Glioblastoma multi form tissue type classification by means of 1D convolutional neural networks: a deep learning method

Abstract

Introduction Diffuse gliomas are characterized with spatial intra-tumor variability within their microenvironment, which is partly responsible for their grim prognosis ¹. Currently, contrast-enhanced T1-weighted (CE-T1w) and T2- weighted MR imaging are applied for guiding targeted biopsy and surgical/treatment planning procedures for gliomas ^2,3. Nonetheless, relying upon these images cannot sufficiently stratify tumorous regions including the most active tumour component and infiltrative edema (IE) from each other and from the normal tissue (NT). Deep learning algorithms have recently been proposed for MRS data quantification and are becoming a novel tool to solve difficult signal processing problem for in vivo MR spectroscopy ⁴. In this study, we investigate two approaches including: 1-D convolutional neural network (1D CNN) and SVM to explore the competitive/complementary roles of CSI spectral data for differentiation of biopsy-approved Tumoral tissue, Infiltrative edema and Normal tissue. Moreover, to be more general bone, and none brain area regions included in our study.