This thesis focused on quantifying the benefit of daily replanning for lung cancer patients and characterizing the tradeoffs of adaptive benefit with replanning frequency. In addition to these main objectives, a software tool was created to facilitate an adaptive paradigm by
extending data objects and evaluation methods in the temporal domain. This provided a platform for conducting the research, and useful methods for evaluating planning data.
The goal of adaptive therapy is to improve targeting and spare normal tissues by
dynamically modifying treatment in response to observed variation. Reductions in normal-tissue dose allow for target increases while maintaining acceptable levels of toxicity. Dose escalation in lung cancer is desirable as it increases the probability of local control.
Various studies have demonstrated dose sparing and associated escalation for an adaptive approach in lung cancer, and suggest an increased benefit with greater adaptive frequency; however, previous studies implement at most weekly replanning, leaving questions regarding the full potential of adaptive therapy.
Daily adaptation requires images for each replanning time-point to provide up-to-date anatomic information and a basis for dose calculation. In this work synthetic CT images were generated from an existing set of weekly images for a cohort of lung cancer patients using PCA. A novel method was implemented in the sampling process which preserved temporal trends such as tumor regression in the model and resulted in a set of image mappings that were used to map the first image in the weekly series. The use of synthetic images was advantageous because images and contours associated with a given time-point were based on a common mapping to the initial image resulting in consistency between images, contours, and dose volumes; additionally,
extents of designated sub-clinical disease could be accurately tracked throughout treatment avoiding ambiguities regarding adequate target volumes for replanning time-points.
Daily replanning resulted in statistically significant dose decreases for all risk structures considered in this work. An average decrease in mean lung dose of 5% allowed average
increases in tumor dose of 441 cGy when escalating to an isotoxic criteria, and escalation of approximately 17 Gy was achieved for a single patient. Cord tolerances were not exceeded for any escalation in target dose. These values were slightly less than those observed by other authors implementing lesser amounts of replanning frequency. In part the discrepancy may be a result of volumes to which dose was escalated. In this work dose was escalated to the whole of the CTV as opposed to boosting dose to the residual gross disease.
Additional planning studies were carried out that simulated a single adaptation at mid treatment and weekly replanning. Sequential comparison of simulations from the non-adaptive to the full-adaptive revealed incremental reductions that were statistically significant for both mean lung dose and mean esophageal dose revealing an increase in adaptive benefit with each increase in replanning frequency. Interestingly, the magnitude of benefit decreased as planning frequency increased with an average of 60% of mean lung dose reductions associated with daily replanning being achieved after a single mid-treatment replan, and 88% being realized after weekly replanning. Understanding tradeoffs between replanning frequency and adaptive benefit are important because resources are limited and replanning is currently expensive in terms of workload. Assuming that each instance of replanning is equally costly and that a given patient follows the averages stated above, four times the workload of weekly planning (i.e. 28 additional replans) would need to be expended to achieve the final 12% of potential benefit. In addition to boosting CTV as opposed to primary tumor, the trend of decreasing returns may help explain
discrepancies between reported values of adaptive benefit in this thesis and the work reported by others.
A final effort was made to explore potential methods of identifying patients that may benefit from an adaptive approach. Correlations between reductions in mean lung dose and a variety of simple descriptors were investigated including: nodal status, initial PTV volume, and absolute change in PTV; of these, only the latter was found to be correlated with adaptive benefit. Relationships between patterns of target volume regression and adaptive benefit were
investigated, noting that some irregularity (e.g. patterns of target volume regression followed by a trend of volume increase) seemed to be associated with patients that exhibited no adaptive benefit which suggests that temporal patterns may be important; however, sample size was not large enough to be conclusive.
Indicators of adaptive benefit are important to select which patients warrant allocation of additional resources inherent in replanning, and future work should be dedicated to this end. The software tool described in chapter 2 enables analysis of multiple signals that could be used in such efforts including temporal course of volume, mass, or centroid positions.
The major contributions of this work are 1) it identifies a need for tools suited to an adaptive paradigm and gives an example of such tools; 2) it introduces a novel method for sampling of synthetic geometries that preserves temporal anatomic trends providing a basis for adaptive studies; 3) demonstrates the benefits associated with daily adaptation; and 4) it
characterizes the tradeoff between adaptive frequency and workload. These developments: provide insight into the value of adaptive radiation therapy for lung cancer, help inform decisions regarding it’s implementation, and provide a basis for future studies.