Towards real-time plan adaptation for MRI-guided radiotherapy

The introduction of hybrid MRI and linear accelerator (MRI-linac) machines enables the online volumetric imaging during radiation delivery with the superior soft tissue contrast of the diagnostic quality MRI. In this context, conventional radiotherapy workflow will gradually transfer from an offline to an online setting, where in every treatment fraction radiation will be adapted on-the-fly to the changing anatomy of the patient. A new generation of planning systems is therefore needed for online plan adaptation based on the MRI-derived position and motion data. This work formulates the requirements for such a system and proposes a new sequencing methodology to facilitate fast online replanning along with indicative inter- and intrafraction adaptive applications. The newly developed Adaptive Sequencer (ASEQ), is able to incorporate anatomical changes into the IMRT planning process, essentially allowing the successful optimization and delivery of a static clinical prescription on a dynamic patient anatomy. ASEQ is an iterative process which gradually converges to an input prescribed dose. Each iteration produces unique segments which target the latest patient anatomy. By coupling ASEQ to a Segment Weight Optimization (SWO) in a conventional static environment, we show that valid clinical plans can be generated for multiple treatment sites. Furthermore we demonstrate that by omitting SWO, and instead transferring any missing/excess dose in a voxel-by-voxel basis to the next fraction's prescription (Inter-Fraction Scheme (IFS)), the intended dose can be successfully delivered enabling the non-deterministic plan adaptation during treatment. We then demonstrate that ASEQ can be utilized for intrafraction plan adaptation based on 3D anatomical deformations in kidney cases with artificially induced baseline shifts. In both single and multifraction treatments ASEQ converges to the prescribed dose and spares the surrounding structures outside of the target region, by generating segments that target the different instances of the moving patient anatomy. By reimplementing ASEQ to facilitate very fast applications we showcase inter- and intrafraction adaptive treatments for the current and future MRI-guided clinic. We present a daily replanning application for rotational correction in prostate radiotherapy based on implanted fiducial markers. The inclusion of the daily recorded prostate rigid transformations into fast online replanning leads to consistent target coverage and OAR high dose exposure. This in turn enables the reduction of the planning margins and thus further healthy tissue sparing. Finally, we present a single fraction SBRT treatment based on 3D deformations calculated from online MR data for renal tumours. The MRI pipeline includes a pre-beam 4DMRI and multiple sets of orthogonal 2D-cine MR images acquired during the beam-on phase which are processed by a statistical motion model to produce high frequency 3D deformation vector fields along with their corresponding volumes. We simulate fast inter-beam replanning utilizing a novel mid-position update scheme while including the previously delivered dose to the patient, accurately calculated on the online anatomies. We demonstrate that this method is able to account for baseline variations/drifts that might occur during treatment and leads to higher target coverage and dose escalation while greatly decreasing the dose delivered to the surrounding tissue