gms | German Medical Science

Objective: The revised 2016 WHO-Classification of CNS-tumors has deepened the necessity to examine the biological and molecular properties of glial brain tumors for accurate diagnosis. In our prospective study, we aimed to investigate the predictive value of MR-spectroscopy in order to establish a solid preoperative molecular stratification of these tumors. We further developed a novel MR-spectroscopy-based algorithm that allows the specific molecular analysis of these tumors through a radiomics analytics pipeline.

Methods: 105 patients treated for WHO-Grade II, III and IV gliomas at our institution from 2016 to 2018 received preoperative anatomical and chemical shift imaging (MRS) (5x5x20mm voxel size). Tumor regions were segmented and co-registered to corresponding spectroscopic voxels. Feature extraction was performed by a computational model (total number of features n=15760) and included into a machine learning approach. Histological diagnosis was postoperatively matched. Finally, a Convolutional Neural Network (CNN) was trained to predict the molecular subtypes of glial tumors.

Results: A cluster analysis identified four robust spectroscopic clusters based on the mean silhouette widths. Molecular subgroups (cluster 1: Glioblastoma IDH mutated, cluster 2: Astrocytoma WHO-Grade III IDH mutated, cluster 3: Glioblastoma and Astrocytoma IDH wild-type, cluster 4: Oligodendroglioma, 1p19q co-deleted, and IDH mutated) were significantly associated with the computed spectroscopic clusters (Fisher’s Exact test p<0.01). The CNN deep-learning network was trained on 2/3 of the data and validated by 1/3 of our data. The network performed with a total accuracy rate of 93.8%, although 1p19q co-deletion was not well separated from the IDH mutated Glioblastoma cluster.

Conclusion: MR-spectroscopy is a powerful tool that supports the prediction of the molecular subtype of glial brain tumors, adding important diagnostic information to the preoperative diagnostic work-up of glioma patients and supporting surgical and oncological decisions to improve personalized tumor treatment.