Evaluating a method of treatment plan selection from daily plans on the MR-Linac

Abstract

Purpose or Objective Adaptive radiotherapy using the MR-Linac has the potential to greatly improve the quality of patient treatments. Plans can be adapted based on a daily scan that account for changes in position using an adapt-to-position (ATP) workflow, or adapted to both anatomy and positioning using an adapt-to-shape (ATS) workflow. A full optimisation is currently used to give the best plan for the anatomy that day. However, the mean time taken for optimisation alone has been approximately 14 minutes. As a library of adapted plans is built up for each day that a patient is treated, it is possible that on a given day a suitable plan already exists. This work investigates a method for automatically selecting a plan from this dynamic library to improve the workflow efficiency. This could improve the patient experience, and reduce uncertainties due to anatomical change between scanning and delivering treatment. Materials and Methods Evaluation of using library plans based on image similarity has been carried out for 4 prostate and 2 cervix patients at the 10th fraction. Image similarity of daily scans with the previous 9 treatment images was assessed using ADMIRE (Elekta). This performed a rigid registration and extracted similarity measures in a region of interest including the planning target volume (PTV) with an additional 3 cm margin. ATS plans from previous fractions (library plans) were copied on to the 10th fraction using an adapt segments method which shifts segments based on rigid registration between the two scans. An ATS plan optimised from fluence was used as a gold standard against which library plans were compared. Gamma analysis (γdose , 3%/3mm) of library plans against the gold standard was performed. This measure was used to test for correlation with similarity to determine whether a given metric may be suitable. Finally, dose statistics were compared for the resulting library plans. Results For the two available cervix patients, a positive correlation was found between gamma pass rates and image similarity using mutual information in the PTV and surrounding 3cm where r=0.68 and r=0.71 respectively (see Figure 1). Across all 6 patients, 69% of library plans gave a closer agreement than an ATP plan (72% for cervix). In prostate patients correlation with image similarity was not seen. This is thought to be due to lower inter-fraction anatomy change in these patients. Higher similarity library plans used for the cervix patients often gave dose statistics in closer agreement with the reference plan. However, there was variation within this as shown. Conclusion For patients with cervix cancer image similarity could provide a method for automatic selection from library plans accumulated on the MR-Linac. This saves a significant amount of time between scanning and delivering treatment. However, correlation of plan quality with similarity may be improved by refinement of methods. This work will be validated on a larger cohort as part of ongoing work.