In our study, we observed significant inter-fractional tumor volume reduction with the right lung volume expanding daily. Although the adaptive plan was selected for every RT session because of superior dose conformity and lower dose to the spinal cord, we became aware of three main problems as the volumes of the tumor and the right lung changed. First, we examined the sCT after each treatment session and found that the geometry of sCT became increasingly deviated from CBCT over time. Second, over time we became less confident in the accuracy of the online adaptive plan evaluation because the sCT resembled the initial simulation CT rather than the daily CBCT. Third, the timing of re-simulation was difficult to determine, as the target volume was reduced every day compared with the previous treatment session.
To address the problems, we retrospectively evaluated the treatment plan for each fraction. First, we overrode the sCT density of the dissipated tumor and realized that the DVH of the initial treatment plan was no longer applied to the adaptive plan. In fact, no single DVH is representative of the actual dose-volume relationship of the delivered adaptive plans for the entire treatment course. Thus, we plotted the mean, minimal, and maximal DVH for the target volumes and the right lung for the adaptive plans with and without override, which demonstrated consistent underestimation of the dose distribution for the target and the OAR (the right lung). When examining the DVHs carefully, we can see that all four graphs in Fig 3 underestimated both Dmean and Dmax, whereas the latter was underestimated to an even greater extent (Table 2).
The anatomical location of the most significantly underestimated dose was near the location of the most significant inter-fractional tumor volume dissipation (Fig 4) and was on the path of the beam arrangement. This is understandable because the dose underestimation resulted from a change in dose attenuation as the tumor volume shrank. As the tumor dissipated, the attenuation on the beam path decreased, leading to an increased dose distribution in the tissues along the beam path.
Based on the above findings, for clinical cases receiving oART with significant inter-fractional tumor volume reduction, we concluded that:
- Both Dmean and Dmax could be underestimated in the on-couch adaptive plans, with the DVH curve left-shifted compared to the actual delivered dose.
- The location of hotspots and the volume of hot area shown on-couch during adaptive plan evaluation may be displaced compared to the actual delivered dose.
As tools for personalized radiotherapy develop in recent years, use of online ART is becoming increasingly common in clinical settings. Nonetheless, new technologies can also introduce unexpected errors, and this study shows that radiation dose generated based on sCT should be evaluated with caution. Both CBCT-based and MRI (magnetic resonance imaging)-based oART rely on sCT for dose calculation. However, the suitability of sCT generation has been debated since oART was clinically applied15–18. This debate results from the fact that sCT is an image generated through deformation algorithm, and thus may not accurately reflect the real anatomical structure which the CBCT presents. A recent article regarding its commissioning and dose uncertainty was published, which simulated common clinical scenarios including weight loss and weight gain, target inter-fraction displacement, and gas changes. For the gas change scenario, the researchers observed that if no gas was present during CT simulation, no matter how much gas was present during treatment, there will be no gas in the sCT. They also observed that large changes in gas were not deformed accurately in the sCT. The difference in calculated and measured point dose they observed for the gas change scenario was around 2% to 4% for scheduled and adapted treatment plans14. On the contrary, our clinical scenario was tumor shrinkage, with tumor tissue density present at the time of CT simulation, but later dissipated during treatment. We also observed dose difference in treatment plans with and without density override in this scenario. However, in contrast to gas changes that can be overridden with tissue density during treatment planning, the possible tumor volume changes cannot be predicted and thus is impossible to override at the time of treatment planning. Thus, for now, we can only rely on real-world clinical cases to observe the impact of the target volume change on dose distribution.
The lung is a specific tumor site that often requires re-simulation and adaptation. Pleural effusion, atelectasis, tumor regression, and tumor displacement are all common indications for ART19,20. One study reported an average of 38% GTV shrinkage in NSCLC patients receiving definitive RT19. A published study evaluated the accuracy of auto-contouring and the resulting dosimetry in lung oART, but it did not evaluate the accuracy of sCT generation nor its dosimetric impact8. Another published study assessed the suitability of deformable image registration (DIR) software by calculating and comparing the dose of planning CT and sCT for locally advanced head and neck cancer17. This study found that when patient’s contour changes were over 1 cm, the DIR software could not correctly deform the soft tissue. Other studies also reported that the DIR image quality struggled when large changes in volumes were present21–24.
The current study echoes and complements previous research literature. Moreover, it is of clinical significance for oART users, especially for physicians and physicists treating the lungs. The observations in this study are practical and applicable to oART for pleural metastasis and mesothelioma. The limitations of DIR-generated sCT should be kept in mind during oART treatment delivery with large changes in tumor volume.