Advancement and Trends in Medical Case-Based Reasoning: An Overview of Systems and System Development
Markus Nilsson, Mikael Sollenborn · 2004
Diagnostics based on time series are sometimes difficult to perform, particularly when the time series is continuous and non-stationary, i.e. they seldom contain recurring patterns which makes it difficult to identify similarities with other time series. This doctoral thesis presents an artificial intelligence approach to the analysis of continuous non-stationary signals for diagnostic purposes. One way to solve this kind of problem is to break down the series into new forms that are more easily interpreted, and to identify familiar patterns within them. The newly formed series is analysed, using the Case-Based Reasoning paradigm. Known problemsolution pairs are stored in memory and reused for solving problems by classifying new patterns occurring in time series obtained subsequently. Reasoning is conducted on the basis of the knowledge available and a best-guess solution obtained using the available knowledge is presented. The memory need not therefore contain a problem, which has been solved previously and is identical with the problem which is to be solved. This approach to problem solving has been applied to physiological time series as a clinical decision support system. The system provides decision support by classifying patterns of respiratory sinus arrhythmia from heart rate and capnography time series.