Automated Lattice Treatment Planning for Bulky Tumors

Abstract

Spatially-fractionated radiotherapy has potential to improve local control in bulky tumors by achieving both focally high intratumoral doses and acceptable dose fall-off to surrounding normal tissues. However, a standard of care does not exist for the optimal values and distribution of peak and valley doses in treatment plans. To evaluate these questions systematically using dosimetric analyses, the number of alternative treatment plans required exceeds the practical limits of manual planning. In this study, we report an approach for automated planning of volumetric arc-based lattice therapy resulting in over 1100 unique treatment plans. An automation script was generated using Python v3.6 and imported into departmental treatment planning software. A spreadsheet was created with varying combinations of the number of fractions, peak dose, valley dose, and peak-dose sphere size and spacing for 7 patients with bulky tumors (volume range, 159-593 cc). These values were imported into the script after which all planning was automated. For each combination, a bounding box was generated around the gross tumor target volume and divided into equal segments in the x, y, and z planes. Individual spheres were constructed for a given parameter set and those fully encompassed within the target contracted by 2 cm were used to create peak dose targets. The valley dose targets were generated by subtracting an 8 mm expansion of the peak dose spheres from the overall target volume. Multi-criteria optimization (MCO) was then applied to generate a final plan after which the script moved onto the next permutation in the spreadsheet. Following the creation of each plan, the script populated the source spreadsheet with data including doses to the target, fall-off structures, and organs-at-risk. Plans which resulted in an MCO infeasibility error were also noted. A total of 1188 combinations of peak dose, valley dose, and peak-dose sphere size and spacing for 1- and 5-fraction plans were evaluated. Peak and valley doses ranged from 6 to 27 Gy and 0.4 to 20 Gy, respectively, for 1-fraction plans and 30 to 50 Gy and 2 to 20 Gy, respectively, for 5-fraction plans. This resulted in a range of peak/valley ratios from 1.35 to 67.5. Of these combinations, 594 plans were successfully generated and 594 were infeasible, most often due to sphere distribution or peak/valley dose ratios greater than 5. The median time for generation of an individual plan was 21 minutes (range, 15-46). Our script allowed the evaluation of 1188 unique combinations of spatial dose distribution using automated planning. This rapid, yet systematic, assessment of parameters for spatially-fractionated radiation will be used to inform pre-clinical and clinical studies and would not have been feasible with manual planning. Moreover, the median plan generation time of 21 minutes demonstrates potential for wider use of this script in standard clinical settings.