The immune system can specifically identify and target tumor cells on the basis of tumor-specific antigens that are expressed as a consequence of the plethora of genetic and epigenetic transformations that characterize oncogenesis [1]. The term “immune checkpoints” refers collectively to the multiple inhibitory pathways built into the immune system to maintain self-tolerance and inhibit auto-immunity. There is now an increasing understanding of how tumors can commandeer one or more immune checkpoint mechanisms to evade immune surveillance [2]. Following Krummel and Allison’s landmark report on CTLA-4 [3], programmed cell death protein 1 (PD-1) became the second T cell checkpoint receptor to be discovered, and it potently inhibits T cell effector function and expansion when engaged by its ligands PD-L1 and PD-L2 [4]. The CTLA-4-blocking antibody ipilimumab has received FDA approval for melanoma, while the PD-1-blocking antibodies pembrolizumab and nivolumab have received FDA approval for Non-Small Cell Lung Cancer and melanoma. High PD-L1 in tumors is associated with unfavorable prognosis in glioblastoma multiforme (GBM) [5], while increased expression of PD-1 on CD4+ and CD8+ T cells in peripheral blood correlates with disease progression [6]. These data support the use of pembrolizumab in patients with high grade gliomas. Abnormal tumor vasculature, partly caused by pro-angiogenic factors such as VEGF, can lead to increased infiltration of immunosuppressive cells such as tumor-associated macrophages and regulatory T cells [7,8]. The anti-VEGF antibody bevacizumab [9] can therefore potentially improve glioma response to pembrolizumab. Hypofractionated radiotherapy has been demonstrated to synergize with anti-PD-1/PD-L1 blockade in multiple pre-clinical cancer models [10,11]. We have rationally combined pembrolizumab with bevacizumab and hypofractionated radiotherapy (HFSRT) in this study. The objective of this work is to develop a non-invasive response assessment measure using multiparameter MRI (mpMRI) scans which may precede volumetric changes.In this phase I trial, subjects with recurrent high grade glioma are treated with HFSRT (25 Gy in 5 fractions), bevacizumab (10 mg/kg q 2weeks) and pembrolizumab (100 or 200 mg based on dose level, every 3 weeks). T2-weighted, FLAIR, Diffusion-Weighted, T1-weighted (pre- and post-contrast) and perfusion MRI scans of the brain are acquired at baseline and following each cycle of therapy. The presence of infiltrative malignant tumor into functioning brain makes the task of tumor segmentation challenging. We have built on previously reported approaches [12,13] to objectively segment the brain into normal and abnormal volumes-of-interest (VOIs) using a combination of five co-registered MRI sequences (Figure 1, Figure 2). Voxels within the “abnormal VOI” are further subdivided into high and low intensity clusters independently on FLAIR, ADC map and T2-weighted sequences; each voxel in the “abnormal VOI” is thus assigned to one of 8 distinct “habitats” (Hi/Lo ADC, Hi/Lo FLAIR, Hi/Lo T2w).At the time of abstract submission, 3 patients had post-treatment MRIs and the study is ongoing. Figure 3 depicts the variation in habitat composition of the abnormal VOI, at baseline vs. first follow-up scans after cycle 1, in 3 study subjects. A large change in the percent volumes occupied by habitats 2, 3, 5, and 6 in particular is apparent for Subject A at the first follow-up relative to baseline (figure 3). We have quantified this change by calculating the entropy of habitats distribution within the abnormal VOI in all 3 subjects. The entropy of habitats distribution in Subject A was lower before therapy than in the other two subjects, and there was a larger increase in entropy for Subject A than in the other 2 subjects (figure 4). Interestingly, Subject A is the only one of the 3 subjects on the study who has shown complete response as assessed by clinical and RANO criteria. At this stage in the study, it is difficult to assess whether pre-therapy entropy of habitats, or post-therapy change in entropy of habitats, is important for determining this apparent correlation with response.Heterogeneity within tumors, especially GBM, is now recognized to be a potentially causal factor that limits the effectiveness of therapies [14]. Entropy of distribution of 8 “habitats” within the “abnormal VOI”, identified by Otsu’s method [15], shows promise as a measure of heterogeneity within recurrent glioma. Advances in the fields of multispectral optical and satellite image analysis [16,17] can be brought to bear to extract quantitative information on GBM heterogeneity from mpMRI scans. We are investigating color-texture coherence based methods to objectively select the number of “habitats” within the “abnormal VOI” [18]. Accrual to this trial is ongoing, and mpMRI analyses of all treated patients will be reported.