Intensity modulated arc therapy implementation in a three phase adaptive ¹⁸F-FDG-PET voxel intensity-based planning strategy for head-and-neck cancer

The first 10 head-and-neck cancer patients treated with adaptive ¹⁸F-FDG-PET-voxel intensity-based IMRT in a randomized phase II dose-escalation clinical trial (NCT01341535) were selected for this study (Table 1). All tumors were biopsy-proven non-metastatic head-and-neck squamous cell carcinomas.

All patients were positioned with a five-point thermoplastic mask (Orfit Industries N.V., Belgium), which extended down to the shoulders, during computed-tomography (CT) isocenter simulation and treatment delivery. Planning CT scans of 3 mm slice thickness were acquired before the treatment and after the 8^th and 18^th fraction. A verification CT was taken at the treatment end. Contrast-enhanced ¹⁸F-FDG-PET/CT (Philips Medical Systems, Germany) was performed before treatment and after the 8^th fraction. ¹⁸F-FDG-PET-images were acquired with a voxel size of 4 x 4 x 4 mm³ as described earlier [9]. Fusion of the planning CT and ¹⁸F-FDG-PET/CT scans was done on a Pinnacle treatment planning system, version 9.0 (Philips Medical Systems, Andover, MA).

Delineation of the gross tumor volume of the primary tumor (GTV_T) and pathological lymph nodes (GTV_N) was done using mutual information of both anatomical and biological imaging. A threshold level of 50 % of SUV_MAX (maximal standardised uptake value) was set for ¹⁸F-FDG-uptake in Pinnacle. Pathologic lymph nodes were delineated separately and noted as the GTV_N1 and GTV_N2. The high-risk clinical target volume (CTV_HR) was created combining the GTV_N and a three-dimensional expansion of the GTV_T with 1 cm and adjusted to the air cavities and uninvolved bones. 3 mm margin to the CTV_HR was used to create the high risk planning target volume (PTV_HR). Delineation of the elective neck regions according to the guidelines of Gregoire et al. [10] resulted in the CTV of the elective neck (CTV_EN) and the elective neck PTV (PTV_EN) after a 3 mm expansion in all directions.

The considered organs-at-risk (OARs) were spinal cord, brainstem, swallowing structures defined as one region-of-interest (superior, medial and inferior pharyngeal constrictor, upper oesophageal sphincter, first 2 cm of the oesophagus and supraglottic larynx), parotids and mandible. Planning OAR volumes (PRVs) were created for the spinal cord and brainstem by three-dimensional expansions of 5 mm and 3 mm, respectively.

Deformable image co-registration (ABAS, version 0.41, Elekta CMS Software, Maryland Heights, MO) was used to propagate the targets and OAR contours from one CT to another in chronological order. All structures were reviewed and edited if necessary by an experienced head-and-neck radiation oncologist.

Treatment phases I, II and III consisted of 10 fractions planned on the 1^st, 2^nd and 3^rd CT set, respectively. Dose-painting was performed in GTV_T and GTV_N during the first 20 fractions. The dose range was between 2.2 Gy and 3.1 Gy per fraction in phases I and II. Only a GTV_T volume ≤ 1.75 cm³ was allowed to receive more than 2.9 Gy per fraction. GTV_N was dose-painted in 4 out of 6 patients with N+ disease; in the 2 other patients, which had a pathological lymph node volume ≤ 4 cm³, the GTV_N median prescription dose was 2.2 Gy per fraction. The total dose range for the GTV_T and GTV_N was 66-83 Gy.

No dose-painting was performed during the last 10 fractions, where a D_95% of 2.0 Gy/fx was prescribed to PTV_HR. Elective neck was irradiated during fractions 1-20 with a median total dose prescription of 40 Gy to PTV_EN. GTV_T and GTV_N biologic conformity was measured by a quality factor (QF), defined as the mean deviation between prescribed and planned dose in each PET/CT voxel [9]. QF was kept below 5 % where possible. Every treatment was planned to a total of 30 fractions and then rescaled to 10 fractions. Maximum doses of 50, 60 and 70 Gy were allowed to < 5 % of the spinal cord (PRV), brainstem (PRV) and mandible, respectively. A maximal dose of less than 45 Gy for the spinal cord, 50 Gy for the brainstem and 27 Gy to < 50 % of the volume of the spared parotids, respectively, were considered clinically acceptable.

The methodology of ¹⁸F-FDG-PET voxel intensity-based DPBN has been previously discussed [9]. Briefly, a dose is prescribed to the voxels in the dose-painted target volume as a function of signal intensity as follows:

where the signal intensities I_high and I_low are determined as 95 % of the maximum ¹⁸F-FDG-PET intensity and as 25 % of I_high, respectively. The extension of the discrete PET intensity data to the continuum was implemented using trilinear interpolation for the randomly seeded points in the delineated volumes. Using the PET-intensity to dose relation, the dose prescription was on a point-by-point base.

All treatment plans were created for an Elekta linac (Crawley, UK) equipped with a standard multileaf collimator with 40 leaf pairs, capable of delivering s-IMRT and IMAT with variable dose rate, gantry and collimator rotation speed. In-house developed software using an anatomy- and ¹⁸F-FDG-PET-voxel intensity-based segmentation tool (ABST, BBST) followed by leaf position and monitor unit (MU) optimization was used for treatment planning [11, 12].

s-IMRT plans consisted of six non-opposing coplanar 6 MV beams with gantry angles of 45°, 75°, 165°, 195°, 285° and 315°. The IMAT class solution was made of 6 MV arcs collimated around PTV_EN (gantry angle from -176° to 176°) and PTV_HR (144° to -144°) with control points (CPs) defined every 8°. The only constraints were on the physical abilities of the linear accelerator to deliver the treatment (maximum gantry speed, maximum collimator rotation speed, maximum leaf speed, minimum dose rate), and a minimum distance constraint of 1 cm for opposite and diagonally-opposite leaves of the MLC. ABST [11] was used to create the starting set of CPs, resulting in multiple initial arcs, avoiding both parotids, the swallowing structures and the PRV of the spinal cord. The CPs were optimized as described previously [12]. ABST generates beam segments with leaf and jaw positions based on a beams-eye-view projection of selected PTVs and OARs. BBST additionally takes into account PET-intensities to create initial beam segments shapes [9]. For a faster delivery, the parts of the arcs with a contribution of less than 2 MUs were eliminated during the optimization leading to the split of the arcs in sub-arcs. A CP refinement was performed by interpolating and generating additional CPs within the arcs, followed by MU and leaf position optimization. This CP refinement limited MU differences, gantry and collimator angle differences, leaf and jaw position movements between CPs and was applied to reach the accuracy constraints used in the treatment verification. After the final optimization, the remaining arcs were linked together in one beam according to the shortest possible delivery time. All dose computations were done in Pinnacle with a collapsed cone convolution/superposition calculation algorithm.

Doses of the 3 treatment plans were summed on the pretreatment CT using in-house developed software [13] based on the deformable CT image registrations made with the ABAS software. The reporting of the region-of-interest (ROI) dose levels was done on the summed doses.

To assess the risk of inducing secondary malignancies, the integral dose was calculated in the patient volume as follows:

where D_mean is the mean dose, V is the volume and ρ the tissue density, which was considered to be 1 g/cm³.

The delivered dose distributions of all IMAT and s-IMRT treatment plans were verified with the 3D dosimetry system Delta⁴ (Scandidos, Uppsala, Sweden). The Delta⁴ phantom has 1069 p-type disc-shaped Silicon diodes with a diameter of 1 mm and axial size 0.05 mm, in a central region (6x6 cm) spaced per 5 mm, outside the central region spaced per 10 mm. Global gamma indices [14] were determined in the Delta⁴ control software for the criteria of 3 % dose difference and 3 mm distance-to-agreement, the normalization dose being the prescribed dose.

The delivery treatment time was also recorded from the start of the first beam till the end of the last beam.

Population average dose-volume parameters of targets and OARs for both strategies are shown in Table 2. Most of the differences between s-IMRT and IMAT for target and OAR dose levels were significant (Table 2). Mean V_27Gy of ipsi- and contralateral parotids were improved by 16.4 % (p = 0.007) and 17.5 % (p = 0.003) in the IMAT plans, respectively. For the volume of interest that comprised the pharyngeal constrictor muscles (PC) and the one that combined the swallowing structures (SS), both D_50% and D_98% levels were significantly improved in the IMAT plans, while D_2% did not show on average any important differences.

Analysis of each summed dose distribution separately revealed larger differences for some cases in comparison with average data. Additional file 1: Figure S1 showed similar or highly reduced D_50% and D_98% levels of PC and SS with up to 54.9 % for IMAT compared to IMRT. D_2% differences of the same structures varied from -2.8 % to 3.6 %. For a cT4a pN2 cM0 oropharynx cancer case the results were plotted in Fig. 1. IMAT ipsilateral and contralateral parotid mean dose was lowered by 24.9 % and 5.3 %, respectively, while V₂₇ was also improved by 24.9 % and 6.7 %, respectively. Additional file 2: Figure S2 provides for the same patient a visual image of how IMAT isodoses better spare the parotids on every treatment phase, except for the contralateral parotid on the third treatment phase. The s-IMRT median dose of the swallowing structures was 22.7 % and 12.3 % higher for the PC and SS structures.

GTV_T quality factors (QF) were significantly better for the IMAT-plans (p < 0.001 for both DPBN-phases) with a maximum difference from IMRT factors of -2.1 % in phase I and II (Table 3). When the QF values of GTV_T and GTV_N were considered as one group, a Wilcoxon test also showed significantly (p < 0.001) lower values for IMAT.

The integral dose inside the patient was lower for IMAT in 7 patients with a maximum difference of 14.4 % (Table 4). For 2 cases, IMAT integral dose was with 2.8 and 3.7 % higher, while for one case it was similar.

The data on dosimetric verification of treatment plans, number of MUs and delivery time are presented in Table 5. The number of MUs was significantly higher for IMAT than for s-IMRT plans. All treatments were delivered on the Elekta linacs while measuring with the Delta⁴ system. Mean percentages of the points with gamma index >1 were 99.7 ± 0.6 % versus 98.7 ± 1.3 % for s-IMRT and IMAT, respectively. On average, s-IMRT treatment times of phases I, II and III were 6:52, 6:39 and 5:00 min, respectively. IMAT delivery was significantly shorter: 4:13, 3:44 and 2:56 min, respectively.

The data used in this manuscript are part of the study registerend on clinicaltrials.gov under NCT01341535. This study was approved by the Ethics Committe of Ghent University Hospital.