MediGraph: Semantic Knowledge Graph for Healthcare

A portion of medical data is available only in written form, such as medical journals, reports, and online health articles. It can be seen that the unstructured written data is complicated and cannot be rearranged and reused in its current form for analysis. Most available solutions deal with only the extraction of medical terminology, or over-rely on some form of empirical or heuristic rule with deep domain knowledge that limits flexibility or practical application. Considering this, the focus of this project is to provide an uncomplicated and practical means to convert healthcare-related unstructured text to knowledge graphs. The system was developed in Python and uses basic natural language processing. It offers a linear and sequential workflow approach where unstructured raw text data is first, cleaned and prepared. Following this, core medical entities are recognized in the processed text, such as those dealing with specific symptoms, diseases, and treatments. Once the recognition of entities is done, contextually appropriate and co-occurring meaningful relationships are created. Using the Neo4j database, relationships or graphs are stored so that users may confirm or verify them visually. Even though some additional pre-processing is still required for the system, it was able to provide reliable results for many common medical terminologies during the testing phase.