|Home | About | Journals | Submit | Contact Us | Français|
The use of thermoplastic immobilisation masks in head and neck radiotherapy is now common practice. The accuracy of these systems has been widely studied, but always within the context and time frame of the radiation delivery—some 6–8 weeks. There is growing current interest in the use of functional imaging to assess the response to treatment, particularly in the head and neck. It is therefore of interest to determine the accuracy with which functional images can be registered to baseline CT over the extended periods of time used for functional response assessment: 3–6 months after radiotherapy. In this study, repeated contrast-enhanced diagnostic quality CT and mid-quality localisation CT from a positron emission tomography/CT scanner were available for five time points over a period of 9 months (before, during and up to 6 months after chemoradiotherapy) for a series of eight patients enrolled in a clinical pilot study. All images were acquired using thermoplastic immobilisation masks. The overall set-up accuracy obtained from this 9-month study of 5.5±3.2 mm (1 standard deviation) and 1.9±1.3° (1 standard deviation) is in agreement with published data acquired over 6–8 weeks. No statistically significant change in set-up error was seen with time. This work indicates that thermoplastic immobilisation masks can be used to accurately align multimodality functional image data for assessment of the response to treatment in head and neck patients over extended follow-up periods.
Head and neck cancer accounts for 5% of cancers worldwide  with approximately 7000 new cases being diagnosed each year in the UK. The majority of patients with squamous cell carcinoma of the head and neck (SCCHN) present with locally advanced disease. Despite recent advances in multimodality therapy and technical delivery of radiotherapy, outcomes remain suboptimal with 5-year survival rates of 50–60%. However, functional image data, provided by positron emission tomography (PET) and dynamic contrast-enhanced MRI (DCE-MRI) or diffusion-weighted MRI (DW-MRI), have been shown to have a number of potentially important applications in external beam radiotherapy for head and neck cancer [2,3]. Functional imaging is routinely used in diagnosis and staging , and there is increasing current interest in its application to localisation and delineation of target volumes  and normal tissues . There is also growing evidence to support the use of functional imaging for early assessment of the response to therapy .
Ensuring accurate registration between functional and anatomical data is clearly of paramount importance and much work has been done to date developing and testing immobilisation systems for use in radiotherapy planning and delivery. Thermoplastic mask systems have been described providing set-up accuracy in the head and neck of 2.5±1.4 mm, with no increase in systematic error seen over an 8-week period . Similar accuracy was demonstrated with a polyvinyl chloride mask system [standard deviation (SD)=2.1 mm], with cut-outs in the mask to improve dose sparing to the skin but not affecting accuracy . Studies using cone-beam CT (CBCT) to assess set-up accuracy have shown mean vector lengths of 4.7±1.7 mm intercranially and 7.3±4.5 mm in the neck for the thermoplastic mask . Very similar SDs have been shown by other groups using thermoplastic shells and repeat CT during therapy (1.9 mm, 1 SD, for the upper neck and 5.7 mm, 1 SD, for the lower neck) . Set-up error with thermoplastic masks has been shown to increase linearly with treatment time, with a SD of 1.2 mm calculated for 32 intercranial patients treated supine over 15 min . In a large recent study, 762 CBCT scans were analysed from 11 patients using standard and skin-sparing nine-point thermoplastic masks . The interfraction population (SD) was 1.6 mm (1.1°) (random) and 1.0 mm (1.4°) (systematic). All set-up errors >2 mm for three fractions were corrected before calculating these figures.
However, if functional image data are to be accurately aligned with baseline (pre-treatment) imaging to assess response at 3 or 6 months after therapy, use of immobilisation systems over longer time intervals than the 6 or 7 weeks of radiotherapy is clearly of interest. It is well known that head and neck patients often experience dramatic shrinkage of nodal masses in the neck and general weight loss during treatment. This in turn can mean that the thermoplastic shells no longer fit perfectly and may be expected to lead to increased set-up error over longer periods of time; alternatively, weight gain following treatment may potentially reduce immobilisation accuracy. In this paper, we describe an investigation of the accuracy of head and neck patient set-up using a standard five-point thermoplastic shell system during and up to 6 months after the end of induction chemotherapy and definitive chemoradiotherapy (CRT).
The data used in this study are taken from a 10-patient clinical pilot study investigating the feasibility and usefulness of functional imaging in CRT for SCCHN. In brief, patients with locally advanced SCCHN scheduled to receive CRT were recruited into a study that had ethics committee approval (EC 08/H0801/132) at The Royal Marsden Hospital, Sutton, UK. All patients received two cycles of induction chemotherapy—75 mg m−2 cisplatin on day 1 and 1000 mg m−2 5-fluorouracil (5-FU) on days 1–4—followed by radical CRT (100 mg m−2 cisplatin on days 1 and 29). 65 Gy was delivered to areas of macroscopic disease in 30 fractions, with microscopic disease receiving 54 Gy in 30 fractions, using a simultaneous integrated boost intensity-modulated technique. All patients underwent anatomical (CT and MRI) and functional [DW-MRI and DCE-MRI and 18-fludeoxyglucose (18F-FDG) PET/CT] imaging at baseline (immediately prior to the start of chemotherapy); following two cycles of induction chemotherapy prior to the start of chemoradiotherapy; following 40 Gy of radiation (excluding PET/CT); and at 3 and 6 months after treatment. The median time between the baseline CT and radiotherapy planning CT was 42 days (range 39–47 days). Patients were positioned on flat-top couches and immobilised in a Posifix® (Civico Medical Solutions, Reeuwijk, Netherlands) five-point thermoplastic mask (Figure 1) in the radiotherapy treatment position for all imaging modalities. Contrast-enhanced CT was performed using a Philips Brilliance® Big Bore scanner (Philips Healthcare, Eindhoven, Netherlands) (0.75 s per rotation; 16×0.75 mm collimation; 60 cm field of view; 120 kV; 350 mA per slice). PET/CT was acquired using a Philips Gemini® PET/CT scanner (Philips Healthcare) (400 MBq 18F-FDG, 4 min per bed) with a non-contrast-enhanced CT from the base of the skull to the carina (50 mA per slice, 120 kVp). Details of the MRI acquisitions are not relevant to this work.
In order to study the accuracy of the five-point thermoplastic mask, all CT data from the Brilliance Big Bore and Gemini PET/CT scanners were imported into the Pinnacle3 radiotherapy planning system (v. 8.0; Philips Healthcare), giving a total of up to nine independent sets of CT data per patient. Data from eight patients were available for this study. Using the radiotherapy planning CT as a reference, every other CT data set was then rigidly registered to the reference CT data set in one of two ways: (1) using four locating holes in the base of the immobilisation head board as fiducial points, automatic point-based matching was carried out to register the immobilisation shells with the shell in the reference set; or (2) setting a bounding box around the spine and base of the skull (to exclude the shell and shoulders), anatomy-based matching was carried out using cross-correlation. The displacements in the x-, y- and z-directions (left–right, anterior–posterior and superior–posterior, respectively) and rotations about the x-, y- and z-axes were recorded for each pair of registrations and for each pair of images. The difference between the transformations required to register the immobilisation system base-plates for any pair of images and to register the patient bony anatomy for the same pair of images gives a measure of the accuracy of the immobilisation system. A total of 116 registrations for 58 image pairs were available for this study (for 2 of the patients recruited early in the trial, immobilisation shells were not available for the baseline time point). The matched images were then visually inspected by a single observer (MK) and adjusted to achieve the optimum bony anatomy alignment on the skull and upper neck. For any given pair of images, the difference between the point-based and anatomy-based registration is a measure of the set-up accuracy achievable with the thermoplastic immobilisation shell system. As well as recording the x-, y- and z-displacements and rotations, the total vector displacement and Euler angle were calculated, giving figures of merit for the overall set-up accuracy.
The repeatability of the point-based matching was assessed by repeating the complete matching process, including placement of the fiducial points on the images, 10 times for 1 representative pair of data sets with a single observer (MK). The accuracy of the anatomy-based registration was also assessed by repeating the process 10 times for a representative pair of data sets with a single observer (MK). Image pairs were randomly shifted and rotated before each repeat registration, although the same bounding boxes were used for all matches.
The overall accuracy of registration using the thermoplastic shell, calculated using all data available for this study, was 5.50±3.23 mm (1 SD) and 1.94±1.30° (1 SD). However, two patients experienced extreme anxiety when using the mask system at baseline, so a deliberate clinical decision was made to fully clip down their mask only for radiotherapy and to leave some clips undone for subsequent follow-up imaging. Inspection of the data showed that the mean set-up errors were larger for these patients when the mask was not fully clipped down, as might be expected. By selectively excluding these, a measure of the “ideal” accuracy of the thermoplastic shell (when used as designed) can be derived of 4.75±2.14 mm (1 SD) and 1.67±0.92° (1 SD) for the patient population presented here. In addition to the vector magnitude and Euler rotations, the individual x-, y- and z-components can also be considered. The mean set-up displacements in the three translational and rotational directions are x=3.42±2.89 mm, y=2.46±2.67 mm and z=2.16±1.71 mm, with rotations about the x-axis θx=1.18±1.29°, θy=1.04±0.92° and θz=0.44±0.50°. These results indicate that, although small differences exist between the different component directions, there is no single direction that dominates and all component errors are of a similar order.
Because half of the data in this study came from a diagnostic CT scanner dedicated to radiotherapy and half from a PET/CT scanner in the nuclear medicine department, differences in set-up accuracy might occur owing to differences in staff experience with immobilisation equipment . (Although identical immobilisation equipment is used by both staff groups and proper training has been given to both, immobilisation equipment is more frequently used in radiotherapy than in diagnostics.) The mean displacement over all diagnostic CT scans is 4.95±2.96 mm compared with 5.95±3.42 mm from PET/CT. The rotations give a similar picture, with 1.82±1.13° for diagnostic CT and 2.03±1.43° for PET/CT. To test whether the 15–20% larger errors seen from the PET/CT data are statistically significant, a non-parametric Wilcoxon signed-rank test for paired samples was applied and failed to meet significance for either translation or rotation (p=0.45 and p=0.99, respectively). From this analysis it can be concluded that the differing levels of staff experience with immobilisation did not lead to statistically different set-up accuracy for this patient group.
The main aim of this study was to quantify the accuracy of the thermoplastic shell system when used over extended periods of time. To achieve this, the data were separated into five time points: baseline, post chemotherapy, week 5 of radiotherapy, and 3 and 6 months post radiotherapy (results are shown in Figure 2). No clear trend indicating increasing error with time is apparent from inspection of Figure 2, which is in agreement with the results of non-parametric Friedman tests, which fail to find significant differences between any pair of time points for translation or rotation. Similar analysis was performed grouping all registrations by patient and looking for patient-dependent differences in set-up accuracy (Figure 3). It is interesting to note that the patients with the largest systematic and random components of set-up error (Patients 2, 4, 7 and 8) all had images acquired without the immobilisation being fully clipped down, although the analysis showed no overall statistically significant difference in set-up accuracy between patients.
The accuracy of the point-based matching method used to align the immobilisation shell base-plates can be assessed using the mean residual distance after registration between the four pairs of fiducial points selected. For 10 repeats of the entire point-based matching procedure, an overall uncertainty of 0.3±0.1 mm (1 SD) and 0.4±0.2° (1 SD) was recorded. Errors in alignment of the base-plate are therefore an order of magnitude smaller than the total registration errors seen, so they do not significantly bias the results of this study. The cross-correlation algorithm also yielded highly consistent results for the anatomy-based matching, with SDs of 0.2 mm and 0.7° for 10 repeats. After manual adjustment, however, the overall accuracy was more limited at 3.0±3.0 mm (1 SD) and 0.2±0.3° (1 SD). Significant manual adjustment was made in 55% of cases, with mean translations of x=1.5 mm, y=2.0 mm and z=1.6 mm and mean rotations of θx=1.0°, θy=0.6° and θz=0.3° being applied. Breaking down the mean vector uncertainty in overall repeatability accuracy into components yields x=0.5 mm, y=0.3 mm and z=3.0 mm, and shows that registration accuracy is limited in the z-direction by the CT slice thickness (2–5 mm), and the overall vector uncertainty is dominated by this. This is to be expected when comparing a single pair of images, since registration is performed to the nearest CT slice; however, it should be noted that the overall registration accuracy (see section Overall accuracy) is averaged over eight different pairs of images, each with arbitrary CT slice positions, so this effect is not seen to dominate.
It should be noted that by using diagnostic images only (rather than treatment time CBCT) no correction for set-up error will have been made to bias the measurement of the intrinsic accuracy of the shell. It is also important to note that, because of this, the figures quoted for set-up accuracy in this work relate only to the accuracy of the thermoplastic immobilisation shell system without further online correction, and are therefore not representative of radiotherapy set-up accuracy typically achieved during treatment. However, these values are useful when interpreting images of response to therapy, particularly as a guide to the accuracy with which features in functional images can be localised.
When comparing the results of the thermoplastic shell registration accuracy with published values, the estimate of the “ideal” accuracy of 4.8±2.1 mm (1 SD) and 1.7±0.9° (1 SD), derived from data in which all patients are properly clipped down in the mask, is the most appropriate to use.
Early studies of registration accuracy using laser alignment of the shell and anteroposterior and lateral radiographs tend to produce lower estimates of set-up error; for example, Thornton et al  estimated a value of 2.5±1.4 mm. This may be partly because three-dimensional vector displacements are underestimated from two-dimensional projections. Boda-Heggemann et al  obtained values from daily CBCT images of 5.86±2.94 mm and 1.26±3.01° when they used bone matching restricted to the neck region, indicating a close agreement with the results of our study. Interestingly, they also note the SDs in the individual component directions: σx=2.07, σy=2.41 and σz=4.79, clearly showing a larger uncertainty in z, which may in part be due to slice thickness in the planning CT. Although greater observer variability was seen in z for our study, the overall SDs in set-up error were not significantly different in the component directions. In the study of Rotondo et al , which used weekly CT scans acquired using the same CT simulator as for planning, overall errors are given for groups of anatomical landmarks in the “upper head and neck” (3.1±1.6 mm), which is more limited than the region covered by our study, and “lower neck” (8.0±4.5 mm), which includes the acromial extremities or the shoulder specifically excluded from our study. However, they do quote values for the individual structures including the odontoid peg (2.9±1.2 mm) and the C7 spinous process (5.0±3.4 mm), which show values broadly in agreement with our results.
Although all of the data used in this study were obtained by conventional X-ray CT, the results are directly applicable to functional imaging. No significant difference in set-up accuracy was seen between data acquired on a diagnostic CT scanner and PET/CT, and the registration accuracy of PET is intrinsically the same. The physical localisation of the patient, governed by the immobilisation system, in MRI would therefore be expected to fall within exactly the same error range. One note of caution should be added for MRI and functional MRI data: the image-forming process in MR can produce geometrically distorted and shifted images, with misalignments that can depend on the scanner, patient and image sequence. Discussion of distortion in MR is outside the scope of this work (and of what can be controlled by physical immobilisation).
The overall accuracy of registration using a five-point thermoplastic immobilisation shell in the head and neck, calculated using all data available for this study, was 5.5±3.2 mm (1 SD) and 1.9±1.3° (1 SD). By excluding cases when, for good clinical reason, patients were not fully immobilised, the estimated accuracy of the immobilisation system when used optimally is 4.8±2.1 mm (1 SD) and 1.7±0.9° (1 SD). Registration accuracy was not seen to change significantly with time, from patient to patient or from scanner to scanner. This work indicates that use of thermoplastic immobilisation shells to register multimodality image data for long-term follow-up of head and neck cancer patients is feasible and has an accuracy similar to the spatial resolution of the functional imaging.
The authors are grateful to all those who were involved in the head and neck multimodality imaging trial that provided data for this study: Shree Bhide, Marco Borri, Cecelia Caro, Gary Cook, Prasad Dandekar, Kevin Harrington, Dow-Mu Koh, Helen McNair, Dee Mears, Chris Nutting, Angela Riddell and Maria Schmidt.
We acknowledge funding from CR UK grant C46/A10588, the CR UK/EPSRC Cancer Imaging Centre grant C1060/A10334 and the Dutch Cancer Foundation (Koningin Wilhelmina Fonds). The NHS provides funding to the NIHR Biomedical Research Centre.