WE-E-BRB-03: Development of a Deformable Dosimetric Phantom for Verification of 4D Dose Calculation Algorithms

To develop a deformable lung phantom to verify voxel mapping and dose accumulation in 4D dose calculation algorithms used under different scenarios of tissue compression.

The phantom consists primarily of a heterogeneous sponge with an embedded tissue-equivalent tumor. The sponge is wrapped in a latex balloon housed in a Lucite cylinder. The balloon is attached to a piston that compresses the sponge to mimic the human diaphragm. The phantom was programmed to simulate different breathing patterns. Radiochromic films and TLD were embedded in the sponge for 4D dosimetry algorithm verification. 37 anatomical landmarks were manually tracked to verify voxel mappings for four deformable image registration (DIR) algorithms: in-house developed Demons and finite element model algorithms, and two B-Spline based Velocity AI registration algorithms performed between end-inhale and end-exhale. A 6MV photon beam was simulated with BEAMnrc/DOSXYZnrc on the end-inhale image with the dose mapped to the end-exhale using voxel-based linear dose mapping (LDM) and particle-based energy-mass congruence mapping (EMCM) methods.

The mean density of the artificial lung was increased by 10.2% as the sponge was compressed by 2.5cm. The reproducibility of the phantom deformation was within image resolution (1×1×3 mm3), and the accuracy of four DIR registrations of the extreme phases was within 3.0mm. With the same registration displacement vector field (DVF), EMCM and LDM had different doses mapped to the end- exhale image. Their difference at the center of a beam was up to 8.3% for a Demons DVF and 5.8% for a Velocity DVF. The maximum difference between EMCM and LDM was 13.2% at beam penumbra.

The developed deformable dosimetric phantom readily demonstrated variations among different dose addition and image registration algorithms, and is appropriate to serve as a QA tool for verification of 4D dose calculation algorithms. The research was supported by NIH/NCI R01CA140341.