Objectives Previously we showed that organ non-rigid deformation is important for targeted radionuclide therapy (TRT) dosimetry. This simulation study aims to improve TRT dosimetry by non-rigid registration for sequential quantitative SPECT images.

Methods We modeled 3 anatomical variations each with 3 In-111 Zevalin activity distributions, i.e., a total of 9 phantoms, at 1, 12, 24, 72 and 144 hrs post-injection using the digital 4D Extended Cardiac Torso (XCAT) phantom. Local deformation of the liver, kidneys, spleen and stomach were modeled of <5 pixels/degrees with <5% volume change. Whole body rigid motion of <5 pixels/degrees of translation/rotation was modeled. Realistic noisy projections were generated by an analytical projector and then reconstructed using OS-EM with 128 updates including attenuation, scatter and collimator-detector response corrections. Individual organs at 5 time points were segmented semi-automatically or using the exact volume-of-interest (VOI) maps from the phantom. Segmented organs were registered to the 24-hr time point by affine+B-spline registration. Voxel by voxel integration was done on registered images over 5 time points, followed by convolution with a Y-90 dose kernel to generate dose images. Organ dose, dose volume histograms (DVHs) and mean square error (MSE) were calculated for DVHs.

Results After non-rigid image registration with semi-automatic VOI segmentation, the mean difference in organ dose was reduced from -17.3±5.98% to 1.55±2.35% and -11.3±2.67% to -1.79±2.56% for spleen and liver respectively. For all organs, MSE improved for 19.8-93% and DVHs showed substantial improvement after registration.

Conclusions Non-rigid registration improved dose estimation for different organs. It substantially improved DVH estimates especially for spleen and liver without the aid of CT, and is important for reliable 3D dosimetry.