Methods Validation with Simulated Data 1.Generate random linear objects in the model coordinate system. 2.Generate a random set of points on each linear object. 3.Apply a random (ground-truth) transformation to all points and add noise. Validation for Tool-to-Model Registration Linear Object Registration Algorithm Given a set of linear objects collected in the sensor coordinate system and a set of linear objects defined in the interventional tool coordinate system: Conclusions References & Acknowledgements Results Introduction Validation with Simulated Data Validation for Tool-to-Model Registration Validation for Volume-to-Model Registration The proposed registration algorithm is sufficiently accurate for practical registration of interventional phantoms and tools without fiducial points. The algorithm is implemented as a SlicerIGT ( module for 3D Slicer ( Future work involves refining the matching step, further automating the algorithm, and additional image registration experiments. [1] Anand, M., King, F., Ungi, T., Lasso, A., et al. (2014) "Design and Development of a Mobile Image Overlay System for Image-Guided Needle Interventions," 36th Annual International Conference of IEEE EMBS. [2] Carbajal, G., Lasso, A., Gomez, A., & Fichtinger, G. (2013). Improving N-wire phantom-based freehand ultrasound calibration. Int J Comput Assist Radiol Surg., 8, [3] Fitzpatrick, J. M., West, J. B., & Maurer, C. R. (1998). Predicting Error in Rigid-Body Registration. IEEE Transactions on Medical Imaging, 17, [4] Ungi, T., Sargent, D., Moult, E., Lasso, A., et al. (2012). Perk Tutor: An open-source training platform for ultrasound-guided needle insertions. IEEE Transactions on Biomedical Engineering, 59 (12), [5] Walsh, R., Soehl, M., Rankin, A., Lasso, A., et al. (2014). Design of a tracked ultrasound calibration phantom made of LEGO bricks. SPIE Medical Imaging. Matthew S. Holden is supported by the Ontario Graduate Scholarship. Gabor Fichtinger is supported as a Cancer Care Ontario Research Chair in Cancer Imaging. Linear Object Registration of Interventional Tools Matthew S. Holden, Gabor Fichtinger Laboratory for Percutaneous Surgery, School of Computing, Queen’s University, Kingston, Canada Motivation Registration allows clinicians to view tracked physical objects and images (from multiple modalities) in a common navigation space. Most applications use point-set registration algorithms; however, landmark points are not present on all interventional tools. Objective The objective is to develop a practical registration method for when landmark points are not available. We propose a method which uses points, lines, and planes (linear objects). The method should guarantee convergence to the optimal solution, even when correspondence is unknown. Table 1. Mean rotational and translational error for the fCal [2], lumbar spine [4], and gel block targeting tool [1] phantoms compared to the ground-truth point-set registration. Fig. 1. Photograph of user collecting points on the a) fCal phantom [2], b) lumbar spine phantom [4], c) LEGO® brick phantom [5], d) gel block targeting tool [1]. References indicated in green. Fig. 3. Plot of error in calculated registration as a function of noise for simulated data. Blue line indicates rotational error; red line indicates translational error. Four navigation phantoms were used for validating tool-to-model registration (Fig. 1). Because the LEGO® brick phantom does not have fiducial points, target registration error (TRE) [3] and point reconstruction accuracy (PRA) [2] were used to evaluate the registrations. Validation for Volume-to-Model Registration Validation for volume-to-model registration was performed using a reconstructed ultrasound volume of the Targeting Tutor phantom [4] (Fig. 2). Because the phantom had no fiducials, target registration error (TRE) [3] was used to evaluate the registrations. Fig. 2. Photographs of Targeting Tutor phantom [4] a) user ultrasound scanning, b) model of phantom, and c) ultrasound volume reconstruction. References indicated in green. PhantomRotational Error (°)Translational Error (mm) fCal Phantom Lumbar Spine Phantom Gel Block Tool AlgorithmMean TRE (mm)Mean PRA (mm) Point-Set Registration Linear Object Registration AlgorithmMean TRE (mm) Point-Set Registration4.05 ± 1.57 Linear Object Registration3.87 ± 0.91 Table 2. Target registration error (TRE) and point reconstruction accuracy (PRA) for point-set registration and linear object registration for the LEGO® brick phantom [5]. Table 3. Target registration error (TRE) for point-set registration and linear object registration for the Targeting Tutor phantom [4]. 1.Map collected points to linear objects via principal component analysis. 2.Match linear objects in the two coordinate frames using distances to a set of reference points. 3.Calculate the linear object centroid in each coordinate frame. 4.Perform point-set registration with known correspondence using centroid projections and direction/normal vectors. 5.Iteratively adjust the translation and rotation: a)Find the closest point on linear objects to collected points. b)Calculate the translational difference between point-sets. c)Calculate spherical point-set registration.