WE‐D‐T‐6E‐06: Implementation of a Fast Incremental Algorithm for Voxel Radiological Path Determination

Abstract

Purpose: Performing adaptive image‐guided IMRT will require rapid computation of dose for frequent IMRT fluence map optimization (FMO). Patients should not have to tolerate waiting more than a few minutes for a new IMRT treatment plan. Most beamlet dose models in IMRT FMO expend considerable time performing voxel based ray‐tracing computations. Improvements in voxel based ray‐tracing algorithms have not been presented in the radiation therapy literature since the seminal work of Siddon in the 1980s. We present an algorithm that significantly outperforms the voxel ray‐tracing algorithm of Siddon when applied to density scaled radiological path length calculations in patient voxel space for beamlet dose computation. Method and Materials: An incremental voxel ray tracing algorithm that simultaneously computes voxel indices and ray‐voxel intersections was developed. The majority of the computational time required by the standard Siddon algorithm was found to be in the concatenation of the intersection parameters (α's) into a single unique set and the conversion of the calculations of the voxel indices from floating point to integer values. This algorithm was compared, as components of our in‐house IMRT planning system, to the standard Siddon technique. The test case had 1.7 million total voxels on a 2.5 mm isotropic dose grid and the 1,466 beamlets in the IMRT plan intersected an average of ∼15,000 voxels per beamlet. Results: Use of the new algorithm for the test case demonstrated a 2.7 fold increase in computational speed on average over the Siddon algorithm. Conclusion: The new algorithm provides a significant improvement in voxel ray‐tracing performance over the standard Siddon algorithm. However, parallel computation is still currently required to perform the computation of a typical IMRT treatment planning rapidly enough for adaptive IMRT FMO.