PBCI-DS: A Benchmark Peripheral Blood Cell Image Dataset for Object Detection

PBCI-DS contains a total of 17092 cell microscopic images, including eight types: Neutrophils, Eosinophils, Basophils, Lymphocytes, Monocytes, Immature Granulocites, Erythroblasts, Platelets. Among them, there are three types of Immature Granulocytes: Metamyelocytes, Myeloccytes, and Promyelocytes, while Platelets also include Thrombocytes. Two medical scientists manually prepare the object file in “.XML” in a format corresponding to 17092 original images (from 06-2022 to 02-2023). In experiment, we used a 5-fold comparison experiment to compare the dataset partitioning for the traditional machine learning experiment. Firstly, we divided the dataset into five parts, with four parts serving as the training set and the remaining part serving as the testing set. Using this method, the best results of the training data are ultimately divided into a training set and a testing set as the dataset.