Active Surveillance (AS) offers an important alternative to radical treatment as more men die with prostate cancer (PCA) than of the disease (1, 2). Texture features (such as Gabor and Haralick features) are metrics calculated within an image to quantify the perceived texture and provide quantitative measurements pertaining to the spatial arrangement of color intensities within a region of interest. Computer-extracted texture features may allow for a “virtual biopsy” in men with elevated PSA to identify patient do not have prostate cancer and hence may be spared a biopsy. This could pave the way for developing a non-invasive MRI marker for better patient selection and monitoring patients during Active Surveillance. In this study, we explore the role of computer-extracted (or radiomic) texture features on a pre-biopsy screening 3 Tesla multi-parametric MRI (mp-MRI) that can predict which men with elevated PSA will have a cancer-positive or cancer-negative biopsy. [AM1] [AM1]Note how I have changed the motivation or rationale for the abstract….please change the rest fo the test accordingly to reflect this. 3T multi-parametric MRI (T2w images and Apparent Diffusion Coefficient (ADC) maps) was acquired for 45 men from a larger IRB-approved study where men with elevated PSA underwent a screening mp-MRI followed by a trans-perineal grid 30-core sextant TRUS-guided biopsy. Based on biopsy findings, the cohort was segregated into 30 biopsy-positive men (group 1) and 15 biopsy-negative men (group 2). All MRIs were manually annotated by an expert for being PCA-positive or PCA-negative based only on PIRADS criteria on each of the T2w MRI and ADC maps. Each T2w MRI acquisition was converted into a pseudo-quantitative T2 map by correcting for inherent scanner variability(3). This correction was also applied to ADC maps. A total of 230 texture features (including gray-level statistical, steerable Gabor, Entropy, Mean, Median, Sobel and Laws; 115 features from each protocol) were implemented in MATLAB (Mathworks, Inc.) and extracted within the prostate cancer annotation regions for each of T2w MRI and ADC map independently from TRUS guided biopsies[AM1] . The Minimum redundancy Maximum relevance (mRMR) algorithm was used to identify the most relevant 20 CETFs (of the total of 230) that could distinguish between the two groups of biopsy positive and biopsy negative men. Ranksum testing was applied to identify which CETFs showed maximum separability between the two patient groups. [AM1]Same question here…was the radiomic analysis done within the regions annotated via PIRADS on the imaging, was biopsy taken into account or not at all? Need to clarify this. Table 1 shows 7 T2w texture features (Gabor, Mean, Standard deviation and Laws) and 3 ADC features (Energy Laws, Gradient and Sobel) were found to have statistically significant differences (p < 0.001) between groups 1 and 2 in ranksum testing. Figure 1 shows box plots of Energy Laws and Sobel texture features distribution for both MRI parameters used. For group 1, the extracted feature values were confined to a relatively narrower range (0.05 – 0.45) than those of group 2 (0.09 – 0.88)[AM1] . The selected texture features correctly identified 14/15 biopsy-negative (compared to 10/15 cases correctly identified by PIRADS) and 23/30 biopsy-positive cases (compared to only 15/30 correctly identified by PIRADS). Figure 2 shows the texture feature heatmap for a correctly identified case by texture features. [AM1]Just provide the range here like you did for group 1. Computerized texture features derived from T2w images and ADC maps appear to enhance differentiation between biopsy-positive and biopsy-negative prostate cancer patients on Active Surveillance compared to PIRADS alone. This can help predicting which men with elevated PSA may have a cancer-positive or cancer-negative biopsy.