Experiments to Create Ontology-based Disease Models for Diabetic Retinopathy from Different Biomedical Resources

According to the World Health Organisation diabetic retinopathy (DR) is a high priority eye disease. This paper investigates a method for creating disease models for DR using the ontologies BioTopLite2 and SNOMED CT and different biomedical resources: 1) consultation notes from anonymised electronic health records; 2) the clinical practice guideline for DR by the American Academy of Ophthalmology; 3) the BMJ Best Practice for DR; and 4) neural language models from Deep Learning (CBOW and Skip-gram) using a 14M PubMed dataset. As SNOMED CT does not contain disease models, the novelty of this study is twofold: a) evaluation of the utility of CBOW and Skip-gram for obtaining DR disease models from the biomedical literature; and b) the proposed method for building ontology-based disease models exploiting SNOMED CT reference sets. In our method, we first propose a representation of SNOMED CT reference sets for DR in OWL by extracting upper modules from SNOMED CT. Secondly, we use content ontology design patterns with BioTopLite2 and SNOMED CT that act as templates to semantically represent clinical content in OWL. We report on the effectiveness of the method.