Assessing and Improving Validation of Biomedical Ontologies towards Semantic Interoperability in Healthcare

Semantic interoperability in health-informatics ensures meaningful exchange and interpretation of healthcare data. Semantic interoperability is accomplished by using biomedical ontologies, which record clinical information like procedures, diagnoses, symptoms in the form of numeric/alphanumeric codes to ensure standardization. Biomedical ontologies are usually a result of sustained team effort and consequently may contain conflicting modelling styles which introduce inconsistencies and degrade their quality. Incomplete and inconsistent representations of clinical concepts in biomedical ontologies limit their expressiveness and prevent healthcare applications from exploiting their full potential. Furthermore, inaccurate and incomplete representations of concepts in biomedical ontologies can have serious consequences in downstream applications like Electronic Health Record (EHR) systems referencing them. Given the critical nature of clinical information, validation of biomedical ontologies to ensure correctness and consistency is of utmost importance. However, the development of efficient auditing techniques to validate biomedical ontologies is a major challenge and still an open problem in the health-informatics domain. Due to the critical nature of medical data and huge variety in the logical definitions of biomedical concepts, complete automation of validation techniques is not yet possible. Researchers are developing semi-automated validation techniques that can audit large portions of biomedical ontologies and highlight erroneous regions for manual inspection, thus saving time and effort of the curators. The existing auditing techniques can be broadly categorized into structural, lexical and ontological methods. The research in this thesis focuses on improving semantic interoperability in healthcare in the following ways. Firstly, we identify gaps in the existing lexical auditing techniques involved in the Quality Assurance (QA) of biomedical ontologies and make a novel contribution by developing a method that targets stopwords to audit biomedical concepts. This method is developed for the world's most widely adopted biomedical ontology, Systematized Nomenclature of Medical Terms - Clinical Terminology (SNOMED-CT). We then assess the impact of the developed method to ease the manual efforts of curators in SNOMED-CT template creation process. Secondly, we assess Semantic Web Technologies (SWT), like Shapes Constraint Language (SHACL), employed in validation of generic knowledge graphs and their relevance to validate knowledge graphs representing specialized domain knowledge such as biomedical ontologies. We then make a novel contribution by proposing a method to create enhanced SHACL shapes incorporated with intrinsic lexical knowledge of concept names for better validation of biomedical ontologies.