Purpose: This study aims to develop a data-driven method to derive a patient-specific CBCT to CT HU mapping function and evaluate the method for dose calculation on CBCT in H&N cancer radiotherapy.
Methods: Eleven patients receiving VMAT for oropharyngeal cancer in a dose de-escalation clinical trial were enrolled in this study. Daily CBCT and a mid-course CT for re-assessment were acquired. The study derived a CBCT-to-CT mapping from one pair of CT and its closest CBCT, and applied to CBCT in the other set for independent dose calculation evaluation. The mapping was created by spatially-constraint fuzzy c-mean clustering (FC) on (CBCT, CT) HU vectors in the regions of air, soft tissue and bone that were thresholded on CBCT. To generate pseudo-CT, the probabilities of a CBCT voxel belonging to subtypes were computed by clustering with the CBCT centroids, and a CT value was created from the CT centroids weighting with the probabilities. The methods of using a HU-density curve (HC) from Catphan phantom imaging and air-water-bone density assignment (DA) were also evaluated for CBCT-based dose calculations. Dose differences between the three methods and CT calculations were assessed.
Results: Mean absolute differences of CT numbers in entire CBCT body of the patients are 110, 136 and 147 between FC, HC and DA method generated pseudo-CT and planning CTs, respectively. DVH differences between the three CBCT-based methods and CT calculated doses were within 3% for the PTV and related OARs. The percentage differences in parotid mean doses were 0.5±0.4%, 0.7±0.5%, 0.9±0.4% for FC, HC, and DA methods, respectively.
Conclusion: The fuzzy clustering on planning CT and 1st-day CBCT could generate a patient-specific CBCT-to-CT mapping for accurate dose calculation on patient CBCT in the subsequent fractions. The method may be used for CBCT-based daily dose assessment and adaptive replanning.
Funding Support, Disclosures, and Conflict of Interest: The work was supported partially by NIH grant UG3CA228699
Not Applicable / None Entered.
Not Applicable / None Entered.