Development and application of tools for large scale population based evaluation and optimization of localized glioma treatment

Surgery and radiotherapy, both localized treatments, are important corner stones of the treatment of patients with glioblastoma. The location of the tumor in the brain is an important factor determining the feasibility of extensive localized treatment. However, it is not customary to use this information when evaluating treatment strategies.
This thesis focuses on incorporating this location information in the evaluation of glioblastoma treatment, by addressing the feasibility and showing the value of large-scale data collection and localized treatment evaluation for glioblastoma patients using routine clinical data.
We have reduced the manual labor required to obtain the segmentations needed for localized treatment evaluation by introducing a sparsified training approach for deep convolutional neural networks, enabling automatic segmentation using routinely obtained MRI scans.
Next, we have studied the performance of three statistical approaches to compare resection probability maps. We have shown that Fisher’s test can reliably be used to quickly compare resection probability maps between two hospitals. A more flexible, Bayesian approach showed similar results although at the cost of increased computation times.
Furthermore, we compared three definitions of progression: the diagnoses made in clinical practice, following the RANO criteria, and the earliest radiological progression. Earliest radiological progression was identified as most suitable to pinpoint the time and location of progression for localized treatment evaluation.
Finally, we have compared hypothetical treatment strategies and found that target volumes using the preoperative tumor as reference structure could potentially allow for smaller target areas or improve progression coverage while maintaining the same target volumes.
We have shown that localized treatment evaluation provides new insights that can optimize future treatment strategies.