The Use of ASR to Make Clinical Documentation in Serbian

Medical information systems are used to manage electronic medical records (EHRs) which store information about patients' health and medical treatments. These records store data daily in a structured, semi-structured, and unstructured form. As unstructured data preserves details about the health of patients written in natural language, artificial intelligence techniques, such as natural language processing (NLP) techniques, can be applied to this part of medical reports. To obtain as useful knowledge as possible from patients’ data, before data processing, it is necessary to make adequate preparation. Due to the limited duration of the examination, physicians often make typos when writing clinical documentation. The processing of misspellings that occurred during the writing of the electronic medical records is one of the steps in data preparation. If the ASR (automatic speech recognition) is used when creating a medical report, some common typing errors can be avoided. In this paper, a set of electronic medical data written in Serbian is read through using the ASR, and differences in the distribution of misspellings is analyzed compared to manually entered anamnesis. The high-level architecture of the healthcare knowledge extraction system has been proposed, which would serve to take a patient's data using ASR, and then further processing of the NLP to correct errors and classify the text.