SU-E-J-135: 3-D Fourier-Based Volumetric Registration for Estimating Intra-Fractional Lung Tumor Motion.

Abstract

To develop a three-dimensional (3-D) volumetric registration algorithm to estimate the intra-fractional lung tumor motion between respiratory phases for improving the accuracy of radiotherapy treatment. The 3-D thoracic CT volumes (512×512×160 voxels, with dimensions 0.97×0.97×2.5 mm3 ) in different respiratory phases were acquired on a General Electric Optima T580 scanner in cine mode. As a preprocess, a bicubic interpolation was used to interpolate the original 3-D volumes along the cephalo-caudal axis to volumes of size 512×512×400 voxels, with dimensions 0.97×0.97×1 mm3 . In each respiratory phase, a sub-volume covering the tumor was roughly specified manually. A 3-D phase correlation of two sub-volumes was computed by using the 3-D inverse Fourier transformation of the normalized cross power spectrum of two sub-volumes. The 3-D displacements along three axes were estimated by finding the location of the highest peak in the 3-D phase correlation. Experiments were conducted on an artificial 4-D CT data set and three clinical 4-D CT data sets. Experimental results shown that the proposed algorithm was capable of estimating the tumor motion between respiratory phases with a high-accuracy (mean square error <1 mm). This work extended the conventional image registration techniques from 2-D to 3-D for tumor motion estimation. This work indicates a potential for significant accuracy improvement in radiotherapy treatment planning. The high-accurate 3-D tumor motion information provides a reliable basis for expanding a clinical target volume (CTV) to a planning target volume (PTV) to incorporate the intra-fractional tumor motion.