Comparative photon and proton dosimetry for patients with mediastinal lymphoma in the era of Monte Carlo treatment planning and variable relative biological effectiveness

After obtaining approval from the University of Washington Institutional Review Board, we retrospectively reviewed 8 consecutive mediastinal lymphoma patients that were simulated between January 1, 2015 and May 1, 2017 at the SCCA Proton Therapy Center or University of Washington. All patients underwent 4D free breathing (FB), DIBH, and a helical contrast CT simulation scans. Patients were simulated with a thermoplastic mask with gentle neck extension and arms down or akimbo (if the axilla was treated). The Active Breathing Coordinator system (Elekta, Stockholm, Sweden) was used for DIBH. A single radiation oncologist with expertise in lymphoma (YDT) contoured all cases using the involved site technique [16, 17] and retrospectively contoured cardiac substructures using a published contouring atlas [18]. One set of the cardiac contours were reviewed with a radiologist with expertise in cardiac imaging (GK) and feedback was incorporated before contours were finalized. For the FB 4D CT scan, internal target volumes (ITV) were drawn on the average intensity projection (AVE-IP) images and edited using the maximum intensity projection (MIP) and cine images. The CTVs were drawn on DIBH images. The CTV/ITV to PTV margin ranged from 5 to 7 mm across patients but was constant for comparative plans for a single patient.

Mediastinal lymphoma patients included those with classical Hodgkin lymphoma (n = 5), primary mediastinal B-cell lymphoma (n = 2), and grey zone lymphoma (n = 1). Median age at radiation CT simulation was 34 years (range, 18–38). Seven of 8 patients had mediastinal disease that extended below the left pulmonary artery (i.e. lower mediastinal involvement). Extent of disease is summarized in Table 1. Among the 8 patients that underwent simulation, 6 were treated with pencil-beam scanning (PBS) proton therapy in free breathing (Table 1), which was calculated using the pencil-beam algorithm. DIBH was not used with proton therapy given lack of volumetric image-guidance at our center. One patient refused radiotherapy after simulation (patient 5), and one patient (patient 6) was treated with 3D conformal photons in DIBH given that her disease was limited to the right neck and upper mediastinum and above the level of the left pulmonary artery. While prescription doses ranged from 20 to 45 Gy, for this dosimetric comparison study, the same prescription dose was used for all patients: 30 Gy in 15 fractions.

PBS proton plans were created on the DIBH and free breathing scans. The free-breathing plans were planned on the CT obtained by averaging the ten phases from a 4D CT acquisition. The DIBH plans were performed on static CT obtained with the patient under breath-hold and monitored using the Automatic Breathing Coordinator™ device (Elekta Inc., Sweden). Proton plans were calculated in the RayStation treatment planning system (version 6) using a PBA-based dose algorithm (version 4.1) and MC-based dose algorithm (version 4.0). The beam model corresponds to the commissioned IBA Proteus Plus beam at Seattle Cancer Care Alliance Proton Therapy Center [19]. Anterior or anterior oblique beams within +/− 30 degree from vertical were used. Single field uniform dose optimization technique was used with at least two beams, and 2X volumetric repainting was performed for patients with excessive motion (> 1 cm) in the target area. The decision to use 2X volumetric repainting is based on results from the Monte Carlo study by Grassberger et al. [20]. In this study, the optimal number of repaintings was calculated to preserve target coverage based on tumor size and its motion. The subsequent dosimetric endpoints provided in the results section account for 2X volumetric repainting, wherever it is applicable. All 8 patients had PTV volumes extending <=7.5 cm to patient’s external contour. Thus, a 7.5 cm water equivalent thickness range shifter of acrylic material was inserted into the beam path to ensure adequate coverage of the target superficially. The lateral spot and energy layer spacing parameters were set to an automatic scale of 1 for most proton plans and reduced in 0.2 increments to below 1 only for plans where coverage could not be met. The spot spacing parameter for proton plans in the RayStation treatment planning system determines the lateral grid spacing for proton spot placement. A higher value increases the inter-spot distance. Similarly, the energy layer spacing parameter determines the longitudinal spacing between Bragg peaks. The optimal value of these two parameters must provide the optimizer sufficient flexibility during the optimization process without unduly increasing the times for plan optimization and treatment delivery. For this study, the default value of 1 was found to be suitable for both these parameters, as also suggested by Alshaiki et al. [21].

All nominal proton plans were optimized so that at least 99% of the CTV/ITV achieved 99% of prescribed dose (CTV or ITV V_{99% RX} > 99%). Similarly, the nominal plans also ensured that at least 95% of the PTV achieved 99% of the prescribed dose (PTV V_{99% RX} > 95%). After achieving the desired target dose levels, the plans were further optimized to reduce OAR doses (in order of highest priority) to the heart, breast tissue (for females), and lungs.

The nominal plans were perturbed for set up and range errors, including under and over ranging of 3% and isocenter shifts of +/− 3 mm in the superior/inferior, anterior/posterior, and left/right directions. Our institution criterion is to ensure that at least 95% volume of CTV/ITV is covered by 95% of prescription dose in perturbed conditions (CTV or ITV V_{95% RX} > 95%). For this study, all the perturbed plans achieved institution criteria with minimum CTV or ITV V_95%RX of 97.9%.

To evaluate the impact of MC dose algorithm, original proton plans that were optimized and calculated with PBA algorithm (PBPB), were retrospectively re-calculated with MC dose algorithm (PBMC). The MC algorithm, as implemented in the RayStation planning system, has been shown to be highly accurate for dose calculation in heterogeneous media [22‐26] as encountered in the treatment of mediastinal lymphoma. The clinical implementation of MC for treatment planning at the SCCA proton therapy center has been published [27]. New plans were also created using MC for plan optimization and final dose calculation (MCMC). These plans were done with a constant RBE of 1.1 and on free breathing scans (FB), therefore resulting in three different study arms: i.e., FB PBPB, FB PBMC, and FB MCMC. MC was also used to optimize and calculate final proton dose on DIBH scans. These plans also used a constant RBE of 1.1 and are therefore referred in Tables 2 and 3 as DIBH MCMC.

Comparative photon plans were generated on the DIBH scan in RayStation with the collapsed cone convolution superposition dose engine (version 3.4). Collapsed cone convolution algorithms accurately predict dose in inhomogeneous tissues such as the mediastinum [28, 29]. Implementation of this algorithm in the RayStation photon treatment planning system was found to be accurate in determining point doses in anthropomorphic thorax phantom with composite doses within +/− 1% [30].

The same DIBH CTV/PTV coverage and OAR dose constraints used for proton planning were used for photon dose optimization. Dynamic MLC form of IMRT was used with 6 MV photon beam energy. Five to seven beams were used for IMRT with the butterfly technique [7]. The photon plans are referred to as DIBH Photon in Tables 2 and 3.

To assess the potential impact of spatial variations in proton RBE within the Bragg peak (i.e., vRBE), we used a published model for DNA double-strand break (DSB) induction [31‐34] implemented into a research version of RayStation (version 6R). As explained in the Appendix, the RBE for DSB induction is closely related to the RBE for cell survival [35, 36] for protons with LET up to ~ 15 keV/μm (kinetic energies > 2 MeV and a continuous slowing down approximation (CSDA) range in water > 0.07 mm). This LET range encompasses all regions of a pristine Bragg peak except those regions a millimeter or so beyond the tip of the Bragg peak (Additional file 1: Figure S1). The RBE-weighted dose, which is a product of RBE and physical dose, is hereafter referred to as biological dose or as dose. Biological dose distributions computed using the vRBE model are compared to biological dose distributions with a constant clinical RBE = 1.1 and to photon dose distributions (implicit constant RBE = 1.0).

Re-calculation of FB PBPB plans with MC (FB PBMC plan) demonstrated significant under coverage of the ITV V99 (p = 0.012) and V95 (p = 0.012; Table 2, Figs. 1 and 2). The average and median reductions for ITV V99 were 24.1% and 19.4%, respectively. The coverage loss was less for ITV V95, with average and median reduction of 2.5% and 1.6%, respectively. Compared to FB PBPB plans, FB PBMC plans showed greater heterogeneity in target dose levels with lower median homogeneity index (D_95%/D_5%): 0.94 versus 0.97. There was also increase in spinal cord dose maximum (p = .05) with average and median increase of 2 Gy and 2.7 Gy, respectively.

ITV coverage in FB PBMC plans was recovered by re-optimizing and calculating plans with MC (FB MCMC plan), with similar global dose maximum and homogeneity index as the original PBPB plans. Dose to the heart and lungs were not significantly different between FB PBPB and FB MCMC plans. Statistically, but not clinically, significant different doses to the mean right breast and maximum esophageal dose were observed (Table 2). There was also increase in spinal cord maximum dose for FB MCMC plans over FB PBPB plans, with average and median increase of 3.1 Gy and 4.1 Gy, respectively.

Pairwise comparison for both FB MCMC and DIBH MCMC proton plans was performed to DIBH photon plans. Proton plans had similar target coverage and homogeneity index as photon plans. Both DIBH and FB proton plans had lower average mean lung dose by 3 Gy and 1.4 Gy, respectively (Table 2; Fig. 3). There was marked reduction in V5Gy lung volume by 20% for proton plans, likely secondary to fewer beams used for proton plans and lack of exit radiation dose.

Proton DIBH plans also showed statistically significant reduction in mean heart dose over photon plans with median and average reduction of 1.8 Gy and 3.1 Gy, respectively (Table 2; Fig. 3). Proton plans also had reduced spinal cord maximum dose with average decrease of >9Gy for both DIBH and FB plans. The spinal cord dose differences could be attributed to beam angles that are required for plans. Proton plans can achieve desired target coverage levels with only anterior beams thus minimizing spinal cord dose. Photons plans, on the other hand, also used posterior beams that traversed through spinal cord thus increasing spinal cord dose.

Dose to cardiac substructures were compared between DIBH photons and MC-optimized proton plans with and without DIBH. Given that proton plans use anterior weighted beams that could range into the heart (i.e., high LET protons with an RBE > 1.1), we evaluated dose to cardiac substructures using a vRBE model for the endpoint of DSB induction. The RBE for DSB induction is close to a linear function of proton LET up to about 15 keV/μm (Additional file 1: Figure S2) and is closely related to the RBE for cell survival [35, 36].

The relative benefit of proton therapy and DIBH on mean dose to cardiac substructures varied across patients (Fig. 4). The potential for an increase in RBE at the end of range was not found to be large enough to offset the dosimetric advantages of proton therapy (Table 3). Dose to cardiac substructures with proton therapy were effectively the same using a constant clinical RBE = 1.1 and with the vRBE model in our mediastinal patient population (results not shown).