Systematic comparison of semi-automated VBM pipelines for assessing chemoradiotherapy effects on the tumoural and non-tumoural brain in glioblastoma multiforme.

Abstract

Objective: Current voxel-based morphometry (VBM) studies of chemoradiotherapy effects on healthy tissues of the glioblastoma multiforme (GBM) brain faces challenges with neuroanatomical distortions (tumour, tumour oedema and resection cavities). Our aim is to compare current semi-automated segmentation methods and evaluate their reliability in investigating the effects of chemoradiotherapy on the tumoural and non-tumoural brain.

Materials and methods: VBM pipelines of CAT12 and FSL were performed and compared on an open-sourced imaging brain tumour dataset (Burdenko's Glioblastoma Progression dataset). The pre- and post-chemoradiotherapy grey (GM) and white matter (WM), and cerebrospinal fluid (CSF) volumes of whole brain, tumour-containing and nontumour containing hemispheres were derived and compared between both pipelines. Brain volumetric agreement and consistency between FSL and CAT12 were assessed using Bland-Altman plots and Intraclass correlation coefficient (ICC).

Results: Post-chemoradiotherapy GM volumes were significantly reduced in whole brain, tumour-containing and nontumour-containing hemispheres with a compensatory significant increase in CSF volumes, while WM volumes had no significant changes. Visual inspection revealed misclassification of tissue classes in the presence of neuroanatomical distortion for both pipelines, but FSL demonstrated superiority in tissue classification in the presence of a haematoma. Consequentially, Bland-Altman plots and ICC showed better agreement and consistency between FSL and CAT12 in GM_ICC (0.70 (0.53-0.82); p < 0.001), WM_ICC (0.75 (0.60-0.85); p < 0.001) and CSF_ICC (0.55 (0.32-0.71); p < 0.001) volumes of nontumour-containing hemispheres than whole brain and tumour-containing hemispheres.

Conclusions: VBM studies of chemoradiotherapy effects on the brain post-tumour resection remain challenging due to neuroanatomical distortions. A reliable alternative is to use nontumour-containing hemispheres with no anatomical distortion.