“…We compare our MK nuclei detection method with an automatic unsupervised MK detection framework [42], [43], which we implement in Matlab, and the supervised MK nuclei detection method previously proposed in [44], which uses an Adaboost classifier to generate the cytoplasm mask. We use 50 bone marrow trephine images, including 73 megakaryocytes, of early-stage ET and PMF diseases in this evaluation; 50% of the images are used for training and 50% for testing.…”