Tackling oesophageal cancer early detection challenges through AI
Barrett’s Oesophagus is a medical condition when the cells lining the lower part of your oesophagus become damaged, which is believed to be caused by acid and bile repeatedly coming up from your stomach. It is a precancerous condition, with Barrett’s Oesophagus patients at 20-30% increased risk of getting oesophageal cancer (although only one in 860 people with Barrett’s go on to develop cancer each year).
Currently, endoscopy is the widely accepted tool for diagnosis and staging of any gastrointestinal related abnormalities, including the identification of Barrett’s Oesophagus. However, the ability to accurately detect those who are at risk (or at the early stages of) developing cancer is highly operator dependent, leaving much room for human error. As a result, many cancers are not detected as early as they could (and when they are more easily treated). Also, as the vast majority of Barrett’s Oesophagus patients never develop cancer, significant resource is wasted in unproductive screening. There is therefore a strong need to reduce this margin of error and improve the quality of endoscopic surveillance.
Dr. Sharib Ali from the Department of Engineering is working on building AI-based technologies for early oesophageal cancer detection and risk quantification. He believes that by introducing AI-based technologies that have learnt from expert knowledge, we can reduce this margin of error. This project has been ongoing since 2017, the first stage of which was focused on overcoming some of the issues related to using endoscopic videos to train AI. Sharib says, “working with endoscopy videos to build any robust methods is often challenging due to inconsistent lighting, hand motion and other debris floating.”
A year ago the team proposed very comprehensive methods using a fully automatic framework to tackle this (which you can read more about here). Since then, Sharib has built AI detection and segmentation methods to better diagnose cancer and cancer like precursors (such as Barretts oesophagus) from endoscopic videos.
He says: “Building AI technologies is getting more common in every stream of science, however, to build a technology that can generalise and adapt to changes such as patient tissue variability or simply choice of modality is a real challenge.”
He adds another challenge is to collaborate with multiple centres with a common goal and be able to deliver something that will actually transform the early oesophageal cancer diagnosis.
To date, Sharib has initiated a collaboration between 6 different institutions world-wide and this number continues to grow. The major outcome of this work is the collection and curation of multi-modality, multi-institutional and multi-population endoscopy videos and frames, which have since been made publicly available for researchers and AI enthusiasts to tackle such concrete and open problems.
As part of this ongoing project, Sharib has organised two upcoming data science challenges and workshops at the 16th and the 17th IEEE International Symposium in Biomedical Imaging which attracts over 500 researchers across the globe. He says, “due to global pandemic, this year we successfully conducted a virtual workshop via zoom and we had over 90 attendees for this. Our data science challenge is attracting more researchers and we had nearly 40% increased number from last years’.”
Sharib Ali who is located at Big Data Institute and works at Rittscher’s lab (Prof. Jens Rittscher) at the Department Engineering, Oxford, is working closely with clinical endoscopists (Dr. Adam Bailey, Prof. Simon Leedham, Prof. Barbara Braden and Dr. James East) at the TGU and molecular biologists from the Ludwig Institute for Cancer Research, Oxford Branch (including Prof. Xin Lu). Sharib’s research is funded by the NIHR Oxford Biomedical Research Center.