ABSTRACT
Background: Morphometric studies based on image analysis are a useful adjunct for quantitative analysis of microscopic images. However, effective separation of overlapping objects if often the bottleneck in image analysis techniques. We employ the watershed transform for counting reticulocytes from images of supravitally stained smears.Methods: The algorithm was developed with the Python programming platform, using the Numpy, Scipy and OpenCV libraries. The initial development and testing of the software were carried out with images from the American Society of Hematology Image Library. Then a pilot study with 30 samples was then taken up. The samples were incubated with supravital stain immediately after collection, and smears prepared. The smears were microphotographed at 100X objective, with no more than 150 RBCs per field. Reticulocyte count was carried out manually as well as by image analysis.Results: 600 out of 663 reticulocytes (90.49%) were correctly identified, with a specificity of 98%. The major difficulty faced was the slight bluish tinge seen in polychromatic RBCs, which were inconsistently detected by the software.Conclusions: The watershed transform can be used successfully to separate overlapping objects usually encountered in pathological smears. The algorithm has the potential to develop into a generalized cell classifier for cytopathology and hematology.