SBIR-STTR Award

The goal of this research is to develop improved techniques for fully automated and computer assisted classification of medical text. A primary focus will be mechanisms that support easy to construct classifiers, thus enabling research on existing collections of free text that are now difficult to analyze. This approach is exemplar based i.e., compare new text with a training set of previously classified texts, and use the classifications of the closest retrieved texts to generate suggested codes for the new text. Natural language extraction techniques will pre process texts to assist the retrieval macfiinery. Phase I will conduct a series of experiments to test the approach, utilizing coded radiology and pathology reports from Brigham and Women's Hospital and HCHP in Boston. Phase II will develop a full software prototype and deploy it for on site evaluation. Advanced retrieval techniques and expert system back ends for alarminR will also be explored in Phase II. The major technical innovation is a novel combination of document retrieval and natural language extraction technologies to permit easy construction of automated medical text classifiers. The major health related contribution is an enhanced ability to classify existing free text records to permit statistical analysis for research and clinical quality measurement initiatives.Commercial ApplicationsImportant commercial application of this technology are coding (ICD9, CPT4, etc.) for hospitals, insurance companies, and pharmaceutical companies (drug safety monitoring). Other important applications include follow on alarming or reminder systems and quality measurement initiatives for medical institutions.National Cancer Institute (NCI)

The goal of this research is to develop improved techniques, both fully automated and computer-assisted, for classification of medical text. The technical approach is exemplar-based: robust information retrieval methods find similar, previously-classified texts, and corresponding codes are used to suggest likely classifications for a new text. Phase I focused upon implementing experimental software to establish baseline performance with several variations of the exemplar-based approach. Phase II builds upon this work to implement a complete Coder's Workstation (CWS). Based upon Phase I results and assessments of commercial opportunities, Phase II will focus upon shorter texts (<12 words), which are best suited for automated methods. A "short-similarity" capability will be added to the Phase I approach to further enhance performance with shorter texts. To evaluate and refine the CWS, Phase II will include extensive "beta testing" of the software at the Brigham and Women's Hospital and the Mayo Clinic. The major technical innovation of this project is the development of highly automated classification software that is sensitive to term similarities. The major health-related contributions are large potential savings in coding expenses, reduced time demands upon physicians for coding, and improved consistency in classification of free text for research studies.Proposed Commercial Applications:The proposed technology will have important commercial application within hospitals, insurance companies, and pharmaceutical companies which currently expend significant resources on coding of free text (ICD9, CPT4, COSTART, etc.). The founders of Belmont Research Inc. have extensive experience in creating and marketing software to support biomedical applications.