Natural Language Processing

We propose a hybrid NLP solution: a medical ontology system from Mayo Clinic, the Mayo Vocabulary Server (MVS) based on SNOMED- CT, MEDCIN and many other UMLS terminologies; and a machine learning system from Carnegie Mellon University (CMU), the Scone semantic knowledge-base that represents knowledge from numerous ontologies and corpuses. Out-of-the-box, the MVS far exceeds the Phase I goals, with a proven recall sensitivity of 99.7%, precision of 99.8% and specificity of 97.9%, using a 5,000 inpatient and the outpatient published data sample set. While the MVS provides a great vertical depth within the medical domain, the Scone horizontally broadens the knowledge into other domains, increasing the realm possibilities. For example, an NLP input to Scone linked to a news feed can automatically understand an event such as a “forest fire in Alaska” and the concept that “fire causes smoke”. The MVS then can be used to inference about smoke and return the concept that “smoke triggers asthma”. This and additional inferences about health and local information can conclude expected inpatient volume, helping first responders in natural and homeland security scenarios. NLP input from speech recognition systems are also addressed to provide real-time prompts to improve medical form inputs.