Renal biopsies form the gold standard of diagnostic and prognostic assessments of renal transplants. With the addition of new quantitative strategies to supplement renal biopsy interpretation such as gene array and metabolomics, the capability to incorporate all quantitative measures for clinical interpretation will require multi-dimensional analyses. Currently, renal biopsies are analyzed manually; the quantitative features of pathology observed on the biopsies are limited to hand counts. Standardized, automated detection of pathology observed in a kidney transplant biopsy will enable the input of these digital images alongside other quantitative measures of new technologies, with potential gains in precision in patient care. We investigate a learning framework to detect pathological changes in biopsy image that addresses two main issues: the inadequate training set and the significant diversity of color and tissue shape on whole slide images. Two case studies, automatic detection of interstitial inflammation and tubular cast, are presented in this work. Afterwards, we propose a fully automated glomerulus extraction framework on micrograph of entire renal tissue, focusing on extracting Bowman's capsule, the supportive structure of glomeruli. Statistical approaches are also introduced to further improve the performance. Human expert annotations of interstitial inflammation and tubular casts in 10 H&E stained renal tissues of nonhuman primates and more than 100 glomeruli are used to demonstrate the superior performance of the proposed algorithm over existing solutions.