Application of the method of multiple thresholding to white blood cell classification

A new method of image segmentation based on the principle of multiple grey level thresholding has been applied to a data set consisting of 1149 white blood cells of 13 different, clinically important types, randomly chosen on 20 blood smears from leukemia patients. Classification of these cells on the basis of quantitative measurements in the segmented images yields an accuracy of 82.6%. Some of the erroneous classifications must be attributed to intrinsic problems in the assignment of a priori labels. Correcting for such cases, the performance of the method, as measured on the present data set, increases to 89.8%. This illustrates the practical applicability of the segmentation method in automated white blood cell and possibly other cytological and histological analysis systems.