Automatic online adaptive radiation therapy techniques for targets with significant shape change: a feasibility study

This work looks at the feasibility of an online adaptive radiation therapy concept that would detect the daily position and shape of the patient, and would then correct the daily treatment to account for any changes compared with planning position. In particular, it looks at the possibility of developing algorithms to correct for large complicated shape change. For co-planar beams, the dose in an axial plane is approximately associated with the positions of a single multi-leaf collimator (MLC) pair. We start with a primary plan, and automatically generate several secondary plans with gantry angles offset by regular increments. MLC sequences for each plan are calculated keeping monitor units (MUs) and number of segments constant for a given beam (fluences are different). Bulk registration (3D) of planning and daily CT images gives global shifts. Slice-by-slice (2D) registration gives local shifts and rotations about the longitudinal axis for each axial slice. The daily MLC sequence is then created for each axial slice/MLC leaf pair combination, by taking the MLC positions from the pre-calculated plan with the nearest rotation, and shifting using a beam's-eye-view calculation to account for local linear shifts. A planning study was carried out using two head and neck region MR images of a healthy volunteer which were contoured to simulate a base-of-tongue treatment: one with the head straight (used to simulate the planning image) and the other with the head tilted to the left (the daily image). Head and neck treatment was chosen to evaluate this technique because of its challenging nature, with varying internal and external contours, and multiple degrees of freedom. Shape change was significant: on a slice-by-slice basis, local rotations in the daily image varied from 2 to 31 degrees, and local shifts ranged from -0.2 to 0.5 cm and -0.4 to 0.0 cm in right-left and posterior-anterior directions, respectively. The adapted treatment gave reasonable target coverage (100%, 90% and 80% of the base-of-tongue, left nodes and right nodes, respectively, receiving the daily prescription dose), and kept the daily cord dose below the limit used in the original plan (65%, equivalent to 46 Gy over 35 fractions). Most of the loss of coverage was due to one shoulder being raised more superior relative to the other shoulder compared with the plan. This type of skew-like motion is not accounted for by the proposed ART technique. In conclusion, this technique has potential to correct for fairly extreme daily changes in patient setup, but some control of the daily position would still be necessary. Importantly, it was possible to combine treatments from different plans (MLC sequences) to correct for position and shape change.