Effect of marker position and size on the registration accuracy of HoloLens in a non-clinical setting with implications for high-precision surgical tasks

Purpose: Emerging holographic headsets can be used to register patient-specific virtual models obtained from medical scans with the patient’s body. Maximising accuracy of the virtual models’ inclination angle and position (ideally, ≤2° and ≤2 mm, respectively, as in currently approved navigation systems) is vital for this application to be useful. This study investigated the accuracy with which a holographic headset registers virtual models with real-world features based on the position and size of image markers.
Methods: HoloLens® and the image-pattern-recognition tool Vuforia Engine™ were used to overlay a 5-cm-radius virtual hexagon on a monitor’s surface in a predefined position. The headset’s camera detection of an image marker (displayed on the monitor) triggered the rendering of the virtual hexagon on the headset’s lenses. 4x4, 8x8 and 12x12cm image markers displayed at nine different positions were used. In total, the position and dimensions of 114 virtual hexagons were measured on photos captured by the headset’s camera.
Results: Some image marker positions and the smallest image marker (4x4 cm) led to larger errors in the perceived dimensions of the virtual models than other image marker positions and larger markers (8x8 and 12x12 cm). ≤2° and ≤2 mm errors were found in 70.7% and 76% of cases, respectively.
Conclusion: Errors obtained in a non-negligible percentage of cases are not acceptable for certain surgical tasks (e.g. the identification of correct trajectories of surgical instruments). Achieving sufficient accuracy with image marker sizes that meet surgical needs and regardless of image marker position remains a challenge