Abstract
Introduction
Obtaining accurate anatomical landmarks may lead to a better morphologic understanding, but this is challenging due to the variation of bony geometries. A manual approach, non-ideal for surgeons or engineers, requires a CT or MRI scan, and landmarks must be chosen based on the 3D representation of the scanned data. Ideally, anatomical landmarking is achieved using either a statistical shape model or template matching. Statistical modeling approaches require multitude of training data to capture population variation. Prediction of anatomical landmarks through template matching techniques has also been extensively investigated. These techniques are based on the minimization or maximization of an objective or cost function. As is the nature of non-rigid algorithms, these techniques can fail in the local maxima if the template and new bone models have noise or outliers. Therefore, a combination of rigid and non-rigid registration techniques is needed, in order to obtain accurate anatomical landmarks and improve the prediction process.
Objective
The objective of this study was to find a way to efficiently obtain accurate anatomical landmarks based on an existing template's landmarks for use in a forward solution model (FSM) to predict patient specific mechanics.
Methods
Initially, the 3D meshes for a template bone and new bone of question are imported into the FSM. Landmarks on the template are also loaded with imported data. Then, the template and new bones are located at arbitrary positions within the global coordinate system. If determined to be placed at significantly different positions, the user will re-align the bones to ensure that they are close enough for the process to commence. After initially aligning the bones, the new bone model will appear closer to the template. The template bone model is then registered to the new model using Iterative Closest Point (ICP) with scaling to find the initial regions of correspondence. For each anatomical landmark on the template, initial corresponding landmarks on the new bone are defined as being its closest point. To refine landmarks on the new bone, local corresponding regions are determined between the template and new bone models. Local corresponding regions on the template and new bone models are then registered again using ICP with a scaling algorithm to refine the landmark locations on the new model as seen in Figure 1.
Results
Regardless of differences in size, geometry, and initial position, the algorithm has proven to be successful in transferring landmarks from the template bone to the new bone model (Figure 2). The results also revealed that predicted landmarks on the new bone (purple) are properly defined with respect to the landmarks on the template bone (green) (Figure 2). This process allows for the FSM to be parametric in nature for patient specific analyses.
Discussion and Conclusion
The FSM successfully transferred anatomical landmarks from a template to a new bone model. It has also been proven to work on more than just the femur and pelvis. Future investigations using this process for surgical planning/implant sizing will be used for both our hip and knee FSMs.
For any figures or tables, please contact authors directly.
