Automatic Segmentation of Pediatric Brain Tumours using Diffusion-weighted MRI: Towards an Early, In-vivo Classification Pipeline.

Diffusion-weighted MRI (DWI) can offer vital quantitative biomarkers to understand pediatric brain tumours. However, extraction of these markers requires delineation of pathological tissue, which is time-consuming, requires expert-resource and can still be highly variable. This study develops an automated segmentation approach, leveraging transfer-learning and multimodal
ensembling to tackle the issues specific to this DWI-only approach. Using a 3D CNN (namely Deepmedic) to perform automatic segmentations, we demonstrate the benefit of transfer-learning and ensembling for this task. However, we find limited accuracy of these segmentations, with limited reliability of the radiomic features extracted from these regions of interest (compare to the ground truth tumour mask). The current study highlights the potential of this approach, indicating the biases and issues which need to be
addressed before it can be implemented in future clinical decision support tools.