Intra-fraction anatomical variations can compromise treatment quality, and while real-time adaptation of proton therapy plans during delivery offers a potential solution, its implementation remains challenging, particularly when integrating accurate Monte Carlo dose engines into a sufficiently fast optimization workflow. Conventional optimization methods rely on pre-computed dose influence matrices for each spot (i.e., beamlets), which are time-consuming to calculate and limit the feasibility of real-time adaptation. This study introduces an intra-fraction adaptive framework based on a beamlet-free optimizer that dynamically updates the treatment plan during treatment delivery. Approach: To eliminate the need for precomputed beamlets, the beamlet-free optimizer integrates the Monte Carlo dose calculation with the optimization process. The proposed adaptive method accounts for delivered dose and anatomical changes and allows for re-delivery of spots when necessary. The optimizer leverages prior plan information from an initial optimization and integrates a plan conformity term to balance plan stability and adaptability. The framework was tested on synthetic phantoms under various motion and deformation scenarios and on three clinical cases (two lung, one cervix) and compared against (1) the initial plan on the planning CT, (2) the initial plan delivered on the evolving anatomy without adaptation, and (3) plans adapted using a conventional beamlet-based optimizer. Main results: Across all test cases, non-adapted plans failed to meet clinical objectives, while both adaptive approaches restored target coverage and organ-at-risk sparing. The beamlet-free optimizer achieved comparable dosimetric quality to the beamlet-based method while reducing computation times by up to 7.7×. Time gains increased with plan complexity and total number of spots. Significance: The proposed beamlet-free adaptive workflow enables intra-fraction updates during treatment delivery with Monte Carlo-level accuracy, allowing flexible spot re-delivery while maintaining dosimetric accuracy.
Dormal et al. (Wed,) studied this question.