Part of the AIDA-E (Analysis of Images to Detect Abnormalities in Endoscopy) challenge




Number of users: 115

Chromoendoscopy (CH) is a gastroenterology imaging modality that involves the staining of tissues with methylene blue, which reacts with the internal walls of the gastrointestinal tract, improving the visual contrast in mucosal surfaces and thus enhancing a doctor’s ability to screen precancerous lesions or early cancer. This technique helps identify areas that can be targeted for biopsy or treatment and in this challenge we will focus on gastric cancer detection.

Gastric chromoendoscopy for cancer detection is a highly mature medical field with solid clinical taxonomies, including the most relevant one introduced by Dinis-Ribeiro [1], which is used in this challenge. According to this taxonomy, CH images are classified into their respective classes based on color, shape and regularity of pit patterns (Figure 1).

Clinical studies show that Group I images are considered normal, Group II cases are considered metaplasia lesions and could lead to cancer lesions. Group III are considered dysplasia lesions. For the purposes of this challenge, we will consider as ‘Normal’ images from Group I, and as ‘Abnormal’ images from Groups II and III.

Figure 1 – Dinis-Ribeiro gastric chromoendoscopy image classification taxonomy [1]


The CH images were acquired using an Olympus GIF-H180 endoscope at the Portuguese Institute of Oncology (IPO) Porto, Portugal during routine clinical work. Optical characteristics of this endoscope include 140o field of view and four way angulation (210o up, 90o down and 100o right/left). Annotations were performed independently by two medical experts (Dr. Dinis-Ribeiro, Dr. Miguel Areia) leading to a gold-standard final annotation which is used here. More details in [2].

The specific goal of this challenge will be to classify the manually segmented region of each image as either ‘Normal’ or ‘Abnormal’. Dataset includes both image files and region annotation files. Although not foreseen for this specific challenge, researchers can use this dataset to also evaluate their segmentation algorithms, as exemplified in [3].



  1. Please read first the challenge terms and conditions.
  2. Please register for the challenge by filling in the online registration form (Registration will open in mid-November 2015). This registration is a mandatory step before downloading data and submitting results to the challenge.
  3. Upon reception of your registration form, you will receive a link to download the dataset.
  4. After the results submission deadline the organizers will evaluate the results, and outcomes will be sent back to the authors for inclusion in their submission.



  • Mid November 2015: Release of the first training dataset.
  • Mid November 2015: Registration opens.
  • October 2015: Challenge website launched.




  1. M. Dinis-Ribeiro. Clinical, endoscopic and laboratorial assessment of patients with associated lesions to gastric adenocarcinoma. Faculdade de Medicina da Universidade do Porto, PhD thesis, 2005.
  2. F. Riaz, F.B. Silva, M. Dinis-Ribeiro, and M. Coimbra, "Invariant Gabor Texture Descriptors for Classification of Gastroenterology Images", in IEEE Transactions on Biomedical Engineering, vol. 59/10, Oct 2012, pp. 2893-2904.
  3. F. Riaz, F.B. Silva, M. Dinis-Ribeiro, and M. Coimbra, "Impact of Visual Features on The Segmentation of Gastroenterology Images Using Normalized Cuts", in IEEE Transactions on Biomedical Engineering, vol. 60, 60/5, May 2013, pp. 1191-1201.