Volume Rendering-Based Patient Registration for Extended Reality

Modern computer-assisted surgery systems require a patient registration functionality to align the preoperative patient data with the patient on the operating room (OR) table. State-of-the-art systems use a two-stage approach where a handful of landmarks known in the preoperative data set are first pointed at on the patient, followed by some surface landmarks where the corresponding points are not known beforehand. The methods employed are, for example, iterative closest point (ICP), to align the digitized surface landmarks on the patient with a surface mesh of the target structure segmented during surgical planning. With the processing power of modern graphics processing units (GPUs), there is a desire to use direct volume-rendered images of a patient’s preoperative data during the planning stage of the intervention. This would greatly simplify the planning, as manual segmentation of anatomical structures would not be needed anymore. However, the current patient registration approaches rely on such segmented structures and can not yet deal with volume rendered data. This paper proposes a two-stage registration method similar to the classical referencing approaches but that uses volume rendered images instead of the common surface meshes. We quantitatively evaluated the proposed concept on a phantom model and could show its applicability even when segmented anatomical structures are not available. This approach shows great potential for reducing the time between image acquisition and operation for preplanned, orthopedic interventions.