Room: Exhibit Hall
Purpose: The purpose of this study was to develop and validate an efficient strategy for generating virtual computed tomography (CTv) for the assessment of adaptive radiotherapy in head and neck (HN) treatment.
Methods: We retrospectively analyzed CT images taken from 15 patients treated with conventional radiotherapy for HN malignancies. The right and left parotid glands, spinal cord, and the gross tumor volume (GTV) were considered in our analysis. We generated 15 CTv using a deformable image registration (DIR) of initial simulation CT (CTp1) and cone beam CT, acquired for patient setup. We validated our approach by considering the real replanning CT (CTp2) as ground truth. We computed the Dice similarity coefficient (DSC), true segmentation coefficient (TSC), and sensitivity (SENS) between the structures on CTp2 and CTv. Furthermore, we analyzed the relationship between the accuracy of DIR and the transformation magnitude factor (TF) which is provided by the DIR software (Velocity AI).
Results: The mean DSC, TSC, and SENS were approximately 0.7 or greater for all structures, which is similar to other reports. However, the lowest values for DSC, TSC, and SENS were 0.236, 0.250, and 0.235 for GTV, respectively. The Pearson`s correlation coefficient factors of the DSC, TSC, and SENS as a function of TF for all structures were over 0.3. Notably, the factors for DSC were over 0.6 and showed a strong correlation with TF. This is because the DIR software did not support the greater image transformation when we need the part of the images requiring large changes.
Conclusion: We clarified that there is a correlation between TF and the accuracy of DIR and showed the potential of CTv using DIR in assessing HN treatment.