Markerless inside-out tool tracking for endoscopic spine surgery: a benchmarking study and clinical dataset

Area of Science:

• Minimally Invasive Surgery
• Surgical Navigation
• Computer Vision

Background:

• Existing surgical navigation systems face limitations due to line-of-sight issues with outside-in optical tracking.
• Markerless inside-out tracking, utilizing tool-mounted cameras, presents a potential alternative for enhanced surgical guidance.

Purpose of the Study:

• To benchmark state-of-the-art inside-out visual Simultaneous Localization and Mapping (vSLAM) methods for surgical instrument tracking.
• To evaluate the performance of vSLAM algorithms in the context of endoscopic spine surgery.

Main Methods:

• Collected a novel dataset from simulated endoscopic spine surgeries in an operating room setting.
• Recorded synchronized stereo images from tool-mounted cameras, ground truth pose data, and endoscopic video feeds.
• Compared the instrument tracking accuracy of selected vSLAM algorithms using the collected dataset.

Main Results:

• The top-performing vSLAM approach achieved a root mean squared absolute trajectory error of 2.0 mm and 1.47 degrees.
• High accuracy, around 1 mm and 1 degree, was reached in specific sequences.
• Performance degraded significantly when encountering occlusions and dynamic scene elements.

Conclusions:

• Markerless inside-out tracking with vSLAM shows high accuracy, suggesting feasibility for navigated endoscopic spine procedures.
• Current vSLAM algorithms require further development to achieve the robustness needed for routine clinical application.
• The study provides a foundational dataset and benchmark for advancing real-time inside-out navigation in minimally invasive surgery.