Expert-Annotated Optical Microscopy Images of Human Sperm for Detection and DNA Fragmentation Assessment

We present a collection of expert-labeled image datasets designed to support automated analysis of human sperm in the context of male fertility diagnostics. The collection includes three datasets: a raw set of over 400 high-resolution, stained bright-field images containing multiple sperm cells; a binary classification set with cropped image patches labeled as either sperm or non-sperm; and a multiclass dataset of individual sperm cells categorized by DNA fragmentation level, based on halo size (large, medium, small, or no halo). All annotations were independently performed by five experienced embryologists, with final labels determined by majority agreement. Together, these datasets provide a foundation for training and benchmarking machine learning models for sperm detection, classification, and chromatin integrity assessment. The data are openly available to facilitate reproducibility and support further research in reproductive health.