Towards the Automatic Discovery of Rules Based on ECG Data for the Support of Personalized Medical Decisions

Abstract

With the increasing amount of available medical data and particularity bio-signal data, there is a need to develop computerized techniques to support medical staff in providing effective and efficient treatment options. Personalized medical treatment requires the physician to provide treatment based on characteristics specific to the individual patient. This process is typically driven by subject matter expertise and established medical standards. The electrocardiogram is the most common non-invasive method for monitoring the health condition of the heart. In order to support the evidence-based decision-making process, available data can be examined to identify patterns that are relevant to the characteristics of the individual patient. For this process to be beneficial, the knowledge representation and discovery approach need to be easily understood and interpreted by humans as well as machine-readable. Also, the developed model must be flexible to accommodate the evolution of available medical data over time. In this dissertation, we proposed an approach for the development of a decision support system based on ontology and learning classifier systems. The use of ontology supports the integration of data from different sources and various formats, and it provides a mechanism for knowledge enrichment based on available medical standards and the addition of newly available data. Learner classifier systems can support the automatic discovery of rules that can be used to investigate the factors impacting the medical condition of the patient and conduct investigations in the form of what-if scenarios to identify the underlying conditions that are affecting the health of the heart. The approach we propose intends to allow physicians and researchers to use the inference between the attributes of the patient data to create custom what-if scenarios and determine otherwise hidden relationships between the data.