This document discusses knowledge discovery and data mining of free text radiology reports. It outlines challenges with semantic indexing of medical text due to variations in terminology. An expert system called MEDAT is demonstrated that uses semantic parsing to represent sentences in a radiology report as predicate-argument structures mapped to medical concepts. While current systems can index about 60% of reports, fully automated semantic indexing remains a challenge due to implicit knowledge, phrasal synonyms, and representation of concepts not covered in existing ontologies. Further research is needed in rule-based semantic indexing and integrating statistical and rule-based approaches.