The ultimate goal of this project is exploring high and low-resolution images techniques for performing high-level inferences over complete tissue medical slides, which presents a few challenging aspects regarding storage and processing. The high-level inferences consist of learning patterns between summarized slides and target labels for solving a specific problem. For example, it could be interesting to calculate the abundance (by area) of regions like angiogenesis, necrosis, and infiltrating lymphocytes to infer patient-level characteristics. A tissue slide can easily reach around 20GB in-memory, which turns intractable many approaches of feeding those slides entirely to a classifier. One recent method for addressing this problem is sampling regions based on low-resolution features (i.e. texture, boundaries) for extracting information about the tissue. Moreover, this method can also help with speeding up the processing, since after cropping the proposed regions, the problem would be turned into a tractable learning problem.