Structured light for touchless 3D registration in video-based surgical navigation

Arthroscopic surgery, with its inherent difficulties on visibility and maneuverability inside the joint, poses significant challenges to surgeons. Video-based surgical navigation (VBSN) has proven to have clinical benefits in arthroscopy but relies on a time-consuming and challenging surface digitization using a touch probe to accomplish registration of intraoperative data with preoperative anatomical models. This paper presents an off-the-shelf laser scanner for noninvasive registration that enables an increased area of reachable region.

Our solution uses a standard arthroscope and a light projector with visual markers for real-time extrinsic calibration. Nevertheless, the shift from a touch probe to a laser scanner introduces a new challenge-the presence of a significant amount of outliers resulting from the reconstruction of nonrigid structures. To address this issue, we propose to identify the structures of interest prior to reconstruction using a deep learning-based semantic segmentation technique.

Experimental validation using knee and hip phantoms, as well as ex-vivo data, assesses the laser scanner's effectiveness. The integration of the segmentation model improves results in ex-vivo experiments by mitigating outliers. Specifically, the laser scanner with the segmentation model achieves registration errors below 2.2 mm, with the intercondylar region exhibiting errors below 1 mm. In experiments with phantoms, the errors are always below 1 mm.

The results show the viability of integrating the laser scanner with VBSN as a noninvasive and potential alternative to traditional methods by overcoming surface digitization challenges and expanding the reachable region. Future efforts aim to improve hardware to further optimize performance and applicability in complex procedures.