Object dimension measurement based on monocular endoscope and 5-DOF sensor

Background. During endoscopy, dimension measurement of the object is quintessential in diagnosis, e.g., the necessity and extent of further kidney stone fragmentation. The monocular endoscope lacks enough visual cues to measure accurately in the narrow cavity.

Methods. Firstly, a tiny five-degree-of-freedom (5-DOF) sensor is embedded into the tip of the endoscope to obtain the pose of the lens. Then, a semantic segmentation network is designed to segment the object masks automatically. Finally, the masks' dimension and the lens's pose are fused to compute the object dimension in physical units, e.g., mm.

Results. To mimic an endoscopic scenario, the authors placed kidney stone fragments of known size in the isolated porcine kidney. The mean absolute percentage error (MAPE) of measurement was 7.30%.

Conclusion. The proposed approach is safe and feasible for endoscopy. It enables automatic object dimension measurements in the limited cavity without changing existing endoscopy procedures.