Abstract
Multiple Myeloma (MM) is a cytogenetically heterogeneous clonal plasma cell proliferative disease whose diagnosis is supported by analyses on histological slides of bone marrow aspirate. In summary, experts use a labor-intensive methodology to compute the ratio between plasma cells and non-plasma cells. Therefore, the key aspect of the methodology is identifying these cells, which relies on the experts' attention and experience. In this work, we present a valuable dataset comprising more than 5,000 plasma and non-plasma cells, labeled by experts, along with some patient diagnostics. We also share a Deep Neural Network model, as a benchmark, trained to identify and count plasma and non-plasma cells automatically. The contributions of this work are two-fold: (i) the labeled cells can be used to train new practitioners and support continuing medical education; and (ii) the design of new methods to identify such cells, improving the results presented by our benchmark. We emphasize that our work supports the diagnosis of MM in practical scenarios and paves new ways to advance the state-of-the-art.