Despite multimodal aggressive treatment, the median survival time after diagnosis for Glioblastoma Multiforme (GBM) patients is only 12 months, with 5-10% of the patients surviving for more than 3 years (long-term survivors). Within the confined environment of the brain vault, tumor growth, especially in an aggressive tumor such as GBM, forces the compression of surrounding brain tissue resulting in increased intracranial pressure, exacerbation of vasogenic edema, and brain herniation¹. A study on GBM pathology identified herniation or gross distortion of the brainstem as the proximal cause of death in 60% of GBM cases². In light of these observations, a few efforts have been made to identify a relationship between extent of tumor mass effect as manifested on MRI and overall survival. However, the results have been conflicting and ambiguous^{3, 4}, partly on account of a simplistic 2-dimensional distance-to-the-midline measurement used to capture midline shift due to mass effect on MRI. More recently, several efforts have been made to explore radiomics-based approaches to capture per-voxel textural differences across enhancing and edematous regions between long-term and short-term GBM survivors^{5, 6}. In this work, we present the first attempt of its kind, to statistically identify and characterize neuro-anatomical differences on MRI caused due to mass effect, in populations of GBM patients with long-term versus short-term survival characteristics. The anatomical differences are captured using tensor-based morphometry (TBM), a neuroimaging tool that involves computing a deformation field to identify differences in the relative positions of brain sub-structures across the two populations (with respect to a healthy template). The deformations are computed as a log (Jacobian) measurement to measure the expansion (log (Jacobian)> 0) or contraction (log (Jacobian)< 0) of local brain matter, relative to the healthy template. The hypothesis of this work is that mass effect as manifested on anatomical MRI, differentially deforms the neighboring structures in the brain, and will lead to significantly higher deformations in patients with more aggressive GBM (short-lived), from those with less aggressive disease (long-lived).A total of 40 pre-operative Gadolinium (Gd)-contrast T1w, T2w, and FLAIR MRI studies with pathologically proven GBM were retrospectively obtained from The Cancer Imaging Archive. Our cohort had a median survival of 15.1 months, and was used as a cutoff to divide the population into long-lived and short-lived studies (Table 1). Pre-processing steps included co-registration of all protocols from a given study, bias field correction, intensity standardization, and skull stripping. Enantiomorphic normalization was used to remove tumor lesions from T1w images to allow for the creation of a minimum deformation population template. This template was then used within GLISTR, a tumor-aware registration tool that is driven by healthy and tumor tissue segmentations obtained from multi-parametric MRI, to register every patient study with respect to the template⁷. Warp fields generated from the registration were converted to Jacobian determinant images, and were used in a voxel-wise two-tailed t-test with a highly conservative cluster-mass based family-wise error correction with a cluster-forming threshold of t=3.3, to identify areas of the brain where deformation associated significantly with overall survival. To decrease the potential confounding effects of age and the large deformations in the immediate vicinity of tumors, these factors were included in the statistical model as nuisance regressors, to mitigate their effect as outliers. Figure 1 shows the per-voxel quantitative statistical map as quantified via the t-statistics (cluster corrected with p(cluster) < 0.001, at a conservative cluster forming threshold of t=3.3) demonstrating areas that show marked decrease in volume in short-lived relative to long-lived studies (yellow-red) and areas that show marked increase in volume in short-lived relative to long-lived studies (blue-light blue). Relative compression of cerebrospinal fluid (CSF) was observed in short-lived patient population around the cerebellum, which may imply that subarachnoid space is being compressed to a significantly greater extent in individuals with poor outcomes, and may ultimately be a cause of brain herniation. Similarly, the expansion of cortical volume at the medial surface of the cerebral hemispheres and around the ventricles in short-lived patients is consistent with deformation of the cortex across the midline (midline-shift) and into the ventricles. We presented the initial results of a tensor based morphometry approach to understand the impact of mass effect on different substructures in the brain as observed on treatment-naïve MRI, and its association with patient survival in GBM patients. Our preliminary results suggest that prominent deformation changes in CSF in cerebellum and the medial cortical surfaces at the central fissures on treatment-naïve MRI may serve as early indicators of poor survival in Glioblastoma.