Model Generation for Knee Surgery Simulation Mitsubishi Electric Research Laboratory Knee Surgery Simulation In arthroscopic knee surgery, the surgeon uses a scope and a probe to explore and repair the knee (left). The view from the scope is shown on a monitor. The simulator uses a Phantom force-feedback device to feel a virtual 3D model of the knee (right). Building Realistic 3D Models To make the simulation realistic, accurate models of the anatomical structures must be built from actual Magnetic Resonance data. Since the resolution of the MR slices is limited, terracing artifacts result in the 3D models. An axial scan (far left) results terraces in the axial direction (middle left), and a sagittal scan (far right) results in terraces in the sagittal direction (middle right). Simply smoothing these model will account for a loss of important details, so instead the scans are combined to form one model that has high resolution in all directions. Registration of Images To combine information from both directionalities, the two MR scans must be aligned or registered into the same coordinate system. As the knee may been bent differently in the two scans, one structure is registered at a time. In the image to the right, the femur is aligned in both scans. Combining the Images Both scans are segmented and the surface points are superimposed (left). The axial scan is in blue and the sagittal is in red. A structure called a SurfaceNet is created for each model and then the two nets are relaxed to remove the terraces, while staying faithful to the original data. Building More Accurate Models The result of the Dual-SurfaceNet is shown as a 3D surface model on the right. Notice that all the terracing artifacts are gone, yet the model still contains the details present the in scans. If the original models were simply smoothed out to remove the terraces, much of the important details would have been lost. Patent "Surface model generation for visualizing three-dimensional objects using multiple elastic surface nets."         Sarah Gibson and Michael Leventon.         Patent # US-6,362,821