Electronic medical record systems (EMR) contain a wealth of clinical data that is invaluable for biomedical research, but because there are no satisfactory methods to build coherent specialized knowledge bases, which represent the information in free text medical records, data mining and clinical discovery are held back. Medical Reporting Solutions, Inc. has developed advanced technology, which we propose to extend, refine, and test for constructing specialized semantic knowledge bases. These knowledge bases will encode the clinical information in medical reports, and enable automated natural language processing systems for extracting clinical knowledge. Our research and development uses methods in corpus linguistics and sentential logic to represent the knowledge in free-text medical reports in an efficient, codeable manner. We have created tools to map sentences in a medical domain to unique codeable propositions. Our method for creating knowledge ontologies makes it easy for biomedical researchers to get semantic information at the appropriate level of detail. The knowledge base and mapping tables allow us to analyze medical reports in near real-time. One knowledge base, under development, is derived from hundreds of thousands of reports in the radiology domain, and we intend to analyze other medical domains using the methods we have pioneered. Our phase one project plan includes further improving our knowledge editing tools, substantially enlarging our semantic knowledge base to cover over 60-70% of the radiology domain, and extensively test our knowledge representation schema against actual radiology reports. We plan to make the knowledge base freely accessible to the biomedical research community, while providing commercial services to codify free text reports found in EMRs.
