marrow smears stained with Giemsa were scanned with a video camera under computer control. Red cells (89 normal and 44 megaloblastic) were sampled. Each cell was digitized into 70 × 70 pixels, each pixel representing a square area of 0.04 μm 2 in the original image. The pixel gray values ranged between 0–255. Zero stood for white and 255 represented black, while the numbers in between stood for the various shades of gray. The cells were first classified into six distinct classes, each representing a red blood cell differentiation state. The canonical discrimination functions derived from this classification were then utilized for grading cell differences in a continuous fashion. Each canonical discrimination function is represented in this multidimensional space by an orthogonal axis. The distance between two points (or cells) reflects their degree of similarity. By relating this measure to a reference state, it is possible to quantitate red blood cell differentiation changes. This approach is applicable to any differentiating tissue.