Towards the development of a blockchain-enabled voice-to-text transcriber plugin in an electronic medical record for doctor-patient conversations

• Main Author: WENCESLAO, STEPHEN JOHN MATTHEW
• Format: text
• Published: Archīum Ateneo 2019
• Subjects: Medical records > Data processing; Intelligent agents (Computer software) > Design; Medical informatics; Medical Records Systems, Computerized > organization and administration; Natural language processing (Computer science); Automatic speech recognition.
• Online Access: https://archium.ateneo.edu/theses-dissertations/246; http://rizalls.lib.admu.edu.ph/#section=resource&resourceid=2027547010&currentIndex=0&view=fullDetailsDetailsTab
• Institution: Ateneo De Manila University

Electronic medical records (EMR) in general provide significant benefits to healthcare organizations and clinicians. However, most physicians have poor experiences with the usage of EMRs in their workflow. EMRs hinder with their ability to communicate effectively with their patients. This study aims to develop a voice-to-text transcriber plugin for an EMR to improve the documentation process during consultation and to allow physicians to communicate more effectively with patients. An editable summary of the clinical encounter is presented to the user in Subjective, Objective, Assessment, and Plan of Action (SOAP) format, and then saved to the EMR once the note has been finalized. Blockchain technology for the speech recording is also explored to enable transparency and accountability to speech-enabled consultations by logging the interactions made on the record on a blockchain network. Initial tests show that the accuracy of the speech recognition implementation is negatively correlated with recording duration by calculating the Levenshtein (LV) distance against the expected transcript, using datasets from Kaggle and Ezdi. Tests on annotation of patient symptoms show a 7% accuracy in terms of string similarity with the expected result. Improvements for speech to text transcription in EMRs should include: SOAP tagging, shortened conversations, improve the natural language processing (NLP) aspect to capture only the important concepts, and testing for actual doctor-patient interaction.