PCMMD: Plasma Cells for Multiple Myeloma Diagnosis

Description

PCMMD (Plasma Cells for Multiple Myeloma Diagnosis) is an important resource for training and validating Artificial Intelligence (AI) approaches to automate cell identification, supporting studies on MM in developing countries, and discovering approaches regarding other diseases affecting various cell types in diverse populations. The dataset is organized into three sets of cells stained with Wright-Giemsa: - The first set contains 3,546 cells, with 54% labeled as non-plasma and 46% as plasma cells; - The second set includes 2,021 cells organized by patients, along with their respective diagnoses; - The third set comprises 1,615 plasma and 1,931 non-plasma cells, individually segmented without background. All sets were used to train Deep Neural Network models, whose details and parameters are detailed in the following repository: https://github.com/LabIA-UFBA/MMDB

Steps to reproduce

In the first stage, groups of patients with and without MM, who were submitted to a bone marrow aspirate procedure, thus resulting in a set of histological slides. The BM slides were examined using a Nikon ECLIPSE CI visible light optical microscope with immersion oil, employing a 100x objective lens and a 10x ocular lens. Observations were conducted at the smear's feathered edge. Nucleated cells were photographed using a smartphone camera mounted on the microscope with a universal holder. Notably, our dataset comprises both mono- and multinucleated cells. In the last stage, cell images from patients with and without MM were individually labeled and segmented by using the LabelImg and AnyLabeling tools, respectively.

Institutions

Categories

Funding

Licence