Histopathological Digital Image Dataset of Upper Aero Digestive Tract Tumor
Description
The UADT tumor dataset is sourced from the Department of Pathology at Government Medical College Hospital in Vellore, Tamil Nadu, India. Reference number 2941/ MEI (2)/ 2018, Government Order No. 1258/ Health & Family Welfare (MCA3). Following the removal of 217 samples from various tumors due to poor staining, improper waxing, and corrupted image quality (blur, water blob artifact, and fractured tissue sample) the dataset comprises 1,471 biopsy slide images obtained from a retrospective investigation involving 20 Indian patient biopsy slides from the pathology department. The data samples contain Haematoxylin and Eosin-stained microscopic images from both small and large biopsies, totalling 1,471 images. The digital images of the biopsy slides were obtained using three distinct magnifications: scanner, low power, and high power. The collection includes scanner view Whole Slide Images (WSI), Low Power view at 10X magnification, and High-Power view at 40X magnification. The digital images were recorded at a resolution of 640 × 480. The images were taken using a 5.1megapixel camera connected to the light microscope. The dataset comprises various classifications of UADT biopsy images, including inflammatory, no evidence of malignancy, insitu, mild dysplasia, moderate dysplasia, severe dysplasia, Basaloid Squamous cell carcinoma (BSCC), well-differentiated squamous cell carcinoma (WDSCC), moderately differentiated squamous cell carcinoma (MDSCC), and poorly differentiated squamous cell carcinoma (PDSCC). This dataset was collected to support digital and AI-based solutions for glass slide analysis, including artifact management, slide quality enhancement, automated extraction of epithelial tissue and intercellular bridges, detection of cell and nucleus shapes to assess tumor normality and spread, identification of cell overlap and necrosis, contrast enhancement and background separation, and detection of staining artifacts collectively aiming to minimize interobserver variability and reduce glass slide review time.
Files
Steps to reproduce
Ethical clearance and sample sourcing: Biopsy slides were retrospectively obtained from the Department of Pathology, Government Medical College Hospital, Vellore, Tamil Nadu, India, under institutional approval (Reference No. 2941/MEI(2)/2018, Government Order No. 1258/Health & Family Welfare (MCA3)). A total of 20 Indian patient biopsy slides were retrieved from archived departmental records for the study. Sample classification: A senior pathologist (Associate Professor, Department of Pathology) examined and classified each biopsy slide according to standard histopathological diagnostic criteria, categorizing samples into inflammatory, no evidence of malignancy, in-situ, mild dysplasia, moderate dysplasia, severe dysplasia, basaloid squamous cell carcinoma (BSCC), well-differentiated squamous cell carcinoma (WDSCC), moderately differentiated squamous cell carcinoma (MDSCC), and poorly differentiated squamous cell carcinoma (PDSCC). Slide preparation and collection: Two trained laboratory technicians assisted in the retrieval and preparation of the glass slides. All slides were stained using the standard Hematoxylin and Eosin (H&E) staining protocol prior to image acquisition. Image acquisition setup: Each glass slide was examined under a light microscope fitted with a 5.1-megapixel digital camera mounted directly onto the microscope head to enable image capture at multiple magnifications. Image capture: Digital images were captured at three magnification levels for each slide — a scanner-view Whole Slide Image (WSI), a Low Power View (10X), and a High Power View (40X) — to preserve both the overall tissue architecture and cellular-level detail. A total of 1,471 microscopic images were captured across the 20 biopsy slides. Real-time image analysis: MagVision microscopic imaging software was used in conjunction with the mounted camera to view, capture, and analyze the tissue regions of interest during the imaging session, allowing the operator to verify focus, field selection, and magnification in real time. Annotation and organization: As each image was captured, it was labeled according to its corresponding diagnostic classification (as assigned by the pathologist in Step 2) and magnification level, and was immediately sorted into designated folders corresponding to its diagnostic category, ensuring a traceable and organized dataset structure at the point of acquisition. Dataset compilation: The captured and annotated images were compiled into the final dataset comprising 1,471 H&E-stained biopsy images spanning ten diagnostic categories, intended for downstream use in digital pathology and AI-based image analysis research.
Institutions
- Vellore Institute of Technology UniversityTamil Nadu, Vellore