A retrospective feasibility study reported that an AI-based Dense-UNet model can reconstruct 3D stomach models from routine gastroscopy videos. Across eight patients with superficial gastritis, the approach increased the number of 3D points by more than 2.93× compared with traditional methods. The reported gains focus on the practical ability to generate usable 3D geometry from standard clinical imaging streams, which could support planning and visualization tools in endoscopy and minimally invasive workflows. Because the study is explicitly feasibility-focused and small, it primarily informs technical direction rather than clinical impact or diagnostic accuracy claims.