This study presents the development and evaluation of a novel dosimetric method for assessment of the accuracy of the XLTS tumour tracking system on the CyberKnife machine by comparison to the FTTS system for various target motions and modalities.
4.1 Number of fiducial markers
The data presented in Table 2 showed that when the target was not allowed to rotate, the gamma pass rate for the single fiducial Synchrony-based treatment delivery was 97.82%, comparable to results observed for the static deliveries and exceeding the performance of the 3-fiducial Synchrony-based treatment delivery by 4.08% for the 3%/2mm gamma criteria. This was expected since the target motion was restricted to superior-inferior direction and therefore, the benefit of applying rotational corrections using 3-fiducial markers was irrelevant. In particular, during the FTTS treatment delivery with 3 fiducial markers, the system was unable to exactly match onto the fiducial array in a number of kV images, predominantly for the two most inferior markers which were 16 mm apart. In such cases, the system reported a correlation model error in the range of 3–6 mm, resulting in erroneous data points being added to the Synchrony model. In this study, the operator had to manually remove or rebuild the Synchrony model in the case where the 5 mm correlation model error tolerance was exceeded, which was the PTV margin used in this study. In addition, in these cases, the system would often recommend the application of unnecessary translational and rotational corrections to better match onto the fiducial locations for treatment delivery, however, upon applying such corrections the fiducial matching often worsened. In the case of 1-fiducial tracking for the same phantom setup, the FTTS found it easier to lock onto the top-most fiducial for target-tracking without any of the aforementioned problems and consequently, this was likely a contributing factor to the better dose agreement observed for the 1-fiducial Synchrony-based delivery when compared to the 3-fiducial Synchrony-based delivery with the same setup. Such errors could potentially have arisen due to differences in the relative positions of the markers within the fiducial array during delivery when compared to the treatment plan. In this study, however, fiducial drifting was not possible, and the phantom alignment was carefully carried out according to clinical protocols prior to treatment delivery.
When the target was allowed to rotate, the dose agreement for the 3%/2mm gamma criteria when tracking with only a single fiducial marker was degraded from 97.82–91.24%. This result is expected as when tracking with only a single fiducial marker the system is unable to account for the rotational motion of the target during delivery. Notably, however, the 3-fiducial Synchrony treatment delivery for this phantom setup could not be completed. This occurred as the rotational corrections required by the Synchrony system to accurately match onto the 3-fiducial array were observed to exceed the tolerances set out by the Synchrony system at various points throughout the respiratory cycle, preventing the treatment from being completed. This result has some notable connotations, as the 1-fiducal treatment delivery was able to proceed unhindered, with the reported correlation model error never approaching a level which would raise any concerns to the operator, suggesting that by using 3 fiducials to facilitate target tracking, delivery errors which might go unnoticed with 1-fiducial tracking could be mitigated. This explains the degradation observed in the gamma comparison result for the 1-fiducial Synchrony treatment delivery, as despite the significant degree of target rotation, the delivery was not interrupted, resulting in deviations between the delivered radiation dose when compared to the TPS prediction.
It was noted that with larger target rotations, the dose delivery accuracy afforded by tracking with 3 fiducial markers was likely to exceed what was achievable using a single fiducial marker for target tracking. Despite this, in the absence of target rotation, for the target and motion considered in this study, a comparable accuracy was observed for the 1 and 3 fiducial tracking modalities. Consequently, based on results of this work it is recommended that for minimal target rotations where the advantages of tracking with 3 fiducial markers are insignificant, the option of employing tracking with a single gold seed be considered, as the benefits associated with 3-fiducial tracking may not outweigh the detriment associated with the insertion of the additional 2 fiducial markers in such cases. To confirm this hypothesis, further deliveries covering a range of target geometries, motions and rotations would need to be considered, as in this study only a single patient model was observed. In selected patients at a particularly high risk of pneumothorax, tumour rotation can be analysed prior to insertion of a fiducial marker with a 4D respiration correlated CT scan, which would allow an assessment of tumour rotation, followed by an informed decision on the most suitable target tracking modality based on the degree of target rotation. However, other advantages of 3-fiducial tracking must also be considered, including the ability to tell if one fiducial marker has migrated relative to the other fiducials in the planning CT scan.
Other authors, such as Subedi et al. (2015) have recommended against treating patients with a single fiducial marker. Notably, the authors made this recommendation on the basis that when fewer fiducials were used for target-tracking, the system was found to report a higher confidence when falsely locking onto image noise while the use of greater number of fiducials led to increased precision of targeting for true locks . Despite this, to induce false locks, the authors intentionally degraded the image quality so that they could assess the effect of image quality on targeting accuracy . Consequently, when clinical protocols were followed such that the image quality was optimized for visualisation of the fiducial markers, false locks onto image noise were not expected to have a significant effect on the targeting accuracy and were not observed during the study when target-tracking was employed with a single fiducial marker.
4.2 XLTS 2-view vs. 1-view tracking
All of the tested XLTS-based deliveries in this study were performed using the phantom setup which allowed for target rotation. From the results presented in Table 2, the XLTS 1-view Synchrony treatment delivery showed better dose agreement than the XLTS 2-view Synchrony by 3.44% for the gamma criteria of 3%/2mm, which is a surprising result, as one would expect higher accuracy in target localisation and consequently higher dose delivery accuracy when the target could be adequately visualised and tracked by both of the X-ray imagers,. Despite the XLTS 1-view demonstrating comparable accuracy to the XLTS 2-view, the latter would be preferable, as in clinical practice a smaller PTV expansion margin is used for the XLTS 2-view compared to the XLTS 1-view. The result may be due to differences in the targeting accuracy between the two X-ray imagers, which may have occurred due to the different imaging parameters which were optimised for each of the X-ray images during treatment simulation. During treatment simulation, 8 images were acquired in each of the X-ray imagers to ensure adequate visualisation of the tumour and while the tumour was adequately visualised in all 8 images with one camera, the target was accurately localised in only 7/8 (87.5%) of the images acquired with the second Synchrony camera. However, as only 1 image failed the simulation, the 2-view treatment delivery was able to proceed. Consequently, during delivery as an XLTS 2-view, it is possible that for certain repeatable positions in the respiratory cycle, the target may not have been accurately localised by the X-ray imager which exhibited poorer simulation results, leading to the addition of erroneous points into the Synchrony model, which may go unnoticed by the operator, degrading the accuracy of dose delivery. In particular, when compared to the FTTS, the XLTS was less accurate in identifying the target position when the most inferior half of the target was obscured by the overlying bony rib structure, which occurred predominantly in only one of the X-ray imagers. Nakayama et al. calculated the correlation and prediction model uncertainties associated with the XLTS 2-view and 1-view treatment deliveries . They found no differences in error between the XLTS 2-view and 1-view tracking modalities; however, it should be noted that such analysis performed through the CyberKnife log-files cannot provide an independent measure of the dose-delivery accuracy, since log-files are generated by the system itself .
4.3 Accuracy of tracking with XLTS compared with FTTS
With the exception of the XLTS 0-View treatment delivery, this study showed that the accuracy of the XLTS was comparable to, and even higher than FTTS for treatments planned and delivered using the same phantom setup. As all XLTS deliveries were performed using the phantom with target rotation, the XLTS has demonstrated the ability to deliver the desired treatment planning dose distribution with a high degree of geometrical accuracy for an irregular patient breathing pattern in the presence of target rotation under realistic treatment conditions. This result is in agreement with findings by Jung et al. (2015), who concluded that the segmentation accuracy of the XLTS was comparable to that of the FTTS through analysis of the log-files generated by the CyberKnife system during treatments delivered to a lung phantom utilizing 1 fiducial marker for target tracking in addition to the XLTS 2-view target tracking modality . Jung J. et al. also performed measurements of the 2-D dose distributions through the central plane of the lung for each of these treatment deliveries. The dose distributions measured for fiducial-based target-tracking utilizing a single gold seed were compared against the corresponding XLTS 2-view treatment delivery and the average gamma pass rate for the 3%/3mm, 2%/2mm and 1%/1mm criteria were found to be 100%, 99.6% and 86.8%, respectively . Despite the apparent excellent results observed in this study, it should be noted that Jung et al. (2015) conducted target-tracking utilizing only a single gold seed with a phantom setup which did not include the spine. Therefore, spine alignment would not have been possible, and the authors would have been unable to apply rotational corrections prior to or throughout the course of treatment delivery. As a result, it would have been necessary for the authors to exactly recreate the phantom setup for treatment delivery without the aid of a spine setup for phantom alignment and a number of assumptions would have to be made about the phantom setup and alignment, which would not be possible in a clinical environment. In their study, treatments were delivered using an isocentric beam delivery technique with a fixed 30 mm collimator, which is an oversimplified beam geometry and not clinically used, as it would not be possible to achieve sufficient target coverage except for the case where the target was perfectly spherical (or close to it).
4.4 Dosimetric comparison
The sagittal dose distributions in Figs. 2 to 4 showed that for all Synchrony-based deliveries, higher doses were delivered in the high-dose region, in some cases in excess of 100 cGy higher than the treatment plan, which was a contributing factor to the dose delivery error for all treatment deliveries. To ensure that the reason was not the film scanner systeman alternate film scanner was used which provided similar results. The other potential cause would be the QUASAR respiratory motion phantom which has a thickness of only 120 mm, and consequently, as lung insert moves superiorly and inferiorly it is possible that the most superior and inferior beams which traverse through the phantom in the treatment plan may instead pass through regions of air above and below the insert during treatment delivery. This would result in decreased attenuation and consequently higher dose delivered to the PTV. This effect would have been present for both XLTS and FTTS based treatment deliveries, but would not affect real patients.
4.5 Effect of Monte-Carlo dose calculation uncertainty
The average gamma pass rate of all three dose calculations was found to be 97.68% with a standard deviation of only 0.13%. Therefore, the impact of the variability in the Monte-Carlo dose calculation algorithm on the resulting gamma pass rates was small and unlikely to have any significant effect on the results.