A marrow cell dataset and baseline evaluations for precision classification of malignant tumor diseases

When physicians diagnose malignant tumors, such as acute leukemia, the rapid and precise classification of the disease is imperative for guiding subsequent treatment protocols. Traditional diagnostic methods, which rely on peripheral blood smears, are limited to assessing a narrow spectrum of cell types, thereby constraining the accurate determination of the disease subtype. Hence, a more exhaustive cytomorphologic analysis, supplemented by a bone marrow smear, is essential to devise a tailored treatment approach. Given the scarcity of comprehensive bone marrow blood smear datasets, we have developed the inaugural high-quality bone marrow cell fine-grained classification dataset (BMC-FGCD). This dataset comprises nearly 100,000 samples, encompassing 40 detailed categories. The assembly of the BMC-FGCD was overseen by a cadre of expert physicians, who meticulously managed the sample collection, image processing, and high-precision annotation. Furthermore, we employed nine well-established deep learning models to conduct baseline evaluations. These experiments provided a uniform benchmark for the assessment of bone marrow blood cell classification models. The BMC-FGCD dataset is poised not only to enhance the accuracy of bone marrow blood cell classification models in the future but also to significantly contribute to the field of rare or atypical blood cell classification due to its granularity and comprehensive data