KAUHC
Description
Wireless Capsule Endoscopy (WCE) has revolutionized the diagnosis of small bowel (SB) abnormalities by offering a non-invasive and comprehensive view of the SB, previously challenging for traditional endoscopic techniques. This dataset introduces the King Abdulaziz University Hospital Capsule (KAUHC) dataset, a significant resource comprising annotated WCE images. The dataset addresses the scarcity of labeled endoscopic imaging resources, particularly in the Middle East, and aims to facilitate the development of local Deep Learning and Computer Vision systems. The dataset includes 10.7 million frames from 157 studies, categorized into normal, Arteriovenous Malformations, and ulcer classes. This paper details the dataset's collection, annotation process, and evaluation using machine learning models, demonstrating its potential in enhancing automated diagnostic tools for SB abnormalities.
Files
Steps to reproduce
The collection process entails seven phases. Initially, the process commences with patient admission, preparation for GI tract screening, and the ingestion of the OMOM capsule. The second phase involves the capture, wireless transmission, and recording of a sequence of videos using the OMOM system. Then, Gastroenterologists conduct a manual review of the recorded videos in order to not only identify points of interest (POIs) but also choose the most relevant frames in terms of pathological abnormalities. Subsequently, the POI frames are exported, labeled, and submitted for the validation phase. These annotated frames serve as training and testing sets and are utilized as as input for ML and DL classification models. Finally, the evaluation metrics were applied to assess the study's outcomes.