Semantic Segmentation of Myeloid Cells on Blood Smears

Semantic segmentation is pivotal in the diagnosis of hematological conditions as it enables precise identification and classification of cell types, notably stem cells. While traditional methods have been extensively utilized, they encounter difficulties in distinguishing between different cell components like cytoplasm and nucleus, as well as varied morphological, color, and texture attributes observed in bone marrow cells. In this work, we developpe a novel semantic segmentation approach employing state-of-the-art deep learning techniques. Through the integration of convolutional neural networks and data augmentation methods, we have developed a tailored UNet model capable of accurately segmenting cytoplasm, nucleus, and red blood cells evident in microscopic images of bone marrow smears. This innovative methodology enhances efficiency and accuracy in cell identification and categorization within samples, offering substantial advantages to pathologists and researchers engaged in hematological investigations.