Accessible Digital Twins for Medical Laboratories with 3D Gaussian Splatting

Abstract

Creating digital twins of medical learning environments usually requires considerable expertise and effort. 3D Gaussian Splatting (3DGS) offers a promising basis for more accessible reconstruction workflows. In this paper, we compose and compare three workflows for non-expert reconstruction, including our own open-source approach “Splatflow”. We conducted an exploratory pilot study with five technically inclined first-time users, reconstructing both an object-scale and a room-scale laboratory scene. We evaluated the workflows with usability and workload analyses as well as a subjective assessment of reconstruction quality. The pilot results suggest that all workflows are generally usable, with the specialized hardware achieving slightly better results than smartphone-based approaches.