Term recognition from electronic medical records of Singaporean hospitals

Unlike other Medical text, Electronic Medical Records (EMRs) are written by doctors in a clinical setting. As EMRs are relatively new themselves, they lack well defined standards for input and doctors usually write them in free-form dictation These form of text lack proper sentence structure and contain numerous acronyms that only appear in such context, making it difficult to apply current parsing techniques to discover semantic meanings of the text. In this project, methods to identify unknown acronym terms in EMRs and disambiguate their meaning are explored and developed. To better apply automated processes, the EMRs are first pre-processed to restructure their text into a more defined format. Finally these unknown acronym terms are marked up for future use. Results obtained from the formatting of the EMRs showed a precision of 100% and a recall of 67.93%. Initial extraction of acronym-value pairs from the EMRs resulted in 2329 extracted terms with 68.14% true positives, the implementation of an exclusion list, consisting of recurring false positives increased the true positives to 90.99%. While the automated disambiguation process of these true positives could not be implemented, distinct patterns for 31 acronyms were identified which can be used to disambiguate their occurrence in EMRs.Bachelor of Engineering (Computer Science