For patients with colorectal cancer, surgical resection (typically following down-staging chemotherapy) is the only curative therapy available [1]. Pre-operative planning in colorectal cancer is highly dependent on invasion of tumor into the mesorectum and perirectal fat [2]. However, the measure of tumor distance to the mesorectal wall (circumferential resection margin, CRM) is only measured until after surgery, delaying adjuvant intervention which could ensure better patient outcome. CRM values are classified as either positive (if the tumor is within 1 mm of the mesorectal wall) or negative (if the tumor localized to the mesorectum). Using a restaging MRI alone to identify extent of tumor invasion is difficult, as experts may not be able to accurately identify extent of tumor versus confounding treatment effects in vivo. Radiomic features, or computer-extracted attributes from radiographic images, have shown promise in quantifying subtle microarchitectural differences between different tissue regions, potentially enabling more accurate characterization of tumor extent after chemoradiation, but prior to surgery [3]. We present preliminary results of our approach for identifying radiomic features predictive of tumor invasiveness on a restaging MRI in colorectal cancers.24 patients diagnosed with colorectal cancer were imaged using an axial T2-weighted (T2w) turbo-spin echo MRI sequence post-neoadjuvant chemotherapy and prior to surgery. A total of 78 first-order statistical, gray level, gradient, Haralick, and multi-scale oriented Gabor radiomic feature maps were extracted from entire T2w volumes on a per voxel basis in order to capture structural homogeneity and heterogeneity characteristics within the region of tumor [4,5]. Expert annotations of the tumor were made on the original T2w images. Our approach consisted of two modules: (1) identifying top radiomic features by correlating feature statistics (median, variance, kurtosis, skewness) with the ratio of carcinoembryonic antigen (CEA) levels measured pre- and post-treatment, and (2) performing principal component analysis (PCA) to calculate the optimal combination of these top features in discriminating between patients who were identified following surgical resection as having either a positive or negative CRM. A cohort of 14 patients were used for training purposes to score the radiomic features, and a separate cohort of 10 patients were used to test the top 10 features in discriminating between positive and negative CRM classes (size of cohorts were limited due to available clinical information).In our testing cohort of 10 patients, 5 were identified as having a positive CRM post-surgery (tumor within 1 mm of the mesorectal wall), and 5 patients were identified as having a negative CRM (tumor localized to the mesorectum). A scatterplot of samples in 3D PCA space appears to demonstrate excellent clustering of patients with positive and negative CRM with an accuracy of 90% (one misclassification). While the original T2w intensity between images of positive and negative CRM patients does not visually indicate marked differences between the two patients, a heat-map of a top-ranked Gabor feature demonstrates feature expression levels that are markedly different between positive and negative CRM patients at the voxel level. Boxplots of feature intensities between the two CRM classes further indicate that Gabor features express differentially between positive and negative CRM patients, by capturing multi-scale gradient information that may be reflective of tumor invasiveness. We have presented initial results of a quantitative approach to predict tumor invasiveness using radiomic features in the context of colorectal cancer, via restaging MR images. Gabor features were found to be most discriminatory, by capturing microarchitechtural oriented gradient differences between patients with positive and negative CRMs. Validation of these features on a larger cohort of patients will enable us to develop a novel radiomic score for predicting patient CRM class prior to surgery, allowing for better pre-operative surgical planning.