Online adaption approaches for intensity modulated proton therapy for head and neck patients based on cone beam CTs and Monte Carlo simulations

Abstract

To develop an online plan adaptation algorithm for intensity modulated proton therapy (IMPT) based on fast Monte Carlo dose calculation and cone beam CT (CBCT) imaging. A cohort of ten head and neck cancer patients with an average of six CBCT scans were studied. To adapt the treatment plan to the new patient geometry, contours were propagated to the CBCTs with a vector field (VF) calculated with deformable image registration between the CT and the CBCTs. Within the adaptive planning algorithm, beamlets were shifted following the VF at their distal falloff and raytraced in the CBCT to adjust their energies, creating a geometrically adapted plan. Four geometric adaptation modes were studied: unconstrained geometric shifts (Free), isocenter shift (Iso), a range shifter (RS), or isocenter shift and range shifter (Iso-RS). After evaluation of the geometrical adaptation, the weights of a selected subset of beamlets were automatically tuned using MC-generated influence matrices to fulfill the original plan requirements. All beamlet calculations were done with a fast Monte Carlo running on a GPU (graphics processing unit). Geometrical adaptation alone only worked with small anatomy changes. The weight-tuned adaptation worked for every fraction, with the Free and Iso modes performing similarly and being superior than the two range shifters modes. The mean V95 and V107 were 99.4 ± 0.9 and 6.4% ± 4.7% in the Free mode with weight tuning. The calculation time per fraction was ~5 min, but further task parallelization could reduce it to ~1–2 min for delivery adaptation right after patient setup. An online adaptation algorithm was developed that significantly improved the treatment quality for inter-fractional geometry changes. Clinical implementation of the algorithm would allow delivery adaptation right before treatment and thus allow planning margin reductions for IMPT.

Dosimetric influence of deformable image registration uncertainties on propagated structures for online daily adaptive proton therapy of lung cancer patients

“…In general Monte Carlo (MC) calculations have smaller dose calculation uncertainties, especially in areas with large density heterogeneities [57][58][59][60]. However, in online DAPT, time is a critical factor and even with the fastest available proton MC dose calculations [61][62][63][64], a full plan optimization takes ~5 minutes, while analytical algorithms reoptimise the plan within a few seconds [27]. Additionally, the analytical algorithm used in this study has been shown to compare favourably with full Monte Carlo (TOPAS) dose calculations, with 97.7% and 91.9% of all voxels agreeing within 5% and 3% for a variety of indications, including lung [65] and the difference between the different dose calculation algorithms were negligible compared to those caused by anatomical changes [22].…”

Daily Adaptive Proton Therapy: Is it Appropriate to Use Analytical Dose Calculations for Plan Adaption?

“…We are aware that a fast plan optimization with MC would be preferable. However, even though recent developments achieved large improvements for the speed of proton MC dose calculations (29,(42)(43)(44), a full plan optimization takes ~5 minutes and is therefore still much slower than analytical approaches, which only take some seconds (30). In an on-line daily adaptive approach the calculation time is a critical factor, therefore for this application an ultra-fast analytical approach is preferable.…”

“…Current adaption protocols are however time and resource consuming. Times required for re-planning using MC optimization, which take 5 minutes or more at best, are incompatible with daily online adaption and a high throughput patient workflow (28,29). Re-planning using an analytical algorithm on the other hand can be extremely time efficient, reducing the complete re-planning process to just a few seconds (30), enabling rapid plan adaption on a daily basis.…”

Quantifying the accuracy of deformable image registration for cone‐beam computed tomography with a physical phantom

“…Users of CBCT should ensure that their CBCT system has adequate image quality for daily image guidance of intensity‐modulated proton therapy (IMPT). Recently, Botas et al developed an online plan adaptation algorithm for IMPT based on fast Monte Carlo dose calculation and CBCT imaging . Those authors concluded that clinical implementation of their developed algorithm would allow adaptation of dose delivery immediately before treatment, thereby allowing planning margins for IMPT to be reduced.…”