An annotated dataset of gram stains from positive blood cultures

Bloodstream infections (BSIs) of high morbidity and mortality are across the age spectrum, and urgent for correct intervention. Gram stain interpretation of positive blood cultures (PBCs) is crucial for early diagnosing bloodstream infections, yet this manual process is labor-intensive, time-consuming, and highly operator-dependent. Artificial intelligence-assisted microscopic interpretation of stained smears presents beneficial to microbiology diagnostics. Addressing the auto-identification of blood-culture gram stains, this study introduces a dataset of Gram-stain smears collected in clinical practice. The dataset includes 505 pictures, and covers species variation up to 57 species of BSIs with 7,528 annotations, containing Gram-positive cocci in clusters, Gram-positive cocci in chains/pairs, Gram-negative rods, or fungus based on staining and morphology. The published dataset will help developments that utilize artificial intelligence to auto-interpretate the Gram stains from PBCs for routine clinical application.