Data Mining for Knowledge Discovery from Imaging Reports

As medical practice has become increasingly dependent"nupon imaging for clinical diagnosis, surveillance, and"ntreatment; the storing and retrieving imaging cases"nare an important activity for education and clinical"nresearch. There is an urgent need for development of"na new generation of computational tools to assist the"nradiologist in the extraction of useful data from the"nrapidly expanding volumes of digital data in medicine"n(which is textual, numerical and graphical). The"ndevelopment of this database technology will expand"nour ability to record, track and analyze reported data,"nalong with the potential to create data-driven and"nautomated decision support technologies at the point"nof care. For the radiologist community, this could"nimprove the report content through an objective and"nthorough understanding of uncertainty, identifying its"ncausative factors, and providing data-driven analysis"nfor enhanced diagnosis and clinical outcomes. The"nconcept of identifying useful patterns within large"ndatasets has been referred to as data mining, which"nhas been largely restricted to statistical analysis."nData mining tools for imaging reports will improve"nthe productivity of academic radiologists in clinical,"neducational, research and administrative tasks. For"nexample, data mining used to analyze report uncertainty"ncan also be applied to analyze other components"nwithin the imaging report including pathologic"nfindings, descriptive data (such as size, morphology"nand focality), follow-up recommendations, temporal"nchange and differential diagnosis. In this research,"nour goals were to describe a tool to allow radiologists"nto directly and efficiently mine data from years of"nimaging reports while protecting patient privacy and"nto explore how such a tool may be used in an academic"nradiology department."nKeywords: Data Mining, Knowledge Discovery,"nImaging Reports