Inter-fraction robustness of intensity-modulated proton therapy in the post-operative treatment of oropharyngeal and oral cavity squamous cell carcinomas

Abstract

Objective:

To evaluate dosimetric consequences of inter-fraction setup variation and anatomical changes in patients receiving multifield optimised (MFO) intensity modulated proton therapy for post-operative oropharyngeal (OPC) and oral cavity (OCC) cancers.

Methods:

Six patients receiving MFO for post-operative OPC and OCC were evaluated. Plans were robustly optimised to clinical target volumes (CTVs) using 3 mm setup and 3.5% range uncertainty. Weekly online cone beam CT (CBCT) were performed. Planning CT was deformed to the CBCT to create virtual CTs (vCTs) on which the planned dose was recalculated. vCT plan robustness was evaluated using a setup uncertainty of 1.5 mm and range uncertainty of 3.5%. Target coverage, D_95%, and hotspots, D_0.03cc, were evaluated for each uncertainty along with the vCT-calculated nominal plan. Mean dose to organs at risk (OARs) for the vCT-calculated nominal plan and relative % change in weight from baseline were evaluated.

Results:

Robustly optimised plans in post-operative OPC and OCC patients are robust against inter-fraction setup variations and range uncertainty. D_0.03cc in the vCT-calculated nominal plans were clinically acceptable across all plans. Across all patients D95% in the vCT-calculated nominal treatment plan was at least 100% of the prescribed dose. No patients lost ≥10% weight from baseline. Mean dose to the OARs and max dose to the spinal cord remained within tolerance.

Conclusion:

MFO plans in post-operative OPC and OCC patients are robust to inter-fraction uncertainties in setup and range when evaluated over multiple CT scans without compromising OAR mean dose.

Advances in knowledge:

This is the first paper to evaluate inter-fraction MFO plan robustness in post-operative head and neck treatment.

Introduction

Radiotherapy for head and neck cancers is challenging due to the proximity of normal structures. Irradiation of healthy tissue can lead to long-term toxicities such as xerostomia, dysphagia and dysgeusia that have a negative impact on quality of life.¹ The finite range of high energy protons makes proton beam therapy an attractive treatment option for squamous cell cancers of the head and neck by limiting dose to normal tissues beyond the target volume. However, the sharp distal fall-off of proton beams makes the dose distribution sensitive to setup, motion and range uncertainties. In proton treatment planning, potential sources of error resulting from uncertainty in the location of the Bragg peak need to be considered. Potential sources of error include: daily changes in the patients’ position or anatomy, organ motion, delineation variation, image artefacts, inaccurate conversion of CT Hounsfield units to the proton stopping power and changes in the beam path due to variable tissue densities.^2,3 Patient immobilisation, image guidance, expansion margins and use of a dual energy CT may reduce errors but in clinical practice centre-specific setup uncertainties and ±3.5% uncertainty on stopping powers are typically used to account for residual uncertainties.^4,5 Despite attempts to reduce setup uncertainties, some daily variations are unavoidable and may result in under- or overdosage of the target volume and organs at risk (OARs) respectively.

Robustness to uncertainty may be attained in many ways. One such approach is through robust plan optimisation whereby uncertainties are accounted for in the optimisation process. The optimisation process can reduce dose gradients within the plan and consequently, the occurrence of hot and cold spots due to uncertainties in the proton beam range. In robust optimisation, patient setup errors and the corresponding perturbations in the dose distribution are simulated by shifts in the isocentre position on a single CT scan. Range uncertainties are simulated by scaling the CT to stopping power ratio calibration. Setup variation, however, is more complex than a rigid translation of the CT image and can involve soft tissue deformation, variations in spine alignment, neck tilt and shoulder position. In the literature, different methods of robust plan optimisation exist, including probabilistic treatment planning, the use of worst-case scenarios, selective robust and minimax optimisation.^6–8 Probabilistic treatment planning described by Unkelbach incorporates both range and setup as random variables into the optimisation process. This contrasts with worst-case optimisation described by Liu at al whereby the optimisation is based on the worst case scenario.^9,10 However, the robust optimisation process only accounts for positional and range uncertainties, and does not evaluate robustness against anatomical changes during the treatment course. Commercial systems using multiple patient images prior to treatment are being studied for anatomical robust optimisation but are yet to be adopted in routine clinical practice.^11,12 Using pencil beam scanning, two methods for plan optimisation can be used: single field optimisation (SFO) and multifield optimisation (MFO). In the latter, all beam spots are optimised together using multiple fields and beam angles to modulate and shape the dose. A superior dose distribution can be delivered compared with passively scattered proton beams or SFO, in particular in cases of complex geometry. MFO, however, can be prone to inter-fractional uncertainties due to anatomic changes, such as from weight loss or tumour response, resulting from increased in-field dose gradients. These non-linear changes in patient’s anatomy due to tissue deformations are not explicitly modelled in robust optimisation; it is unknown if robustly optimised MFO can still accommodate these unplanned range variations.

Conventionally, uncertainties are accounted for with the use of a planning target volume (PTV). In a study of fourteen patients by Liu et al, a robust clinical target volume (CTV) based optimisation approach produced a superior plan for targets and OARs compared with a PTV approach.⁵ Stutzer et al showed MFO to be superior to SFO in original plan robustness in the treatment of oropharyngeal cancer.¹³

MFO is potentially more sensitive to tissue deformation during the treatment course, which is likely to be different along different beam angles. The effectiveness of robust plan optimisation in MFO has not been studied in post-operative head and neck treatment planning. There is also no agreed consensus as to which optimisation method is superior in evaluating MFO plans partly due to variable literature and lack of standardised protocols. In this study, we aim to evaluate how robust MFO plans are to uncertainties in set up and range error for patients treated post-operatively for oropharyngeal and oral cavity cancers using weekly cone beam CT (CBCT) scans. Such CBCT scans capture realistic inter-fraction setup variations as well as tissue deformations during the entire treatment course.

Methods

Patient selection

Six head and neck patients treated between July 2017 and April 2018 and planned with multifield robust optimisation in Eclipse (Eclipse v. 13.7, Varian Medical systems, Palo Alto, CA) were retrospectively selected. Inclusion criteria were: patients with oropharyngeal or oral cavity cancers requiring post-operative proton beam therapy to the primary site and elective neck, and the availability of weekly CBCT images. All patients were immobilised in a 5-point thermoplastic mask and positioned using daily orthogonal kV imaging. The relative percentage change in each patient’s weight was recorded weekly during treatment which was subsequently correlated with CTV and OAR coverage.

Planning approach

Each patient was treated with two or three CTV dose levels. CTV1 was defined as the surgical bed with CTV2 and CTV3 defining “at-risk sites”. CTV1 received 60–63 Gy (RBE), CTV2 received 54–60 Gy (RBE) and CTV3 received 54 Gy (RBE) all treated with a dose of between 1.8 and 2.1 Gy (RBE) per fraction. Each treatment plan was based on a three-field beam arrangement consisting of two posterior obliques and one anterior field with range shifters (7.2 cm anterior 8.0 cm posterior) in situ as shown in Figure 1(a) , and (b). The posterior oblique fields cover the superior portion of the target while the anterior field covers the inferior target. The posterior oblique and anterior fields overlap in the superior–inferior direction over a 2 cm region and are optimised to be robust against 3 mm longitudinal isocentre perturbations between the posterior oblique and anterior fields. In this way, smooth dose gradients are achieved in the superior–inferior directions for each field at the overlap region without the need to feather the match line. Range shifters are positioned close to the patient to minimise the post-range shifter air gap and subsequent increase in spot size.

Figure 1.

Three field IMPT beam arrangement (a) and (b) with two posterior oblique fields and an anterior field indicated by red arrows. In situ range shifters located anterior and posterior to the patient are used to minimise the air gap between the range shifter and the patient. Representative lateral (c) and anterior (d) DRRs of one planning CT and maximum intensity projection of 5 vCTs (e) and (f) with red contours representing inter-fraction setup variation of cervical vertebrae. DRR, digitally reconstructed radiograph; IMPT, intensity modulated proton therapy; vCTs, virtual CTs.

The total prescribed dose to the original treatment nominal plan was defined such that 95% of each target volume received a minimum of 100% of the prescribed dose to this CTV level. Clinically acceptable plans require 95% of the target volume in the worst-case scenario to receive at least 95% of the prescribed dose and D_0.03cc to receive ≤110% of the prescribed dose.

Analysis of robustness

Plans were optimised using MFO. CTVs were robustly optimised using a setup uncertainty of 3 mm and a range uncertainty of 3.5%. Setup perturbations were simulated along three orthogonal directions and combined with range uncertainties in each position to produce 12 scenarios used in the robust optimisation, (x ± 3 mm), (y ± 3 mm), (z ± 3 mm) and CT Hounsfield unit (HU) scaling of ±3.5%.

Daily positioning and setup correction were accomplished using orthogonal kV imaging. Each patient underwent weekly online CBCT, which were then used to create reliable virtual CTs (vCTs). If a CBCT was acquired, a 3D–3D match was performed followed by a confirmatory kV imaging without additional couch motion. The vCTs were generated by deforming the planning CT onto the CBCT using a diffeomorphic implementation of the Morphons algorithm.¹⁴ The deformable registration provided a high-quality image in the treatment geometry on which the planned dose could be evaluated to estimate the delivered dose. Each target CTV on the vCT was compared with the original treatment nominal plan and manually edited to modify the superior and inferior extension to ensure consistency of anatomical landmarks. The vCT plan robustness evaluation was performed using a residual setup uncertainty of 1.5 mm and range uncertainty of 3.5%. The 1.5 mm setup robustness evaluation considers the uncertainty in the coincidence between the imaging and radiation isocentres, intra-fraction motion, as well as variations in user dependent choice of region of interest for evaluating image registration between the vCT and the planning CT. Actual setup variability of the cervical neck vertebrae is illustrated in the digitally reconstructed radiographs depicted in Figure 1(e) , and (f) for Patient 5. A video depicting the setup variation is available in the supplementary data.

Dose calculation

All doses and dose–volume histograms (DVHs) were calculated using proton convolution superposition (PCS) v. 13.7 within the Eclipse^™ treatment planning system (Varian Medical Systems, Palo Alto, CA). Plan metrics were extracted from the calculated DVHs using a MATLAB (Mathworks, Natick, MA) script. Maximum and minimum values under uncertainty for D_95% for each CTV dose level and D_0.03cc for the high dose CTV were extracted and compared with the original treatment nominal value. Mean dose in the vCT-calculated nominal case to ipsilateral and contralateral parotid glands, oral cavity, pharyngeal constrictor muscles, larynx and maximum dose to the spinal cord were calculated. Relative % change in weight from baseline to end of treatment was recorded.

Results

The demographics of the six patients are outlined in Table 1. In all patients, the vCT-calculated nominal treatment plan D_95% across the vCTs was at least 100% of the prescribed dose. The median time from baseline radiotherapy planning scan to first CBCT within week 1 of treatment was 29 (range 22–52) days.

Table 1.

Patients’ demographics

Patient number	1	2	3	4	5	6
Age (years)	73	50	67	70	55	74
Gender	M	M	M	F	M	M
Site	Base of tongue	Tonsil	Oral tongue	Tonsil	Vallecula	Oral tongue
Laterality of tumour	Right	Right	Right	Right	Left	Left
Stage	T2N2bM0	T2N2aM0	T2N1M0	T2N1M0	T4aN1M0	T2N0M0
HPV status	Positive	Positive	Negative	Positive	Positive	Negative
Chemotherapy	None	Weekly Cisplatin	None	Weekly Cisplatin	Weekly Cisplatin	Weekly Cisplatin
Prescription dose/Gy RBE	60	63	60	63	63	63
Baseline weight	78	92	79	78	99	109
Weight loss in kg at the end of treatment (relative % change)	0	3 (3%)	0	5 (6%)	1 (1%)	7 (6%)

F, Female; M, Male; RBE, relative biological effectiveness.

Across the six patients, robustness evaluation of CTV coverage using inter-fraction uncertainties of 3 mm setup and 3.5% range error for the initial plan (day 0) and 1.5 mm for the residual setup and 3.5% range for the vCT-calculated plans are shown in Figure 2a and b. The dose to the vCT-calculated nominal plan is close to the maximum CTV dose band across all six plans. Clinically acceptable D_0.03cc values in the high dose CTV levels are shown in Figure 2a and b with the exception of Patient 6. In Patient 6, the highest D_0.03cc value for the vCT-calculated nominal plan is 113% (71 vs. 63 Gy) of the target dose. The hot spot in patient six is shown in Figure 3 and can be explained by a shift in the jaw position and variation of the posterior neck. When uncertainty analysis is included, the highest prescribed D_0.03cc for this fraction was 116% of the target dose (73 vs. 63 Gy if scaled to the full treatment dose). Analysis of daily kV setup image revealed that this relatively large variation in jaw position occurred only once during the entire course of treatment. The D_0.03cc values for the other two vCT-calculated nominal plans of Patient 6 in Figure 2b were within the uncertainty bands of the initial plan.

Figure 2.

Figure shows an overview of the change in D_95%, D_0.03cc in the high-risk CTV and patient weight from baseline to the end of treatment. Uncertainties over all error scenarios are shown as a light band, CTV 63 (green), CTV 60 (red), CTV 54 (blue). The solid line shows the nominal dose. Dashed lines indicate weight changes of ±5% from the baseline weight recorded at the time of the planning CT scan. CTV, clinical target volume.

Figure 3.

(a) Sagittal view of planning CT patient six with CTV1 contour and (b) corresponding slice of the first vCT-calculated nominal plan. The hotspot is attributed to differences in the jaw position (shown by the dashed red reference line) and the setup variation of the posterior neck region. CTV, clinical target volume; vCT, virtual CT.

None of the six patients lost >10% weight relative to their baseline throughout treatment. The largest absolute (relative %) weight loss were noted in patients 4 and 6 both of whom lost 6% by the end of treatment. Two patients lost no weight and in the remaining two patients their relative % weight loss remained within 5%. Figure 4 illustrates the mean and maximum doses to the OARs with time.

Figure 4.

Figure illustrates the changes in mean dose to the oral cavity, pharyngeal constrictor muscles, larynx, ipsilateral parotid gland and maximum dose to the spinal cord with time. Dose constraints to each OAR are; oral cavity mean dose 20 Gy or ALARA, pharyngeal constrictor muscles mean dose 50 Gy or ALARA, larynx mean dose 20 Gy or ALARA, ipsilateral parotid gland mean dose 20 Gy or ALARA, maximum spinal cord 45 Gy. ALARA as low as reasonably achievable; OAR, organ at risk.

The impact of inter-fraction setup, anatomical variation and changes in patient weight on the OAR doses is depicted in Figure 5. While vCT-calculated nominal doses do vary for each patient, dose constraints did not exceed planning objectives for any OARs.

Figure 5.

Axial (a) and sagittal (b) views of setup variation in the neck region between planned and actual treatment positions for Patient 4. Blended images of planning CT and vCT shown demonstrating differences in the neck angle as indicated by red arrows. Nominal dose distribution on the planning CT is shown in (c) with CTV 54 contour in blue and the corresponding vCT dose. Green body contour of the planning CT is shown in (c) and (d). Red dotted circle in (d) indicates region of hotspot outside of CTV 54 and cold spot within CTV 54. Red arrow in (d) indicates region with variation in setup position. CTV, clinical target volume; vCT, virtual CT.

Discussion

This novel pilot study to evaluate dosimetric consequences of anatomical variations and range errors in MFO plans in the post-operative setting has demonstrated:

(i) plans are robust to uncertainties in setup and range, (ii) weight loss within 6% does not negatively impact dosimetric coverage, (iii) OAR exposure for all vCT-calculated nominal plans are within tolerance.

MFO plans have been shown to be robust to uncertainties early in the treatment course enabling plan adaptation if necessary. Weight loss of >10% from baseline in head and neck radiotherapy is influenced by radiation related toxicities and significantly associated with global poor health-related quality of life.¹⁵ Changes in weight can affect the positioning of OARs such as the parotid gland resulting in potential overdosage. The use of proactive enteral feeding is the cause of much debate. In this study, none of the six patients lost ≥10% weight throughout treatment and required enteral feeding. Of the two patients who lost the largest amount of weight (6%) from baseline, plans remained robust to uncertainties. This highlights the potential importance of aggressive and proactive symptom and supportive management during treatment, so as to minimise weight loss.

Proton planning, particularly in head and neck radiotherapy is challenging due to uncertainties in setup and range on a background of complex anatomy and close proximity of tumours with normal tissues. Optimisation techniques may help minimise the effect of uncertainties, accepting the trade-off between delivering an acceptable integral dose and normal tissue sparing.¹⁶ In this study, dose constraints to the oral cavity, pharyngeal constrictor muscles, larynx and parotid glands were achieved without compromising plan robustness.

MFO proton beam plans in head and neck radiotherapy are susceptible to the development of hot and cold spots due to anatomical motion and changes in the position of the head and neck. During radiotherapy, as patients relax, the neck and jaw position can move, which potentially alters the dose distribution. In all six patients, the distribution of hot spots in the vCT-calculated nominal plan as defined by D_0.03cc were clinically acceptable (within 110% of the prescribed dose and 115% for the worst case). In Patient 6, D_0.03cc of 113% in the vCT-calculated nominal plan was noted at Day 34 from baseline. The development of a hot spot correlated with a shift in the jaw position as shown in Figure 3. The dose distribution corrected to within the clinically acceptable range on subsequent weekly CBCTs. Changes in the patients’ position is a random variable unlike weight loss which is more systematic throughout treatment. More regular imaging with daily CBCTs could be considered in those who experience hot spots, D_0.03cc >110% and weight loss >10% to ensure an adequate dose distribution to the target and maintain OAR constraints. Another example of the impact of setup variation of the neck and its impact on dose distribution of MFO plans is depicted in Figure 5 where the hotspot is seen outside of the CTV. Current robust optimisation techniques do not take into account anatomical deformation.

Although this is the first study to evaluate optimisation of MFO plans in post-operative oropharyngeal and oral cavity patients, the sample size is small and not all patients underwent the same number of CBCT scans during treatment. In addition, the mean doses at each vCT-calculated nominal plan reflect dose to the full prescription but the measurement point is for a single fraction. A more thorough analysis could be performed with daily CBCT using deformable dose accumulation on the vCTs. Unfortunately, daily CBCT was not available for this study. Challenges of maintaining robustness may be different in other anatomical sites in the head and neck, such as in the paranasal sinuses and skull base, where random, daily changes in sinus filling may occur. Despite these limitations this is the first paper to our knowledge to evaluate MFO robust optimisation in post-operative oropharyngeal and oral cavity cancer patients and may help in the process of developing optimisation protocols with agreed parameters to help identify plans that require additional individualisation.

Conclusion

In this novel pilot study, MFO plans in post-operative oropharyngeal and oral cavity patients were observed to be robust to uncertainties in setup and range in regard to CTV coverage when evaluated over multiple CBCT scans. Dose constraints to OARs remained within tolerance and changes in weight from baseline did not appear to affect CTV coverage or plan quality. Development of a robust analysis protocol for MFO plans may improve consistency of reporting and plan evaluation amongst radiotherapy centres.