Endoscopic Bladder Tissue Classification Dataset

Creators

Lazo, Jorge F.
Rosa, Benoit
Catellani, Michele
Fontana, Matteo
Mistretta, Francesco A.
Musi, Gennaro
de Cobelli, Ottavio
de Mathelin, Michel
De Momi, Elena

Description

This is the dataset used in the article Semi-Supervised Bladder Tissue Classification in Multi-Domain Endoscopic Images [1]. The dataset is composed of 1,754 endoscopic images from 23 patients undergoing Trans-Urethral Resection of Bladder Tumor (TURBT) and labeled according to their respective histopathology analysis from the resected tissue. This endoscopic procedure is usually carried out using White Light Imaging (WLI) and when available Narrow Band Imaging (NBI) is also used.

Four different classes are defined taking into consideration the general classification of Bladder Cancer as defined by the WHO and the International Society of Urological Pathology; two categories were considered for cancerous tissue: Low-Grade Cancer (LGC) and High-Grade Cancer (HGC). Additionally, two extra categories were considered for No Tumor Lesion (NTL) which comprehends cystitis, caused by infections or other inflammatory agents, and Non-Suspicious Tissue (NST).

The annotation file contains the name of each frame, the imaging type (NBI or WLI), the tissue type (HGC, LGC, NTL, NST), and the dataset in which they were used to train the classifier (train/val/test). For more details, please check the referred publication associated with this dataset.

Reference

[1] J. F. Lazo et al., "Semi-Supervised Bladder Tissue Classification in Multi-Domain Endoscopic Images," in IEEE Transactions on Biomedical Engineering, vol. 70, no. 10, pp. 2822-2833, Oct. 2023, doi: 10.1109/TBME.2023.3265679.