Inclusion of signal analysis in a hybrid medical decision support system

Signal analysis has played an important role in cardiac diagnosis, both as a separate entity and in conjunction with clinical parameters. Hybrid systems are an effective method for developing higher-order decision models in which biomedical signal data can be incorporated.

The hybrid system components include a knowledge-based system that utilizes approximate reasoning techniques, a neural network model based on a potential function approach to supervised learning that uses the general class of Cohen orthogonal functions as potential functions, and a signal analysis component that relies on continuous chaotic modeling to produce a degree of variability in the time series. The hybrid system is illustrated in an application for differentiation among different types of dementia.

Application of this method to cardiac diagnosis shows that chaotic parameters alone contribute significantly to correct classification while the addition of clinical parameters increases the sensitivity, specificity, and accuracy. Applications to electroencephalogram analysis indicate that the second-order difference plots display significant differences for the different types of EEG waves identifiable by frequency, both in shape and degree of dispersion. Hence the identification of these waves, and the duration of their occurrence, may provide suitable variables for chaotic analysis.

Results from studies in cardiology demonstrate that using chaotic measures for ECG analysis provide useful information for classification. Sensitivity, specificity, and accuracy are increased if these methods are combined with other clinical parameters in a hybrid system. This approach has been extended to new applications based on EEG analysis combined with other relevant information.