A comprehensive raw dataset of Ziehl-Neelsen-stained sputum smear microscopy images for Mycobacterium tuberculosis detection.

This data article presents a comprehensive dataset of 1,438 sputum smear microscopy images containing 11,447 ground-truth bounding-box labels for the detection of Mycobacterium tuberculosis. To address the scarcity of diverse real-world medical imaging data, the clinical specimens were captured using two distinct digital microscope camera systems (Hayear and Optilab). Instead of utilizing computational enhancements that may introduce synthetic biases, this dataset intentionally preserves the raw illumination and color characteristics of the slides. To facilitate robust model evaluation, the dataset includes a detailed metadata.csv file that categorizes each image based on its natural background color variation (e.g., Greenish, Bluish, Purplish/Pinkish, Yellowish) caused by differences in staining thickness and camera sensor responses. All annotations are natively formatted in the YOLO bounding-box format, making this dataset immediately viable for training, validating, and benchmarking automated object detection models for tuberculosis screening.