Automatic evaluation of 3D registration quality in surgical navigation

Surgical navigation systems rely on accurate registration between pre-operative models and intraoperative data. Currently, registration quality is assessed manually after the registration stage, either by digitizing additional anatomical landmarks or by visually inspecting augmented reality. We introduce a framework for automatic, real-time evaluation of 3D registration quality during the registration process, enabling continuous assessment and helping the surgeon decide when to conclude registration while increasing confidence in navigation performance.

The framework encodes 3D points acquired during registration, called 3D trajectories, along with a registration hypothesis into structured heatmaps. A classification model processes these heatmaps to evaluate registration quality. The framework is anatomy-agnostic and generalizes across different anatomies and surgical workflows.

Evaluated in video-based surgical navigation (VBSN) for anterior cruciate ligament (ACL) reconstruction on five ex vivo femur and tibia specimens, which were not used during development. The proposed framework provides a robust evaluation of the registration quality, is anatomy-agnostic, requires on average less than 20% intraoperative data digitization, and achieves performance comparable to a conventional empirical approach.

The framework enables automatic, real-time, continuous assessment of 3D registration quality, supporting the surgeon throughout the registration process. Although demonstrated in femur and tibia arthroscopy, it can be extended to other anatomies and surgical navigation procedures, potentially reducing navigation time and improving workflow efficiency.