PROJECT SUMMARY The goal of this work is to assess the clinical value of voxel-wise predictive spatial maps of tumor heterogeneity that directly reflect histopathologically defined tumor biology. It is well known that tissue samples used for clinical diagnosis come from a relatively small portion of a vastly heterogenous lesion and are obtained infrequently during the course of the disease. Non-invasive imaging markers that are able to assess intratumoral heterogeneity and serially monitor biological properties of the tumor are critical for assessing response to therapy and directing patient care. The modalities that have shown the most promise in quantifying surrogate markers of malignant characteristics in patients with gliomas include diffusion-weighted MRI, perfusion-weighted MRI, and 1H MR spectroscopic imaging (MRSI). We have accumulated multi-parametric physiologic and metabolic imaging data from pre- surgical scans in order to target over 2000 tissue samples from more than 750 patients with glioma. These samples are unique in that they have each been specifically selected to target heterogeneous regions of tumor biology, including: hypoxia, proliferation, cellularity, gliosis, and malignant transformation using a combination of anatomic, physiologic, and metabolic imaging. Using this well-characterized cohort, our novel approach will leverage multi-parametric imaging features in conjunction with advanced statistical-, machine-, and deep- learning models to predict tumor biology, molecular phenotype, and progression. Aim 1 focuses on predicting intra-tumoral heterogeneity and the extent of infiltrating tumor and in newly- diagnosed glioma in order to identify areas of malignant characteristics that will direct tissue sampling for a more accurate diagnosis and predict the spatial location and characteristics of residual disease. Aim 2 will define characteristics of treatment related changes vs recurrent tumor and malignant transformation within lower grade molecular sub-groups of glioma within patients undergoing surgery for suspected tumor progression. This supplement will allow for the development and incorporation of new machine learning approaches on our existing data as well as learn the imaging features that are predictive of newly-defined molecular subgroups of glioma that are more prognostic of outcome than previously defined 2016 criteria by the World Health Organization. The result will enhance and expand current strategies for evaluating patients with glioma and provide a framework for incorporating newly identified imaging, molecular, and genomic markers that can be integrated with current response assessment criteria for evaluating standard and experimental treatments.