SU‐F‐J‐109: Generate Synthetic CT From Cone Beam CT for CBCT‐Based Dose Calculation

Abstract

Purpose: The use of CBCT for dose calculation is limited by its HU inaccuracy from increased scatter. This study presents a method to generate synthetic CT images from CBCT data by a probabilistic classification that may be robust to CBCT noise. The feasibility of using the synthetic CT for dose calculation is evaluated in IMRT for unilateral H&N cancer. Methods: In the training phase, a fuzzy c‐means classification was performed on HU vectors (CBCT, CT) of planning CT and registered day‐1 CBCT image pair. Using the resulting centroid CBCT and CT values for five classified “tissue” types, a synthetic CT for a daily CBCT was created by classifying each CBCT voxel to obtain its probability belonging to each tissue class, then assigning a CT HU with a probability‐weighted summation of the classes’ CT centroids. Two synthetic CTs from a CBCT were generated: s‐CT using the centroids from classification of individual patient CBCT/CT data; s2‐CT using the same centroids for all patients to investigate the applicability of group‐based centroids. IMRT dose calculations for five patients were performed on the synthetic CTs and compared with CT‐planning doses by dose‐volume statistics. Results: DVH curves of PTVs and critical organs calculated on s‐CT and s2‐CT agree with those from planning‐CT within 3%, while doses calculated with heterogeneity off or on raw CBCT show DVH differences up to 15%. The differences in PTV D95% and spinal cord max are 0.6±0.6% and 0.6±0.3% for s‐CT, and 1.6±1.7% and 1.9±1.7% for s2‐CT. Gamma analysis (2%/2mm) shows 97.5±1.6% and 97.6±1.6% pass rates for using s‐CTs and s2‐CTs compared with CT‐based doses, respectively. Conclusion: CBCT‐synthesized CTs using individual or group‐based centroids resulted in dose calculations that are comparable to CT‐planning dose for unilateral H&N cancer. The method may provide a tool for accurate dose calculation based on daily CBCT.