Cookies on this website

We use cookies to ensure that we give you the best experience on our website. If you click 'Accept all cookies' we'll assume that you are happy to receive all cookies and you won't see this message again. If you click 'Reject all non-essential cookies' only necessary cookies providing core functionality such as security, network management, and accessibility will be enabled. Click 'Find out more' for information on how to change your cookie settings.

Early pre-cancerous malignant condition such as Barrett’s esophagus (BE) and quantification of associated dysplastic changes is critical for early diagnosis and treatment. While endoscopic videos are corrupted with multiple artefacts and procedure require investigating extended areas such as stomach, it is inevitable that there is risk of missing areas that may potentially harbour neoplastic changes and require immediate attention. A complete guidance assisting navigation is thus vital. Visually obvious neoplasia and suspected areas both needs be flagged for biopsies. Due to the thin demarcation between BE and early neoplasia, it is often challenging to identify subtle changes even by experts. We propose a self-supervised learning technique for a fully assistive esophageal endoscopy surveillance system. The self-supervision step allows to learn complex representations using a pretext task of solving a jigsaw puzzle. Here, the idea is to enable network to distinctly learn inconspicuous features that are characteristics of neoplasia and other classes. In order to enable an optimal decision boundary we propose to incorporated angular margin in our fine-tuning process. Our proposed framework showed a boost of 3% on overall accuracy compared to fully supervised approach with similar backbone.

Original publication




Conference paper

Publication Date



13581 LNCS


14 - 23