MRI and CT are used in radiotherapy (RT) planning of Glioblastoma Multiforme (GBM) despite their limitations. 1H spectroscopy can detect and quantify choline (Cho) and N-acetyl aspartate (NAA) which have been strongly associated with GBMs.1 Previously, Metabolic Tumor Volumes (MTVs) were investigated for better delineation of the Clinical Target Volumes (CTVs) in RT of GBM.2 The objective of this study is to investigate the associations between metabolic RT volumes and radiomic features in GBM. 17 patients with GBM were selected from a larger study based on availability of all three MRSI, contrast enhanced (CE) T1w and T2w sequences.3 MRI CE and necrotic volumes were identified on T1w, while edema was identified on T2w. All were manually contoured, and reviewed by a radiologist. Figure 1 shows volumes for necrosis (VN), contrast enhancing (VCE), and edema (VEd), and derivative volumes: “pure” enhancing volume (VPCE = VCE - VN), and “pure” edema (VPEd = VEd – (VCE ⋃ VN)). Shape irregularity was calculated by expanding and contracting the corresponding volumes by 2 cm. The resulting smooth contour was subtracted from the original one and labeled SN, SCE and SEd. The fraction of irregularity to corresponding volumes were also investigated (SN/VN, SCE/VCE and SEd/VEd). RECIST measures of volumes, RSCE, RSCE, RLN, RSN, RLED, RSED were automatically calculated.4 Images were reviewed by neuroradiologist and 13 semantic imaging features were recorded, described in Diehn et al.5, and Gutman et al.6 Current practice for RT of GBM defines two target volumes: (i) CTV receiving 46 Gy, defined as edema with an additional 2 cm margin as CTV46. The second volume, CTV60, includes CE with a margin of 2.5 cm and receives an additional boost of 14 Gy for a total dose of 60 Gy. MRSI volumes and clinical MRI’s were obtained to construct both MTVCho and MTVNAA based on volumes with high Cho and low NAA relative to normal-appearing tissue.2 To uncover the relationship between radiomics and MTVs, Spearman correlations were computed. For visualization of results, heatmaps were generated using hierarchical clustering. Eight of the13 semantic features were replaced with quantitative imaging parameters, calculated directly from the contoured imaging volumes. The correlations between radiologist’s reported categorical semantic features and automatically estimated ones were statistically significant. In Figure 2 an autocorrelation heatmap of radiomic features is shown. The radiomic features are represented by two major clusters; the top right cluster consists of high positive correlation mainly of features related to the size of the lesion: VN, VCE, VPCE, VEd , RLN, etc. The lower left cluster groups composite features; VPED/VCE, SCE/VCE, SCE/VPCE, VCE/VN, SEd/VEd, SN/VN, VCE/(VEd ∪ VN), etc., demonstrating that both groups of features provide independent characteristics of the tumor. In Figure 3A an autocorrelation heatmap of 18 metabolic features are shown where 8 are redundant (r>0.95) as seen in upper right (CTV46 ∩ Cho, MTV Cho, CTV60 ∩ Cho, CTV ∩ Cho), and lower left (MTV NAA, CTV46 ∩ NAA, CTV60 ∩ NAA, CTV ∩ NAA). Retained features are displayed in Figure 3B. Figure 4 shows the heatmap of two way correlation coefficients between radiomic and metabolic features. NAA is strongly correlated with VN, VCE, VPCE, VN, VED, and SN. Conversely: choline volumes show weak correlations with radiomic features indicating that choline is independent of radiomic feature tumor characteristics and thus provides additional information about the tumor phenotype. This study demonstrates the ability to replace subjective, categorical variables with quantitative analysis that allow for an objective method for evaluating GBMs. The potential usefulness of whole-brain MRSI for RT planning of GBMs revealed that areas of metabolically active tumor are not covered by standard RT volumes. The described integration of MTV into the RT system will pave the way to investigating outcomes of patients treated based on incorporated metabolic information.