A Fast 3-dimensional Full Search Algorithm for Setting Volume of Interests of MR Spectroscopy in Brain Tumor Images

Motivation: No good algorithms are available for searching volume of interests (VOIs) of MRS even if good segmentation methods like neural networks are available. Goal(s): To run a 3-dimensional full search method in practical processing time. Approach: Evaluating all voxels within a VOI is too slow. To overcome this problem, the summed-area table is computed for each set of rotation angles in the 9-dimensional searching parameter space. By using the table, an objective function can be computed with only 8 adding/subtracting operations. Results: By compared with a naïve method, the proposed 3-dimensional full search algorithm accelerated approximate 200 times. Impact: This abstract presents a fast search method using a new fast 3-dimensional full search algorithm for setting volume of interests of MRS using a segmentation network. The proposed algorithm accelerated search approximate 200 times using 3-dimensional summed-area tables.