Evaluation of pattern based point clouds for patient registration — A phantom study

The image to patient registration process is the longest non operative process on navigated surgery. Markerless registration using laser scan surface registration techniques comes with inaccuracies up 9mm due to user errors and to durations of up to over 15 minutes. In this study different patterns were examined in their influence of registration accuracy. On a patient´s head phantom, point clouds were recorded. For each pattern 15 point clouds were examined by matching them with the computer tomographic data of the phantom. The matching was each performed with three different versions of the Iterative Closest Point Algorithm. As a measurement the Root Mean Square Error was considered. For each point cloud the number of considered points was reduced step by step from 800 down to 12. A pattern without minimal effects due to face shift could be found with a registration error below 1.5mm by considering twelve points.