Aims The aim of this study was to validate real-time fully convolutional networks (FCNs)
to delineate the resection margin of early gastric cancer (EGC) under indigo carmine
chromoendoscopy (CE) or white light endoscopy (WLE).
Methods We trained FCNs named “ENDOANGEL”. ENDOANGEL was tested in still images and ESD videos,
and compared with ME-NBI based on post-ESD pathology by endoscopy-pathology point-to-point
marking.
Results In ESD videos, resection margins predicted by ENDOANGEL covered all areas of cancer.The
minimum distance between margins predicted by ENDOANGEL and cancerous boundary was
3.27±1.35 mm, outperforming ME-NBI.
Conclusions ENDOANGEL has the potential in delineating resection extent of EGCs.
