Deep machine learning algorithms and deep neural networks are widely used for accurate classification and realistic image generation, requiring image datasets for training. This paper focuses on biomedical image datasets, comparing known datasets, and highlighting key indicators. It describes the development of datasets containing real cytological, histological, and immunohistochemical images, detailing the preparation, digitization, and storage processes. A database for these real biomedical images was designed and implemented, with tables of the dataset model explained. Additionally, the paper discusses the development of GAN architectures for generating synthetic cytological images, detailing the parameters of the GAN generator and discriminator, and comparing synthetic biomedical image databases. The dataset model for synthetic cytological images is also described, including the detailed tables. The research led to the creation of two certified databases: “IHCDBI” (the copyright certificate No. 118979) for storing digital immunohistochemical images of breast cancer, containing information about medical studies, patients, images, and their characteristics; and “BCADCID” (the copyright certificate No. 123607) for organizing data related to synthetic image creation using GANs.