Automated scoring of micronuclei in binucleated human lymphocytes

Manual and automatic scoring of micronuclei (MN) in binucleated human lymphocytes (BNC) were compared after irradiation of whole blood samples. The blood samples were irradiated with X-ray doses (1, 2 or 3 Gy) and stained with Giemsa. The preparation technique was optimized in such a way that acceptable conditions (cell density, contrast) were obtained for both scoring procedures. To estimate the quality of automatic micronucleus detection, two researchers who had different experience in scoring MN (6 months and 5 years) analysed the samples independently from each other. Automatic scoring was carried out with a digital image analysis system and the recognition procedure was divided into two parts. The BNC positions were detected with low microscope magnification (100x), and the recognition of micronuclei within the cytoplasm of the classified BNC was carried out at high magnification (630x). A fuzzy logic classification system as well as two different segmentation steps (preclassification and postclassification) made it possible that about 94% of all automatically recognized BNC were classified correctly). On the other hand, the classification system was optimized in such a way that false positive decisions were minimized (95% of automatically recognized micronuclei were classified correctly). Failure to recognize micronuclei (8.5%-25% false negatives) was mainly due to extremely small micronuclei, poor contrast with respect to the cytoplasm, and aggregation of micronuclei especially at higher doses.