Related Experiment Video
Updated: Aug 5, 2026
Robotized Testing of Camera Positions to Determine Ideal Configuration for Stereo 3D Visualization of Open-Heart Surgery
Published on: August 12, 2021
A Pilot Study on the Efficacy of Artificial Intelligence-Driven Monocular Three-Dimensional Conversion for Endoscopic
Yosuke Sato1,2, Kosuke Tanaka2
1Neurosurgery, Showa University, Tokyo, JPN.
Background:
Monocular two-dimensional endoscopes inherently lack depth perception, significantly increasing the cognitive load during surgical procedures. Artificial intelligence has the potential to convert two-dimensional video feeds into stereoscopic three-dimensional views in real time, yet its depth-dependent limitations remain underexplored. This pilot study evaluated the efficacy and spatial limitations of an artificial intelligence-driven monocular three-dimensional conversion system.
Methods:
Six participants performed a surgical target-grasping task within a dry box simulator under three distinct visual conditions: a monocular two-dimensional baseline, an artificial intelligence-generated three-dimensional system, and a native stereoscopic three-dimensional system. Targets were placed at three specific working distances of 3 cm, 5 cm, and 10 cm, yielding a total of 90 independent trials.
Results:
The overall grasping success rate was 33.3% for the two-dimensional baseline, 52.2% for the artificial intelligence system, and 68.9% for the native three-dimensional system. Statistical analysis confirmed that the artificial intelligence conversion significantly improved the success rate over the standard two-dimensional environment. Furthermore, the artificial intelligence system drastically reduced the average task completion time to 5.25 seconds compared to 7.93 seconds in the two-dimensional setting. A distance-based sub-analysis revealed a notable physical limitation. The artificial intelligence system provided exceptional spatial support, comparable to native three-dimensional systems, at close working distances of 3 cm and 5 cm. However, at a deeper distance of 10 cm, the monocular depth cues diminished, leading to a decline in estimation accuracy.
Conclusions:
Real-time artificial intelligence-driven three-dimensional conversion successfully enhances spatial perception and operational efficiency in surgical environments. While it serves as a highly effective visual assist tool at standard close working distances, operators must remain aware of its limitations in deeper operative spaces.
