Data-driven approach to generating achievable dose-volume histogram objectives in intensity-modulated radiotherapy planning

Abstract

Purpose: To propose a method of intensity-modulated radiotherapy (IMRT) planning that generates achievable dose-volume histogram (DVH) objectives using a database containing geometric and dosimetric information of previous patients.

Methods and materials: The overlap volume histogram (OVH) is used to compare the spatial relationships between the organs at risk and targets of a new patient with those of previous patients in a database. From the OVH analysis, the DVH objectives of the new patient were generated from the database and used as the initial planning goals. In a retrospective OVH-assisted planning demonstration, 15 patients were randomly selected from a database containing clinical plans (CPs) of 91 previous head-and-neck patients treated by a three-level IMRT-simultaneous integrated boost technique. OVH-assisted plans (OPs) were planned in a leave-one-out manner by a planner who had no knowledge of CPs. Thus, DVH objectives of an OP were generated from a subdatabase containing the information of the other 90 patients. Those DVH objectives were then used as the initial planning goals in IMRT optimization. Planning efficiency was evaluated by the number of clicks of the "Start Optimization" button in the course of planning. Although the Pinnacle(3) treatment planning system allows planners to interactively adjust the DVH parameters during optimization, planners in our institution have never used this function in planning.

Results: The average clicks required for completing the CP and OP was 27.6 and 1.9, respectively (p <.00001); three OPs were finished within a single click. Ten more patient's cord + 4 mm reached the sparing goal D(0.1cc) <44 Gy (p <.0001), where D(0.1cc) represents the dose corresponding to 0.1 cc. For planning target volume uniformity, conformity, and other organ at risk sparing, the OPs were at least comparable with the CPs. Additionally, the averages of D(0.1cc) to the cord + 4 mm decreased by 6.9 Gy (p <.0001); averages of D(0.1cc) to the brainstem decreased by 7.7 Gy (p <.005). The averages of V(30 Gy) to the contralateral parotid decreased by 8.7% (p <.0001), where V(30 Gy) represents the percentage volume corresponding to 30 Gy.

Conclusion: The method heralds the possibility of automated IMRT planning.