AndroGen: Open-source synthetic data generation for automated sperm analysis.

BACKGROUND AND OBJECTIVE: Automated sperm analysis systems rely on machine learning models, which must be trained on large and diverse datasets. However, obtaining real labelled microscopic samples is a costly, time-consuming process, often limited by privacy concerns. This article presents AndroGen, an open-source software tool that allows generating customized synthetic images of male reproductive cell samples from different species. METHODS: AndroGen takes into account the needs of the researchers in the field. It features a user-friendly graphical interface with several predefined configurations, for the sake of easiness. Additionally, it provides dialogue controls that allows the users to set the parameters for the creation of a custom dataset, without the need to rely on a large number of real images or to use generative training models. The architecture of the system is presented in detail. RESULTS: The performance of the proposed system was evaluated on three case studies using quantitative and qualitative metrics. The similarity of the generated synthetic images to real ones is estimated with two different metrics, Fréchet Inception Distance and Kernel Inception Distance. The obtained quantitative results and the qualitative analysis show that this tool is able to generate realistic image datasets. CONCLUSIONS: AndroGen offers researchers a fast and interactive way to create realistic labelled sperm datasets. The generation process allows customization of all relevant fields in the dataset, including cell morphology and movement.