Abstract
Image-guided spine surgery requires registration between the patient anatomy and the preoperative computed tomography (CT) image. We have previously developed an accurate and robust registration technique for this application by using intraoperative ultrasound to acquire patient anatomy and then registering the ultrasound images to the CT images by aligning the posterior vertebral surfaces extracted from both modalities. In this study, we validate our registration technique across 18 vertebrae on three porcine cadavers.
We applied the ultrasound-registration technique on the thoracic and lumbar vertebrae of the porcine cadavers using both single sweeps and double orthogonal sweeps. For each sweep pattern at each vertebra, we also randomly simulated 100 different initial misalignments and registered each misalignment. The resulting registration transformations are compared to gold standard registrations to assess the accuracy and the robustness of the technique.
Orthogonal-sweep acquisition was found to be the sweep-pattern that performed the best and yielded a registration accuracy of 1.65 mm across all vertebrae on all porcine cadavers. It was found that the target registration errors (TRE) stay relatively constant with increasing initial misalignment and that the majority (82.7%) of the registrations resulted in TREs below the clinically recommended 2 mm threshold. In addition, it was found that the registration accuracy varies by the sweep pattern and the vertebral level, but neighbouring vertebrae tend to result in statistically similar accuracy.
We found that our ultrasound-CT registration technique yields clinically acceptable accuracy and robustness on multiple vertebrae across multiple porcine cadavers.
