Gross tumour volume (GTV) changes occur during the course of radiotherapy (RT) for lung cancer, and as a result, adaptive radiotherapy (ART) has been suggested to increase tumour control and spare surrounding healthy tissues. The safety of performing ART for lung tumour regression depends on whether any microscopic disease is left behind, i.e. determining the mode of tumour regression: elastic (surrounding tissues moving along with the regressing tumour) versus non-elastic (tumour regresses with static surrounding tissue). However, most studies focus on the volumetric changes of the visible GTV and do not distinguish between modes of tumour regression. This thesis aims to develop methods to describe, understand and predict lung tumour changes using routine cone-beam computed tomography (CBCT) images acquired during treatment. First, a method to investigate longitudinal intensity changes across the tumour-lung boundary was developed to identify patterns of tumour regression. Eight concentric shells were defined to sample the intensity across the tumour-lung boundary. Mean density change over time for each region was computed, and slopes of density change clustered. Results confirmed that distinct patterns of tumour changes exist during RT of lung cancer. However, no significant difference in overall survival was found between the identified groups and clusters could not be attributed to identifiable patterns of tumour regression. The methodology was then adapted to maintain spatial information of tumour changes. A novel approach was developed where the determinant of the Jacobian was sampled in regions across the tumour-lung boundary. This aimed to better understand modes of tumour regression by investigating local volume changes happening to the tumour and surrounding lung tissue. This voxel-wise approach allowed modes of tumour regression to be identified. This work showed that pure elastic regression did not occur for any patient in our cohort and instead tumours showed mixed modes of tumour changes throughout treatment. Finally, individualised voxel-wise regression models were developed to automatically predict tumour shape and volume at later weeks of treatment. These models used CBCT images acquired up to mid-treatment (week 2). The results showed a linear fit performed better than higher degree polynomials. This study provides a means to proactively predict patients who will show large GTV reduction, and therefore, would benefit from ART. Early prediction would give departments enough time to determine the best adaptation for each patient. Importantly, results in this thesis show, for the first time, that sub-regions of lung tumours display different modes of tumour regression. This information is essential to enable safe clinical ART for lung cancer patients, avoiding potential underdosing of microscopic tumour deposits and ensuring optimal patient outcomes. Lastly, early identification of good responders will improve the feasibility of ART for many patients.
|Date of Award
|1 Aug 2021
- The University of Manchester
|Eliana Vasquez Osorio (Supervisor), David Cobben (Supervisor), Andrew Green (Supervisor), Alan Mcwilliam (Supervisor) & Marcel Van Herk (Supervisor)
- Tumour changes