Interactive initialization of 2D/3D rigid registration.

Area of Science:

• Medical Imaging
• Computer-Assisted Surgery
• Image Registration

Background:

• 2D/3D registration is vital for image-guided navigation systems.
• Clinical adoption of existing registration algorithms is limited by the need for accurate initial transformations (TRE < 10 mm).

Purpose of the Study:

• To present two interactive initialization approaches for 2D/3D registration in the operating room (OR).
• To achieve the necessary accuracy for x-ray/MR and x-ray/CT registration, facilitating clinical integration.

Main Methods:

• Developed two interactive methods: gesture-based alignment and augmented reality (AR) using a tracked tool.
• Preoperative images (MR/CT) are visually aligned to intraoperative x-rays using volume rendering and either gestures or a tracked tool.
• Methods evaluated using three public 2D/3D registration datasets.

Main Results:

• Gesture-based x-ray/MR registration: mean TRE of 9.3 ± 5.0 mm (146.3 ± 73.0 s).
• AR-based x-ray/MR registration: mean TRE of 7.2 ± 3.2 mm (44 ± 32 s).
• Gesture-based x-ray/CT registration: mean TRE of 7.4 ± 5.0 mm (132.1 ± 66.4 s).
• AR-based x-ray/CT registration: mean TRE of 8.3 ± 5.0 mm (58 ± 52 s).

Conclusions:

• Both gesture-based and AR-based registration methods provide sufficient accuracy for initializing 2D/3D registration in the OR.
• These approaches are suitable for use with or without an existing tracking system.
• The methods address the critical need for accurate initial transformations in image-guided navigation.