Evaluating the Need for Daily Image Guidance in Head and Neck Cancers Treated with Helical Tomotherapy: A Retrospective Analysis of a Large Number of Daily Imaging-based Corrections

Abstract Aims Clinical implementation of image-guided intensity-modulated radiotherapy is rapidly evolving. Helical tomotherapy treatment delivery involves daily imaging before intensity-modulated radiotherapy delivery. This can be a time consuming resource-intensive process, which may not be essential in head and neck radiotherapy, where effective immobilisation is possible. This study aimed to evaluate whether an offline protocol implementing the shifts derived from the first few fractions can be an acceptable alternative to daily imaging for helical tomotherapy. Materials and methods We retrospectively analysed the set-up data of 2858 fractions of 100 head and neck cancer patients who were treated with daily online image guidance. Using summary data from all treatment fractions, we calculated the systematic error (∑) and random error (σ) in each of the three axes, i.e. mediolateral ( x ), craniocaudal ( y ), anteroposterior ( z ). We also calculated the translational vector of each fraction of individual patients. We then simulated two no-action-level offline protocols where set-up errors of the first three (protocol F3) or five fractions (protocol F5) were averaged and implemented for the remaining fractions. The residual errors in each axis for these fractions were determined together with the residual ∑ and σ. Planning target volume (PTV) margins using the van Herk formula were generated based on the uncorrected errors as well as for the F3 and F5 protocols. For each scenario, we tabulated the number of fractions where the residual errors were more than 5 mm (our default PTV margin). We also tried to evaluate whether errors tended to differ based on intent (radical or adjuvant), anatomical subsite or weight loss during treatment. Results Analysis from this large dataset revealed that in the tomotherapy platform, the highest set-up errors were in the anteroposterior ( z ) axis. The global mean was 5.4 mm posterior shift, which can be partly attributed to couch sag on this system. Uncorrected set-up errors resulted in systematic and random errors of ∑ x,y,z of 1.8, 1.7 and 2 mm and σ x,y,z of 1.7, 1.5 and 1.9 mm, with a required PTV margin in x , y , z axes of 5.7, 5.3 and 6.2 mm. Implementing average shifts from the first three or five fractions resulted in a substantial reduction in the residual systematic errors, whereas random errors remained constant. The PTV margins required for the residual errors after three and five fraction corrections were 3