MM-MCD：Multiple Myeloma Bone Marrow Smear Analysis Dataset

Multiple Myeloma Bone Marrow Smear AnalysisMM-MCD is a large-scale, high-resolution, expert-annotated dataset tailored for multi-cell segmentation and fine-grained classification in multiple myeloma (MM) bone marrow smear analysis, addressing the gap of insufficiently diverse and context-rich datasets in hematological AI research. It includes about 20000 publicly available 1024×1024 pixel whole-smear images (with a full collection of 20,000 images for five-fold cross-validation) acquired under standardized clinical conditions, featuring rigorous annotations for 49 distinct bone marrow cell categories spanning plasma cell subtypes, erythroid, myeloid, lymphoid, monocytic, and megakaryocytic lineages—generated via independent labeling, cross-review, and senior hematologist adjudication with Cohen’s Kappa coefficient > 0.85 to ensure high reliability. Provided in the widely compatible COCO format with predefined 8:2 training-validation splits and detailed metadata, MM-MCD preserves the spatial context and cellular heterogeneity of real-world MM bone marrow smears (including dense cell distributions, overlapping cells, and staining variability), outperforming existing MM-focused datasets that only target plasma cells by covering the entire hematopoietic microenvironment. Intended for non-commercial research and educational use, this dataset serves as a robust foundation for developing clinically relevant computer-aided diagnosis (CAD) systems, facilitating advancements in automated bone marrow morphology analysis while supporting fair and reproducible model benchmarking.