Virtual Chromoendscopy with Tunable Visibility Enhancement
arXiv:2607.12416
The paper introduces a virtual enhanced chromoendoscopy (V‑ECE) system that uses a tunable image‑translation model to generate endoscopic images with adjustable visibility enhancement, eliminating the need for physical dye while improving lesion detection.
Abstract
Chromoendoscopy (CE) is a common clinical practice that sprays indigo carmine blue dye onto the gastric surface to improve the visibility of gastric lesions, such as an early cancer. While CE is effective in detecting the lesions, preparing and spraying the dye needs additional cost and time, which is undesirable both for patients and medical practitioners. To overcome this issue, virtual chromoendoscopy (V-CE) was recently proposed, which applies a learned image translation model to virtually generate a CE image from a standard endoscopy (SE) image. In this paper, we propose virtual enhanced chromoendoscopy (V-ECE) that combines V-CE with image enhancement techniques to further improve the visibility of gastric lesions. Because a desired enhancement level depends on the inspected lesion and the practitioner's preference, we introduce a novel image translation model that can generate V-ECE images using an enhancement level tunable by a user. Experimental results demonstrate that our proposed model can plausibly generate V-ECE images with various enhancement levels using a unified model.
7 pages, 8 figures. Accepted at EMBC 2026. Project page: http://www.ok.sc.e.titech.ac.jp/res/VIC/