Project Overview
In this work, we address the practical limitations of indigo carmine chromoendoscopy (IC) in early gastric cancer (EGC) diagnosis, including procedural complexity and additional examination time, by proposing a deep-learning–based virtual chromoendoscopy system using a cycle-consistent generative adversarial network (CycleGAN). Our approach enables dye-free generation of virtual IC images from standard white-light endoscopy (WLE), providing enhanced mucosal contrast without disrupting clinical workflows. We evaluate the proposed system using endoscopic videos, demonstrating that virtual IC improved visibility compared with WLE. By improving the performance of virtual IC equal to real IC, our system may serve as a useful alternative to real IC, supporting its potential clinical utility in enhancing visibility of EGC.
Method Overview
Video Results
Publications
-
Takasu A., Suzuki S., Monno Y., Minai M., Hirasawa T., Yamamoto H., Ishibashi F., Nishizawa T., Okutomi M., Tada T.
Assessment of early gastric cancer visibility in deep-learning-based virtual indigo carmine chromoendoscopy (with video).
Endoscopy International Open, 14, 2026. Paper link -
Suzuki S., Monno Y., Arai R., Miyaoka M., Toya Y., Esaki M., Wada T., Hatta W., Takasu A., Nagao S., Ishibashi F., Minato Y., Konda K., Dohmen T., Miki K., Okutomi M.
Diagnostic performance of deep-learning-based virtual chromoendoscopy in gastric neoplasms.
Gastric Cancer, 27(3):539–547, 2024. Paper link