Clinical ontology is a standardized medical knowledge representation model that facilitates the integration and analysis of a large amount of heterogeneous electronic health record (EHR) data. Using ontologies to represent clinical terms can improve data integration to build robust and interoperable medical information systems. To date, there is no ontology existing to represent the medical knowledge for physical examination of stroke, which has inhibited the stroke physicians to make full use of clinical information captured in EHR data to understand stroke patient's health status and plan effective medication and rehabilitation treatment. In this research, we co-design with two stroke clinical specialists a stroke clinical ontology "StrokePEO"using advanced natural language processing and deep learning techniques to extract terms and their relationships from real clinical case records provided by a tertiary hospital in China. We apply the W3C Resource Description Framework (RDF) data model to represent these clinical terms and relationships, and successfully store all case data in a graph database with StrokePEO. Our experiment results suggest that our methods and the output of StrokePEO can be applied in various medical contexts that require extraction of medical knowledge from free text for decision making. These include, but not limited to, physical assessment, drug and rehabilitation treatment outcome evaluation, medication effect analysis, and patient risk prediction.