Skip to Main Content
Medical time series contain important information about the condition of a patient. However, due to the large amount of data and the staff shortage, it is difficult for physicians to monitor these time series for trends that suggest a relevant clinical detoriation due to a complication or new pathology. This paper proposes a framework that supports physicians in detecting patterns in time series. It has three main tasks. First, the time-dependent data is gathered from heterogeneous sources and the semantics are made explicit by using an ontology. Second, Machine Learning techniques detect trends in the semantic time series data that indicate that a patient has a particular pathology. However, computerized classification techniques are not 100% accurate. Therefore, the third task consists of adding the pathology classification to the ontology with an associated probability and notifying the physician if necessary. The framework was evaluated with an ICU use case, namely detecting sepsis. Sepsis is the number one cause of death in the ICU.