Organ segmentation is a prerequisite for a computer-aided diagnosis (CAD)\nsystem to detect pathologies and perform quantitative analysis. For\nanatomically high-variability abdominal organs such as the pancreas, previous\nsegmentation works report low accuracies when comparing to organs like the\nheart or liver. In this paper, a fully-automated bottom-up method is presented\nfor pancreas segmentation, using abdominal computed tomography (CT) scans. The\nmethod is based on a hierarchical two-tiered information propagation by\nclassifying image patches. It labels superpixels as pancreas or not via pooling\npatch-level confidences on 2D CT slices over-segmented by the Simple Linear\nIterative Clustering approach. A supervised random forest (RF) classifier is\ntrained on the patch level and a two-level cascade of RFs is applied at the\nsuperpixel level, coupled with multi-channel feature extraction, respectively.\nOn six-fold cross-validation using 80 patient CT volumes, we achieved 68.8%\nDice coefficient and 57.2% Jaccard Index, comparable to or slightly better than\npublished state-of-the-art methods.\n