C11152: Method for Localizing Radiation Implants from Limited X-ray Data
Value Proposition:
Johns Hopkins University inventors consider the 3-D implant localization as a seed-matching problem and a combinatorial optimization algorithm was formulated. It resolves the correspondence of identified seeds in every image even when a significant number of seeds overlap one another on one or more observed images. In order to solve the optimization problem within near real time, a dimensionality reduction is performed with pruning algorithm that enables a systematic rejection of infeasible combinations without computing the costs for all feasible combinations, thus achieving an efficient cost computation and significant dimensionality reduction. Inaccuracy in the image pose and patient motion are adjusted by using the reconstructed seeds as a fiducial in an iterative way.
Technical Details:
Johns Hopkins University inventors report a matching algorithm for a localization of radiation implants from limited x-ray data. The localization problem is formulated as a combinatorial optimization problem, and the solution is found in seconds using dimensionality reduction to create a smaller space of possible solutions. Dimensionality reduction is possible since the optimal solution has approximately zero cost when the poses of the acquired images are known to within a small error. This method is also formulated to address the object overlap problem in which objects overlap on one or more observed images and cannot be explicitly identified. A pruning algorithm is proposed to avoid unnecessary computation of cost metrics and the reduced problem is solved using linear programming.
Looking for Partners:
The methods described are applicable to any 3-D reconstruction and localization problem where a significant number of small objects must be reconstructed from a limited number of x-ray projections. It will be most useful in scenarios where the number of target objects are large (~100), the number of images are extremely small (e.g., 3 images), image acquisition angle is very narrow, and the computation time should be minimal.
