Accumulating delivered dose to the rectum to improve toxicity prediction in prostate radiotherapy Leila E. A. Shelley Gastrointestinal (GI) toxicity is a clinical issue suffered by up to 22% of prostate cancer radiotherapy patients. However, the relationship between radiation dose and toxicity is generally poorly understood. In prostate radiotherapy, the rectum is a dose-limiting structure to which treatment planning dose constraints are applied to minimise the risk of toxicity. Current normal tissue complication probability (NTCP) models are based on planned dose data and do not consider the effects of organ motion on true delivered dose. The VoxTox research programme has developed automated solutions for segmentation and dose calculation of the rectum for prostate cancer patients being treated with helical TomoTherapy. Daily image guidance scans, acquired primarily for the purposes of positional verification, are exploited by extracting quantitative information to facilitate the calculation of daily delivered and total accumulated dose to the rectum. Prospectively collected toxicity data at 2 years post-treatment were available for 295 patients across two separate cohorts. In this thesis, the hypothesis being tested is that delivered dose is a better predictor of rectal toxicity than planned dose in prostate radiotherapy. The research has successfully demonstrated, for the first time, that delivered dose produces stronger associations with rectal bleeding and proctitis than planned dose. Analysis was performed using dose surface maps (DSMs) of the rectal wall, allowing spatial aspects of dose to be retained during accumulation. A subsequent analysis on a separate cohort also found stronger links between delivered dose and GI toxicity, stool frequency, and bowel bother, in addition to rectal bleeding and proctitis. Biomechanical finite element (FE) modelling was introduced to provide a more anatomically plausible tool for dose accumulation and allowed more accurate tracking of dose at the voxel level. A sensitivity analysis was conducted which explored the effect of simulated rectal motion on dose, and corresponding change in NTCP. For VoxTox patient dose-toxicity analysis, further dose parameterisation approaches were explored in order to consider the increased resolution of information available. Voxel-based rectal subregions at risk (SRRs) were identified using geometric and statistical approaches. In general, discriminative power improved with FE modelling for both planned and accumulated delivered dose, and associations between accumulated dose and toxicity were strengthened by voxel-based subregion analysis. Multivariate NTCP models were constructed for 12 different toxicity endpoints based on planned and accumulated dose parameters. Model performance was compared between analysis approaches, and models were tested on a validation dataset. In general, FE-based dose models performed best, although the optimal dose parameter selected within the model varied with toxicity endpoint. Overall, results suggest that there is an advantage to incorporating delivered dose into NTCP modelling. However, the differences between planned and accumulated dose can be subtle. Meaningful parameterisation of accumulated dose needs careful consideration, as traditional methods for quantifying planned dose may not be directly transferable. Voxel-based analysis techniques are recommended in order to accurately preserve and register spatial dose information, and have been shown to improve the strength of dose-toxicity associations. Further research into quantifying voxel level dose distributions is encouraged. It is anticipated that the novel scientific contributions presented within this thesis will prove valuable for future development of clinical decision-making tools for adaptive radiotherapy, with the ultimate aim of reducing the incidence of radiation-induced toxicity for prostate cancer radiotherapy patients.
