Conditional Random Fields and Supervised Learning in Automated Skin Lesion Diagnosis

Many subproblems in automated skin lesion diagnosis (ASLD) canbe unified under a single generalization of assigning a label, from an predefinedset, to each pixel in an image. We first formalize this generalizationand then present two probabilistic models capable of solving it. The firstmodel is based on independent pixel labeling using maximum a-posteriori(MAP) estimation. The second model is based on conditional randomfields (CRFs), where dependencies between pixels are defined using agraph structure. Furthermore, we demonstrate how supervised learningand an appropriate training set can be used to automatically determineall model parameters. We evaluate both models' ability to segment achallenging dataset consisting of 116 images and compare our results to5 previously published methods.