Automatic PTV margin determination for midposition-based lung SBRT planning on the Unity MR-linac

Purpose or Objective
For midposition (midpos) based dose planning, a patient-specific PTV margin is typically determined by measuring the peak-to-peak (p2p) motion of the GTV during the breathing cycle. Our goal is to develop a new method to determine the PTV margin automatically for lung SBRT midpos-planning on the Unity MR-linac (Elekta AB, Stockholm, SE). Robustness of the automatic, patient-specific margin expansion is demonstrated using 4D dose accumulation.

Materials and Methods
18 patients with a lung tumor were selected who received 8x7.5Gy (60Gy) SBRT. All patients were scanned with a 10-phase 4D-CT, from which a midpos CT was reconstructed using in-house deformable image registration software. Subsequently, the midpos CT was deformably warped to each 4D-phase, resulting in 10 deformable vector fields (DVFs). The GTV was delineated on the midpos CT. The mean standard deviations (SDauto) in the GTV over the DVFs, in all directions (CC, AP, LR), were calculated. Next, SDauto was inserted as random error in the Van Herk margin recipe, resulting in an individualized PTV margin. Treatment plans were optimized following departmental guidelines for MR-linac lung SBRT using Monaco-v5.40.01 (Elekta AB). To validate SDauto, a manual rigid registration of the lesion in all 4D phases was performed relative to the end-exhale phase. Based on the nine resulting displacement vectors, SDman was calculated for the GTV for all directions. As additional plausibility check, the p2p motion was divided by 3 and used as (semi-)random error (SDp2p) in the Van Herk margin recipe following the clinically used rule-of-thumb. Dosimetric robustness of the automatically derived margins was investigated with 4D dose accumulation. To this end, the midpos plans were recalculated on each phase of the 4D-CT, then the dose was warped to the midpos CT using the energy-mass transfer dose mapping and accumulated. V60Gy[%] in the GTV and V57Gy[%] for an expanded GTV (GTV+2mm) (2 mm isotopic expansion to incorporate delineation uncertainty) was used to assess target coverage.

Results
The SDauto and the SDman / SDp2p agreed for 94%/94% (RL) , 100%/78% (AP), and 72% / 78% (CC) of all patients within 1 mm (Fig 1). The largest differences with SDauto were observed for the CC direction for SDman [range: - 4.6 mm/+0.9 mm] and for SDp2p [range: -2.8 mm/+1.7 mm], in which tumor motion was largest. The automatically derived margin differed from the p2p-based margin in three cases by -1 mm and in one case by +1 mm in CC direction. No differences in margin were observed in AP and RL direction. Average GTV coverage (V60Gy) for the 4D accumulated dose plans was 98.9% (SD: 3.1%, range: 87.8%-100%) and for the planned dose 99.5% (SD: 2.0 %, range: 90.9%-100%) (Fig 2). The average GTV+2mm coverage (V57Gy) was 98.6% (SD: 3.4%, range: 86.1%-100%).

Conclusion
Automatic, patient-specific PTV margins are feasible for midpos-based MR-linac lung SBRT planning, and result in clinically acceptable target coverage