Tumour Volume Dynamics: Quantification and Prediction in Non-Small Cell Lung Cancer Radiation Therapy

Abstract

Radiation therapy (RT) is indicated in the management of all stages of non-small cell lung cancer. During radical RT delivery, on treatment image guidance results in the routine acquisition of cone beam CT scans on a daily or weekly basis. These scans are utilised in the geometric verification of the treatment but have the potential to assess early tumour response.

This thesis investigates longitudinal tumour volume kinetics in patients treated with RT for NSCLC. The existing literature on this topic is conflicting and is limited by the retrospective nature of the evaluation of tumour volume changes.

Firstly, this thesis examined the feasibility of generating large scale datasets of tumour volume measurements using a semi-automated delineation tool. Study 1 tested this approach in high quality imaging acquired for research purposes and found the volumes generated to be acceptable based on geometric and dosimetric comparisons to manual contours. Study 2 piloted the methodology validated in Study 1 on cone beam CT images acquired routinely during RT and found the approach to be feasible and efficient in 66.6% of the population.

Secondly the Proliferation Saturation Index (PSI) mathematical model to predict tumour response to RT was examined in this population. Prediction of tumour response to RT could select patients for tailored adaptive strategies in the clinic. Study 3 trained and tested the model in a cohort of 164 patients across four dose fractionation schedules. The PSI model predicted tumour volume regression during RT with a high degree of accuracy, using early tumour response dynamics. Comparison of the measured vs predicted volumes resulted in R2 values ranging from 0.94-0.97, and Pearson Correlation Coefficients from 0.97-0.98 in the four groups. Study 4 externally validated the findings of Study 3 in a cohort of 71 patients treated with definitive RT alone. Furthermore, it examined the relationship between regression with overall survival, identifying in this cohort that there was a negative relationship between these factors. Study 5 determined that the performance of the PSI model using only pre-treatment data was strong (R2=0.86 and Pearson Correlation Coefficient=0.93) when comparing measured and predicted volumes. An in-silico trial of 6 dose fractionation schedules revealed that the impact of prescription on tumour regression at the end of RT varied widely between individuals.