A Comprehensive Raw Dataset of Ziehl-Neelsen Stained Sputum Smear Microscopy Images for Mycobacterium Tuberculosis Detection

This dataset provides a comprehensive collection of raw Ziehl-Neelsen-stained sputum smear microscopy images for Mycobacterium tuberculosis bacilli detection. The dataset contains 1,438 raw microscopy images and 11,447 manually annotated bounding-box labels in YOLO format. Images were acquired using two digital microscope camera systems, Hayear and Optilab, to introduce multi-sensor variability and better represent real clinical microscopy conditions.

The dataset intentionally preserves the original raw image characteristics, including natural illumination variation, staining differences, background color shifts, sensor response, and microscopic artifacts. No synthetic image enhancement, color normalization, contrast adjustment, or artificial augmentation was applied to the released images. This makes the dataset suitable for evaluating object detection models under realistic clinical imaging conditions.

A metadata.csv file is provided to support stratified analysis. The metadata includes Image_ID, Data_Split, Background_Color, and Camera_System information. The Background_Color field categorizes images into Yellowish, Purplish/Pinkish, Bluish, and Greenish profiles, while the Camera_System field identifies whether each image belongs to the Hayear or Optilab camera group. These metadata fields enable researchers to analyze model performance across different staining appearances, illumination conditions, and camera-system characteristics.