| Popis: |
Currently, the interest in medical informatics is to extract medical terms from medical records for later use as searching for the best treatment and use knowledge in cooperation with other physicians. In literature, several algorithms exist that extract certain information of interest using certain models. An area where accurate, real and in-time information is paramount is the medical prescription one, where information is on the rise, and information on drugs, drug interactions or other interactions is vast and complex. In this paper, we use two CRFClassifier algorithms to extract several terms of interest from the contraindication and dosing sections of patient information leaflets in Romanian and later use them in creating decision-support applications. We use the Stanford NER and Scikit-learn tools to extract medical information and create a corpus to train the algorithms. Following testing, the best-performing tool proved to be Scikit-learn. The obtained data is useful in creating ontologies for further use in decision-support applications reducing medical errors in the prescription process. |
