Many diagnostic tasks performed on the MR (Magnetic Resonance) scanner involve some sort of tissue characterization to determine the type or some other property of the imaged tissues. In some cases, like for stroke, making a characterization is very complex, because there is no contrast that gives clear clues from which all desired characterizations can be made. However, spatial overlap of contrast-specific signal effects between images or the absence thereof is essential for a comprehensive interpretation of a stroke lesion. We have presented the expert system that we have implemented, which is capable of exposing concordance or discordance of pathology-related effects both in the characterization output and as a result of data mining. As such, the data mining algorithm and characterization algorithms could be a helpful tool for the expert involved in stroke tissue characterization. Our system is especially suited for those situations, such as acute stroke, where the pathology, including the final set of relevant tissue classes, is not completely understood.