Intertwining Ontologies with Deep Learning for Events and Temporal Expressions Extraction from Narrative Medical Texts

Medical ontologies play a crucial role in structuring, organizing, and standardizing healthcare data, enabling efficient information retrieval and semantic interoperability. By providing a formal representation of medical concepts and their relationships, ontologies facilitate accurate extraction and integration of knowledge from diverse clinical texts and electronic health records. Their application enhances decision support systems, medical research, and automated reasoning, improving patient care and healthcare analytics. Additionally, medical ontologies help address the challenges posed by the complexity, ambiguity, and evolving nature of medical terminology, ensuring consistency across healthcare systems. Integrating ontologies with natural language processing (NLP) techniques and machine learning models further strengthens medical data analysis, enabling more precise and context-aware information extraction. Thus, leveraging medical ontologies is essential for advancing healthcare informatics, fostering interoperability, and driving innovation in medical knowledge management. This study highlights the effectiveness of using Ontology Web Language (OWL) using the Owlready2 library to enhance healthcare data management, enabling efficient ontology-based reasoning and improving the extraction of meaningful medical information. Our proposal takes an innovative approach, using medical ontologies (Drug, Symptom, and Human Disease) and various convolutional neural networks (CNN) to extract medical events and temporal expressions from narrative medical texts. Several tests were conducted on datasets, including i2b2 2010, i2b2 2012, and i2b2 2018.