Improving surgical decision-making in glioblastoma

This thesis set out to answer the question whether probability maps can be used to evaluate surgical decision-making in glioblastoma. Chapter 1 provides a general overview of glioblastoma. An overview is presented of surgical decision-making, which addresses the tumor location, important patient-related factors in preoperative and intraoperative decision-making, discusses reported variance in surgical decisions and outlines a proposal for evaluation of surgical decisions by probability maps. This chapter ends with the aims and outline of this thesis: to evaluate surgical decision-making in glioblastoma using probability maps, including evaluation of preoperative surgical decisions and postoperative surgical results, and a critical review of the applied methodology as this is a novel instrument. Neurosurgeons often establish their surgical decision based on the tumor’s ‘eloquence’: brain locations considered essential for brain functioning. To aid in decision-making it may help to have a measure of brain function prior to surgery. In Chapter 2 the ‘expected residual tumor volume’ and the ‘expected resectability index’ are presented, which are objective measures of expected extent of tumor removal derived from previous surgical decisions quantified in a resection probability map, and compare these measures with the historical Sawaya’s classification of eloquence grade in predictions of patient outcome. We demonstrate the expected residual tumor volume and expected resectability index predict biopsy decisions, resectability and survival better than the eloquence grade. We conclude the expected residual tumor volume and expected resectability index may serve as adjunct with clinical factors to guide biopsy decisions or provide estimates of extents of tumor removal. The impact of the time between radiological diagnosis of glioblastoma and surgery on patient outcome is unclear. In Chapter 3 we relate this time with the extent of resection and residual tumor volume, performance change and survival, and we explore the identification of patients for urgent surgery. We found equal extent of resection, residual tumor volume, performance status, and survival following longer waiting times. We also found that patients who were selected for urgent surgery had a low preoperative performance status and a large tumor volume. As the vast majority of patients were operated within one month from the initial scan, this seems a reasonable maximally acceptable time between radiological diagnosis and glioblastoma surgery. To evaluate neurosurgical decision-making between multidisciplinary teams the bias from heterogeneity of tumor locations is often unaddressed. Chapter 4 introduces a proof of concept to compare patterns of patient selection and surgical decisions throughout the brain between two multidisciplinary teams using probability maps. We found similar biopsy decisions, but a differential referral pattern for patients with a tumor in the left frontal lobe and a varying resection decision for patients with a tumor in the right caudate nucleus. This treatment variation remained unnoticed by common outcome measures such as the extent of resection and residual tumor volume and we therefore conclude probability maps may be a suitable adjunct or perhaps alternative to these measures. The previous Chapter indicates treatment variation in surgical decisions between two multidisciplinary teams, reflecting the need to evaluate multiple teams on a larger scale to establish a benchmark. In Chapter 5 surgical decisions from twelve multinational teams were quantitated and compared using probability maps. We demonstrate the teams generally agree which glioblastoma-infiltrated location to biopsy or resect, providing benchmarks for probability maps of biopsies and resections for glioblastoma. Two examples of treatment variation were explored, providing the teams with objective information to guide new decisions. Glioblastoma resections are usually assessed by volumetric measurements, implicating that absence of residual tumor is the norm. To integrate surgical decisions made in other patients, a quantification of the deviation from a resection probability map for glioblastoma resections is introduced in Chapter 6 and its association with patient performance change and survival is explored. We found resections with less deviation were associated with longer survival. More deviant resections were not related with changes in performance. This supports the concept of an optimal glioblastoma resection, beyond which longer survival can not be pursued, and below which performance can be compromised. Manual segmentation is subjective and may lead to interobserver disagreement, possibly limiting reproducibility of probability maps. In Chapter 7 we therefore explore the interobserver agreement of glioma segmentations on longitudinal MRI. For glioblastoma, we found the interobserver agreement was excellent between experts for segmentations of enhancement for preoperative scans of glioblastoma and reasonable for postoperative scans. The interobserver agreement was good between novices for preoperative enhancement and poor for postoperative enhancement. We conclude the reliability of construction of probability maps is negligible by uncertainties for preoperative segmentations because of a high interobserver agreement, while the lack of high agreement for postoperative segmentations may introduce measurement bias. Registration of tumor segmentations together with brain deformations towards standard brain space is challenging and may lead to less reliable probability maps. In Chapter 8 we therefore explore the accuracy of MR image registration to anatomical reference space with linear and non-linear transformations using estimated tumor targets of glioblastoma and lower-grade glioma segmented by an expert, and anatomical landmarks at pre- and postoperative time-points using six commonly-used registration software packages. We found no apparent differences in accuracies between packages. We also found linear transformation suffices to summarize glioma locations in anatomical reference space, which therefore may reliably substitute our non-linear approach. Nevertheless, we conclude from this study no algorithm outperforms our registration approach and therefore determine our methodology adheres to current standards. Chapter 9 presents the general discussion of this thesis where the most important findings of this thesis are discussed and the implications for clinical practice are reviewed. Considerations for future research are subsequently provided