Quantifying the trigger level of the vacuum surveillance system of the Gamma-Knife eXtend™ positioning system and evaluating the potential impact on dose delivery

Abstract

Objective

This work evaluates the precision and characteristics of the trigger level of the vacuum surveillance system of eXtend™ on Gamma Knife Perfexion and the effect of the potential displacement on the dose distribution.

Methods and Materials

A total of 20 individually moulded mouthpieces based on human dental models were used to measure translational shift and rotation until the vacuum surveillance of eXtend interrupted the irradiation. The positional accuracy of the movement was 0.01 mm using a computerized numerically controlled positioning system. Rotation was introduced by peripheral pressure in superior or inferior direction on the mould and was measured with a digital inclinometer. In 10 patients with a large brain metastasis the effect of a potential displacement of the centre of the target was recalculated. Two out of the 10 targets were located near the optic nerve and chiasm. In addition, the potential displacement of the chiasm and the optic nerve due to rotation based on their distance to the centre of rotation (COR, the mouthpiece) was evaluated. For the recalculation of the delivered dose after a displacement, GammaPlan Software® was used. Dose distribution with displacement between ±0.1 mm and ±2.0 mm was simulated for ten targets with volumes between 7.7 cm³ and 19.3 cm³. Shifts were applied to the X (lateral), Y (ventro-dorsal) and Z axis (superior-inferior). Dwell time was kept constant within <0.5% total time and 1% individual dwell time. Evaluated parameters were “minimum dose” (D_(99%)), “coverage” and “Paddick Conformity Index” (PCI).

Results

The vacuum between mouthpiece and dental model was broken by a mean translation of the mouthpiece of 0.15 mm (SD ±0.05 mm, range 0.05-0.29 mm) or a mean rotation of 0.33º (SD ±0.15º, range 0.05-1.0º). Rotation caused by pressure in the inferior direction was higher with 0.42º (SD ±0.16º) compared to rotation due to pressure in the superior direction with 0.21º (SD ±0.09º). The mean rotation would produce a mean displacement of 0.64 mm for the target, 0.39 mm for the chiasm and 0.33 mm for the optic nerve, for the group of patients studied. The biggest effect of positional deviation was seen in the minimum dose to the target, with 2.5% reduction for a 0.5 mm and 5% for a 1.0 mm shift in the Z-direction respectively.

Conclusions

The trigger level is sensitive and reproducible. A sudden movement by the patient, such as an attempt to get out or correct an uncomfortable position would interrupt irradiation immediately. The dominant effect for a displacement is due to rotation. The resulting displacement depends on the distance between target and COR (magnitude) and the location (direction). Therefore, lesions near chiasm or brainstem, which are closer to the COR, are less affected from a rotation than peripheral targets.

1. Introduction

To treat brain tumours, fractionated whole brain radiotherapy (WBRT) is often applied but provides limited local tumour control and induces side effects like fatigue or reduced cognitive function which leads to reduced quality of life (QoL).²⁰ Stereotactic radiosurgery (SRS) can reduce several of these side effects by focusing the dose on the target and sparing normal brain tissue.^7,22

Gamma Knife (Elekta Instruments, AB, Sweden) (GK) is an efficient and cost effective treatment unit dedicated to SRS of solid brain tumours.^19,34 For the clinical success of intra-cranial SRS, high positional accuracy is required to place the dose gradient exactly on the border between target and normal tissue (NT). This was originally achieved by a direct, rigid, minimally invasive connection between the skull bone and the G-frame where the G-frame is screwed directly to the skull bone. The G-frame is then connected and aligned to the GK coordinate system. This direct connection between imaging and treatment unit coordinate system is unique to GK and eliminates the positioning error present in other treatment units, thus achieving submillimetre accuracy.^14,31

A high dose per fraction limits the treatable lesion size (<3 cm in diameter) and proximity to organs at risk (OAR).^2,12,25 This can be overcome by fractionating the treatment.^{3,11,12,27,42} Fractionation or hypo-fractionation results in a lower dose per fraction and allows the NT in the gradient region to repair sub-lethal injury between the fractions thus reducing toxicity.^{1,13,18,33,39}

There is no universally agreed definition of exactly what SRS is. The international Gamma Knife society has recently published a standardization report (Torrens et al ³⁸) which suggests regarding the treatment unit as “a system for stereotactic guidance of radiation output with submillimetre accuracy”. Fractionated radiosurgery is expected to achieve that specification (with an arbitrary limitation of five fractions).

The spacious design of the Leksell Gamma Knife Perfexion™ (Elekta Instruments, AB, Sweden) model (GKP) led to the introduction of a new non-invasive repositioning system known as eXtend™ (Elekta Instruments, AB, Sweden). This is based on a head rest with an individually formed vacuum cushion and a vacuum assisted mouthpiece. As a positional verification during the treatment, the vacuum level of the mouthpiece is continuously monitored and the treatment interrupted as soon as a movement-induced drop of the vacuum is detected.

The key part of this vacuum surveillance is a spacer that is placed on top of the mouthpiece towards the upper palate to create a cavity. While the mouthpiece is in close contact with the upper palate the cavity is sealed and a vacuum can be established. Any movement of the patient would create a gap and the vacuum would be lost. A good mould fits tightly and allows setting a vacuum level indication of 50 to 60% (deviation relative to ambient atmospheric pressure). However, for patient comfort in clinical practice, it is recommended to set the level between 30 and 40%.

The vacuum surveillance software of the patient control unit constantly monitors the vacuum level and pauses the treatment as soon as the vacuum drops by 10% of the initially set level. Any movement or displacement until this trigger level is reached causes a deviation of the delivered dose from the calculated one.

Uncertainties of the initial patient setup associated with the use of eXtend™ have been assessed by Ruschin et al ³⁵ and the whole chain accuracy of eXtend™ on GKP was tested by Ma et al.²³

Ruschin et al ³⁵ mounted the eXtend™ system on an Elekta Linac and treated intracranial tumours with a conventional fractionation scheme. Position accuracy was verified using a cone beam CT (CBCT) scan pre and post treatment. The deviation measured by the CBCT was compared to that measured with the repositioning check tool (RCT) of eXtend™. The mean deviation was 1.3 mm with the largest contribution in the superior-inferior direction. The largest mean rotation was 0.6° in pitch. The mean intra-fraction motion (pre and post treatment) was <0.4 mm. In one case the post treatment 3D vector was 3.9 mm. The authors assumed that “ … the patient attempted to turn his or her head when the therapist entered the room … “.³⁵ This is a likely explanation for the large displacement, however, it leaves the question open: at which point does the vacuum surveillance system pause the treatment?

Schlesinger et al ³⁷ reported a series of ten patients undergoing fractionated treatment by GKP. In one case, the treatment was paused by the vacuum surveillance system. After re-set up the treatment was continued without further problems. However, they mention the importance of “further work to characterize the sensitivity of the vacuum monitoring system to detect patient motion.”

To date, to the authors’ knowledge, there has been no report on the trigger level of the vacuum surveillance system which defines the intra-fraction accuracy. This work investigated the trigger level and characteristics of the vacuum surveillance system of eXtend™ and what effect a displacement would have on the dose distribution in the target.

2. Material and Methods

2.1. Dental model, mouthpiece moulding and fixation

For accurate measurements clearly defined surfaces and reference points are required therefore dental models have been used for the measurements. Twenty moulds from dental plaster models of the upper human jaw were produced applying the procedure described by Ruschin et al.³⁵ The dental models were a random selection representing a range of dental conditions and geometric shape. The solid flat surfaces on the model allowed us to define a reproducible reference point for the measurements. Examples are shown in Figure 1.

Figure 1.

some examples of the dental models used for the measurements.

To fit the mouthpiece to the patient in clinical practice a shift in the Z-direction (superior/inferior) and a rotation around the X- and Y- axes is required. These movements are the most likely ones to happen during treatment and are those quantified in this work. Shift in X- and Y-axes as well as rotation around the Z-axis are generally prevented by the large contact area along the dental circumference (Figure 2).

Figure 2.

Movement in X and Y direction were considered as negligible due to the large contact area visualised by the solid line.

2.2. Translational measurements

To measure translational movement the mouthpiece was fitted to a holder with a pole and then mounted with a collet on a computer numerically controlled (CNC) machine (SM2000 milling machine) with positional accuracy of 0.01 mm. The dental model was fixed on the CNC table. For best alignment the mouthpiece was formed in the measurement position (Figure 3). As a surrogate for the mucosa, Vaseline gel was applied to the surface of the dental model.

Figure 3.

the eXtend™ mouthpiece with a pole attached was fitted to the collet of the computer numerically controlled (CNC) machine for precise positioning.

Dental model and mouthpiece were put together in close contact. Then the vacuum level was set and the collet from the CNC machine tightened. This position was set to zero. The mouthpiece was then moved away from the dental model in steps of 0.05 mm and the vacuum level was recorded at each position until it had dropped to zero indication which equals atmospheric pressure. The position where the vacuum level had dropped by 10% of the original setting was recorded as the trigger distance. This was repeated for each mould 30 times at different vacuum levels (10 times at 30% vacuum level, 15 times at 35% the recommended vacuum level and 5 times at 40% vacuum level). Where a higher (non-clinical) vacuum level was achievable, measurements between 45% and 70% were added.

2.3. Rotation measurement

To measure rotation, the dental model was fixed to a base plate with the mouthpiece and dental model in close contact then the vacuum level was set to the initial level. Five measurements each were performed for vacuum levels in a consecutive series of 35%, 30%, 35%, 40% and 35%. The value of 35%, as the usual clinical level, was measured in between the other levels to confirm reproducibility.

Head rotation around X- and Y-axes may cause a significant change in target position. In order to induce a rotation pressure was applied at the periphery of the mouthpiece. Action points were right, left, front and at the back of the mouthpiece (marked as circled crosses in Figure 3). The pressure was applied in the superior and inferior directions. When the vacuum level was set, pressure at each action point was continuously increased and the rotation measured with a digital inclinometer (Fisco Solatronic, Model EN17, resolution 0.1°). The rotation at which the vacuum level had dropped by 10% was then recorded.

2.3.1 Dosimetric effect of positional deviation on dose distribution

In order to estimate the effect of a displacement on the target dose the potential displacement due to rotation and the associated delivered dose distribution were evaluated.

2.4. Target selection

Plans for ten metastatic lesions with a size between 7.7 cm³ and 19.3 cm³ were selected for dosimetric simulation of displacements. Location varied across the brain including three lesions adjacent to the skull bone and two near an OAR (optic nerve and chiasm). Detailed characteristics are given in Table 1. All patients had been previously treated in a single fraction using the G-frame screwed to the skull for imaging and treatment. This plan was used for recalculation of the dose distribution with a displacement induced.

Table 1.

Characteristics of targets used for calculating the dosimetric effects of displacements

Characteristic	Mean	Range
Tumour volume (cm3)	12.8	7.7 – 19.3
Distance to centre of rotation (mm)	105.3	55.7 – 147.8
Prescribed dose (Gy)	17.4	16 - 18
Prescription Isodose (%)	48.7	45 - 51
Minimum dose (Dose to 99% of volume, D (99%))	17.08	15.6 - 18
Coverage (%)	98.6	97 - 99
Total number of shots	23.1	16 - 32
Paddick conformity index (PCI)	0.873	0.765 – 0.918

2.5. Potential displacement due to rotation for target, chiasm and optic nerve

For a given patient, the potential displacement of the target due to rotation depends on its position relative to the centre of rotation (COR, the centre of the mouthpiece at the hard palate). Ma et al ²³ had the point defined at the connection of the vacuum tube to the mouthpiece. In this work it was defined at two centimetres dorsal from the dentures on the level of the hard palate in the central sagittal slice of the MR image. Figure 2 shows the estimated point.

Rotation influences the target position in two ways: a target rotation and a translational displacement. In this work, the target rotation was ignored because of the near spherical shape of the selected targets.

The potential displacement was evaluated from the distance between the COR to the centre of the target (centre of gravity) for the mean and maximal rotation around X and Y axes, respectively measured in the previous experiment. The displacement was calculated for each axis individually as in Figure 4 and equations 1 to 3 for rotation around the Y axis. The same analysis was made for the chiasm and ON.

Figure 4.

A rotation around the Y axis causes a displacement of a target in the X and Z directions. The magnitude of such a displacement depends on the rotation angle and the distance in the Z direction between target and COR for the magnitude of the displacement in the X direction and the distance in the X direction for the displacement in the Z direction.

$${\mathrm{d}}_{\mathrm{x}}\mathrm{\hspace{0.17em}}\approx \mathrm{\hspace{0.17em}}\mathrm{Z}\mathrm{\hspace{0.17em}}\mathrm{*}\mathrm{\hspace{0.17em}}{\mathrm{TAN(}}_{\mathrm{(Y)}}\mathrm{)}$$ (1)

$${\mathrm{d}}_{\mathrm{Y}}\mathrm{\hspace{0.17em}}\approx \mathrm{\hspace{0.17em}}0$$ (2)

$${\mathrm{d}}_{\mathrm{z}}\mathrm{\hspace{0.17em}}\approx \mathrm{\hspace{0.17em}}\mathrm{X}\mathrm{\hspace{0.17em}}\mathrm{*}\mathrm{\hspace{0.17em}}{\mathrm{TAN(}}_{\mathrm{(Y)}}\mathrm{)}$$ (3)

d_X: displacement in X direction (d_Y and d_Z accordingly)

Z= distance in Z direction between COR and target (Y accordingly)

_(Y)= rotation around Y axis

Impact of positional uncertainties on dose distributions

2.6. Keeping shot time constant

Recalculation of the effect of a displacement requires the plan to be calculated with the actually irradiated shot times, i.e. the times from the original plan. GammaPlan Software® (Elekta Instruments, AB, Sweden) automatically adjusts treatment time for all shots to achieve the prescribed dose in relation to the actual maximum dose. A shift of the shots may change the path length and therefore the dose to the modified point. The shot time is automatically “corrected” for this change. This time modification by GammaPlan was reversed to the original planned time by modifying the prescribed dose and where necessary the individual shot weight. This method allowed us to keep the total treatment time variation ≤0.5% and the individual shot time variation ≤1.0%

2.7. Re-calculating dose distribution after displacement

For the original treatment plan the effect of a potential uncertainty was simulated by shifting the shots in all three axes in positive and negative directions for 18 positions in steps of 0.1 mm up to 0.4 mm and then in steps of 0.2 mm up to 1 mm plus shifts of 1.5 and 2 mm. The shift was performed for one axis at a time.

2.8. Plan evaluation

For plan comparisons three parameters were considered: minimal dose to the target D_(99%), percent target coverage and Paddick Conformity Index (PCI) ³² as a commonly used quality index in GK radiosurgery.

The values of the original plan were set to 100% to normalize the values. Parameters calculated from the shifted plans are given in relation to the original plan.

3. Results

3.1. Accuracy of the Vacuum surveillance

3.1.1. Translational measurements: Shift until trigger level is reached

With the vacuum level set in the clinical range between 30% and 40% the mean shift was 0.15 mm (SD ±0.05 mm, range 0.05-0.29 mm) until the vacuum level dropped by 10% of the initially set value. When the vacuum level was set to a non-clinical level of ≥45% the shift until trigger distance was 0.22 mm (SD ±0.09 mm, range 0.09-0.43 mm).

3.1.2. Rotation measurements: Rotation until trigger level is reached

For the clinically used vacuum level between 30% and 40%, the mean rotation was 0.33° (SD ±0.15º, range 0.05º – 1.0º) until the vacuum level dropped by 10% of the initially set value. Analysis in respect of the direction of the applied force showed twice as large a rotation for an inferior force (pressing the teeth on one side into the mouthpiece) than for a force in the superior direction (a patient ‘pulling out’ of the mouthpiece) from one side; being a mean rotation 0.42º (SD ±0.16º, range 0.05º 1.00º) and a mean rotation 0.21º (SD ±0.09º, range 0.05º 0.60º), respectively.

No dependence on the point of action, right (0.31°, SD ±0.19º), left (0.32º, SD ±0.17º), front (0.33º, SD ±0.19º) or back (0.35º, SD ±0.24º), of the rotational force was observed.

For both, translation and rotation, the results were very reproducible. However, it appears that better moulds where a high maximum vacuum level could be set (better quality mould and/or dental condition) have a higher tendency to have an occasional larger displacement than those with lower maximum vacuum level even when the vacuum level is set to the same value for the measurement.

3.1.3. Resulting displacement due to rotation

The mean distance from the COR to the target was 112 mm (SD ±27.9 mm, range 59.7 mm – 147.8 mm), from the COR to the chiasm was 68 mm (SD ±7.1 mm, range 57.2 mm – 75.4 mm) and from the COR to the optic nerve was 57 mm (SD ±5.2 mm, range 48.8 mm – 68.1 mm). These distances result in a potential displacement for the mean rotation of 0.33° of 0.64 mm, 0.39 mm and 0.33 mm, respectively. For the largest rotation measured, 1.0°, a target displacement of up to 2.56 mm total vector might be possible, for chiasm 1.32 mm and for optic nerve 1.11 mm respectively. Individual results for target and chiasm are shown for mean displacement and one standard deviation in Figure 5.

Figure 5.

potential displacement of the target (A)) and chiasm (B)) for ten patients in the different directions due to mean rotation (0.33°) and one standard deviation added (0.48°) as line.

3.2. Impact of position uncertainty on dose distribution

The main criteria for plan acceptance for brain metastasis are minimum dose D_(99%), coverage and PCI. Figure 6 presents the recalculated values as a function of the displacement as a percentage from their original plan values.

Figure 6.

A) Minimum dose drops faster for Z direction (superior/inferior) shifts than for those in X or Y directions (right/left and ventro/dorsal respectively). This is due to the steeper dose gradient in the Z-direction. Figure B) (coverage change) and C) (PCI) show no axis specific variation. The asymmetry seen in the +Z axis is due to a compromise in two plans in order to keep the dose to an OAR below a certain tolerance level. The shift towards the OAR improved the plan quality parameters which do not take the OAR risk into account. All values are normalised to the original plan parameters. Error bars indicate one standard deviation.

All three parameters show little change in the first 0.5 mm shift. With increasing displacement the curves become steeper and indicate compromises in the plan quality. In all three curves an asymmetry between the right and the left side of the +Z direction can be seen. This is due to the OAR protection in two plans.

The minimum dose D_(99%) in the target changes by less than 2.5% for the first 0.5 mm shift and then starts to drop rapidly with 7.5% reduction for 1mm displacement. While there is no significant difference between X and Y axes, the Z axis shows a steeper drop after about 0.75 mm shift. The standard deviation increases with distance (1 SD indicated in Figure 6 A)).

Coverage in Figure 6 B)), represented as percentages of the original parameter taken from the reference plan, shows less change and drops only by 0.5%. In comparison to the minimum dose D_(99%) the coverage in Z direction changes in a similar way as for a shift in X or Y axis on the left side (inferior shift). On the right side coverage drops less for a shift in +Z direction (superior) than for X or Y shift. This is due to OAR protection in two plans.

PCI changes behave similarly for shifts in all three axes (Figure 6 C)). Differences in the +Z direction are again due to the OAR protection in two plans as seen in minimum dose D_(99%) and coverage changes.

4. Discussion

4.1. Accuracy of the vacuum surveillance

The use of dental models provides accurate surfaces and allows the measurement of not only a static set up but to quantify the submillimetre changes with both shift and rotation that can lead to a treatment interruption. It was possible to evaluate different situations such as simulating a patient “pulling out” of the mouthpiece or to “settling in” to a more comfortable position (rotation with one sided pressure in inferior direction). Assuming that in most cases the patient moves with distinct, sudden motion the treatment surveillance is very sensitive and would pause the treatment before any deviation in irradiation could take place. If the patient gradually moves into a slightly different position then rotation is the most critical aspect which usually results in submillimetre displacement of the target but might cause up to 2.56 mm displacement in one of the selected test cases.

A vacuum can only be established if a tight fit is achieved between patient and mould. Ideally, the positional relation is identical as it was during mould forming. In real life this will never be exactly the case. Assuming the trigger level is activated as soon as the vacuum breaks would mean that any position of the patient where a vacuum can be set has to be within the range between “ideal and perfect” set up and “just before trigger level activation”. Several authors evaluated the set up and intrafraction accuracy of eXtend. If this assumption is correct the setup displacement measured in those studies should be within the intrafraction range evaluated in this work.

The RCT system from eXtend™ measures the distance from a predefined reference point to the surface of the head. No discrimination between shift and rotation is available nor is a displacement due to rotation linked to the target position. For this reason, Ruschin et al ³⁵ mounted the eXtend™ system on a linac and evaluated the difference between RCT indicated displacement and CBCT measured displacement from the CT localiser box to the anatomy. They evaluated twelve patients treated for a total of 333 fractions. For RCT deviations of <1.0 mm per axis they reported a total vector mean deviation of 1.0 mm measured with the RCT compared to 1.3 mm measured with CBCT.

Ruschin et al ³⁵ deliberately chose the most distal measuring points. This is probably the most sensitive indication of displacement in case of a rotation. The biggest displacements were measured in connection with a rotation. This is in agreement with our finding that rotation leads to the largest displacement for distal targets. The mean displacement of 1.3 mm they found was slightly larger than that in our experiment. However, no information is given about target location. Furthermore their results included whole chain uncertainties such as imaging, fusion error or CBCT tolerances, in addition to set up uncertainties whereas our results are exclusively the potential difference between patient and mouthpiece.

The uncertainties for the whole chain accuracy excluding the mouthpiece patient interface was measured by Ma et al ²³ using the centre point of the mouthpiece as reference for an adapted Winston-Lutz-Test ²¹. The measured uncertainty included errors from imaging resolution, target definition and mechanical uncertainties of the frame and holder when mounted on GKP. Patient dependent uncertainties were not included. They reported a mean deviation of 0.55 mm, 0.78 mm and 0.72 mm for X, Y and Z axes. Adding these full chain uncertainties of the system to the measurements from our experiment evaluating the intrafraction uncertainties between the mouthpiece gives patient results in good agreement with Ruschin’s results.

Sayer et al ³⁶ reported a good reproducibility of the setup with eXtend™ while treating four patents with three to four fractions each. All setups were within <1.0 mm radial deviation ranging from 0.33 mm to 0.84 mm mean value. From the same institute, Schlesinger et al ³⁷ reported in more detail on ten patients undergoing fractionated treatment. They compared measurements pre and post irradiation to estimate a potential intrafraction uncertainty and found a mean radial displacement (vector) of 0.64 mm and a mean intrafraction displacement of 0.47 mm. Both experiments are based on RCT measurements only and include potential measurement errors from soft tissue effects (e.g. muscle flex on the side when the bite pressure changes, hair at the top) and do not give direct information about the actual target position. Their work shows that a tighter tolerance than 1.0 mm is generally achievable. However, they too mention the importance to choose the RCT measurement points in a way that a displacement due to rotation can be recognised.

4.2. Strategies to deal with displacement uncertainties due to shift and rotation?

To be sure the delivered dose is covering the target, margins can be applied. A commonly used method is proposed from van Herk et al.⁴⁰ He simulated random set up uncertainties and internal motion of prostate treatment for 36 fractions and derived a formula from those data for a margin to achieve the desired tumour coverage. However, with SRS only four or five fractions are treated. A true randomisation that would blur the dose distribution is therefore not possible. So questions remain as to how to add margins, what size and should margins be evenly applied in all directions? Ma et al ²⁴ demonstrated that even small margins increase the volume of irradiated NT significantly. Therefore it is important to minimize margins.

Particularly interesting for Gamma Knife treatments are lesions near the optic nerve, chiasm and brainstem. Several authors reported good results when treating tumours near the optic nerve, chiasm or orbital targets.^{10,13,16,17,30} To minimize margins in these areas is in particular important. The results of this work revealed that the main factor for a displacement is a rotation. Therefore targets closer to the COR (mouthpiece) have a smaller potential for displacement due to rotation than distal targets. Figure 7 demonstrates the difference in displacement between chiasm and a distal lesion for the same rotation and therefore the same RCT values.

Figure 7.

the effect of rotation depends on the target location in respect to distance and direction to the mouthpiece as the COR. With rotation only possible around X- and Y- axes the target displacement is ventro/dorsal for a lesion above the COR and sup/inf for a lesion in the back of the head. (Note: the COR is defined in the central saggital slice. The images shown here are off centre. The COR in these images is an estimated projection from the true COR.)

The change in coverage and minimum dose D_(99%) is small for 0.5 mm displacement as demonstrated in Figure 6. This is likely due to the slightly larger volume of the treatment dose volume compared to the target. So targets near the chiasm or ON might not need a margin at all.

However, peripheral targets may require 2 mm and more positional uncertainties. This could reduce the minimum dose D_(99%) to 80%. To minimize the margin the potential displacement direction due to the target position could be taken into account. For example patient 3 in Figure 5 A) has a target dorsal in close proximity to the skull. Potential displacement in X and Y direction is less than 0.5 mm but might be more than 1.0 mm in Z direction. Minimizing margins in X and Y direction might spare NT. For patient 7 on the other hand, the margin might be minimized in the Z direction.

It may be noted that image resolution in Z direction (cranio/caudal) is typically 1.5 mm which is the resolution for target outlining in this direction. Furthermore, inter-observer variation may vary in many cases by more than the just one millimetre. The magnitude of the set up uncertainties or intrafractionation movement is about within the range of target delineation accuracy.

Due to the proximity to the mouthpiece, being the COR, lesions near the optic nerve, chiasm and brainstem may be very suitable for fractionation with eXtend™. Several authors reported good results when treating tumours near the optic nerve, chiasm or orbital targets.^{10,13,16,17,30}

4.3. Availability of CBCT on the Gamma Knife

While this work was being completed, Elekta launched the ICON model, which adds a CBCT and a High Definition Motion Manager (HDMM) system to the GKP, intended for use with thermoplastic masks and customised made head cushions. Some authors have demonstrated in the past that patient movement during long treatments can be in millimetre range^15,28,29. With the ICON HDMM system patient position is continuously verified to an accuracy of 0.15 mm with a reflector point on the nose tip of the patient. On the treatment console a tolerance level depending on target and proximity to OAR can be set. As soon as the detected position exceeds this tolerance the treatment is paused until the patient position is again within tolerance.

Thermoplastic masks have been used for SRS for a long time. The accuracy of a such a mask is typically around 3.0 – 3.7 mm ± 2 mm. ^4-6,26 For dedicated SRS treatment, critical points are strengthened or supported (Brainlab mask or GTC system^4,9) and with image guidance, mean accuracies are achieved in the range of 1-2mm. The mask system from ICON is aided by a CBCT for initial set up accuracy and the HDMM.

In a pre-clinical setting Chung et al.⁸ evaluated the accuracy of the HDMM using a mechanically controlled device to position the reflector in various known positions. The mask and HDMM system were used on a linac for four patients, treating 28 fractions in total where the HDMM measurement was verified to pre and post CBCT. The agreement for the pre-set position was 0.1 mm under good conditions, with one difference between CBCT and HDMM of 0.5 mm. The intrafraction motion was up to 3.6 mm, in agreement with previous experience of other thermoplastic mask systems. In their study, no detailed information about direction or rotation is given. As seen in previous work of the same group ³⁵ and in our measurements, rotation is most critical for a potential displacement. A single point cannot detect a rotation. In a mask system the COR might more likely be expected to be at the centre of the head, so the reflector at the nose may overestimate the displacement. On the other hand, a rotation around the nose tip might produce an undetected displacement which could potentially be up to several millimetres. Thus pre and post CBCT seem advisable, but can only document a potential displacement during RT not discriminating when the displacement has occurred.

Using the mask is probably more comfortable for the patient but may increase the uncertainty to potentially several millimetres. This can be compensated by gating the treatment. A threshold can be set in the HDMM system so that dose delivery is paused while the position of the reflector is exceeding the set tolerance. The level of the required threshold might be estimated from our measurements presented in Figure 6 always bearing in mind that a single point cannot detect a rotation and therefore target location should be considered too.

The ICON may also impact on imaging uncertainties. MRI, as used for input to GK planning has recognised distortion issues. Special sequences have been developed to minimize this. With a CBCT on the treatment unit a new workflow might be considered, using the diagnostic MR images and matching the target to a distortion free CT scan. Wangeried at al⁴¹ have compared this workflow with the usual one using the distortion optimized MR. They found a mean difference in the resulting isocentre of 0.47 mm. This comparison was simulated and it is unclear which workflow is more accurate in practice.

4.4. Limitations of the current study

This study was performed on dental models. The results are in good agreement with the literature. However, in particular the simulation of the upper/hard palate as a rigid structure has to be considered with care. Real patients might have some upper palate changes between fractions such as swelling. Whether this would make the system more sensitive (different anatomy, more likely to produce a leak) or less sensitive (more soft tissue, better fit with the mould) remains to be investigated.

Furthermore, the movements tested were each applied individually, a shift or a rotation in one specific direction. In a clinical situation it is likely that multiple movements will occur simultaneously. Again, whether this would increase or decrease the intrafraction trigger distance remains to be investigated.

In Figure 6 the curves on the right side are slightly biased due to the plan modification to protect an OAR in two plans. Keeping the dose tight towards the OAR (ON / chiasm) results in a compromise in coverage which is undone when the plan is “moved” towards the OAR. With a shift towards the OAR, the coverage is increased and the parameters seem to indicate a better plan. The effect on the OAR would need a separate investigation.

Our measurements show the potential uncertainties and demonstrate the significant effect of rotation. As has been mentioned by other authors^23,35,37, there is no direct link between target position as evaluated in this work and the RCT values. A rotation would result in a specific RCT deviation but the target displacement is depending on the target position, with the highest displacement in peripheral lesions.

The recalculation of the dose distribution with displacement assumes that the treatment is interrupted by a sudden and distinct movement and any positional displacement is mainly due to set up error. This is a likely scenario. However, it is also possible that the patient gradually changes its position and the displacement of the different shots varies leading to a “blurred” dose distribution.

5. Conclusion

The vacuum surveillance system on the eXtend system for the GKP is sensitive and reproducible. For lesions close to the COR (mouthpiece, top of hard palate), e.g. near the optic nerve, chiasm or brainstem, the trigger level reliably prevents mistreatment within submillimetre accuracy. Potential movement of the chiasm was <0.6 mm for a rotation of 0.48° (mean rotation observed, plus 1SD).

Lesions which are further away might experience deviations up to one millimetre as a result of a mean rotation, which would reduce the minimum dose to 94% of the original value. A 2 mm deviation is in rare situations possible and would reduce the minimum dose to 80% if it were a deviation in the Z direction. Often the potential displacement is dominant in one direction, dependent on the lesion position, so this direction might be given special attention when considering the addition of margins.