Comparison between Artificial Neural Networks and Discriminant Functions for Automatic Detection of Epileptiform Discharges

This study presents the performance analysis between two classifiers when they are used together with mimetic analysis based morphological features to develop a method for automatic detection of epileptiform discharges in EEG signals. We applied mimetic analysis in the form of extracting a set of morphological descriptors, which in this study represent a set of parameters related to morphology features of the EEG waveform. The two tested classifiers are Discriminant Functions (DF) and Artificial Neural Networks (ANN). The DFs are obtained from Discriminant Analysis and are frequently applied in pattern classification problems such as automatic identification epileptiform discharges. On the other hand, the ANNs are an Artificial Intelligence tool commonly used in pattern recognition methods and systems. Simulations showed average efficiency of 84%, sensitivity of 86% and specificity of 82%. While the neural networks presented better sensitivity values, the discriminant functions had better specificity results. Also, it was noticed that the efficiency values for small sized classifiers were equivalent but as the classifier’s size increased the neural networks exhibited better results than the discriminant functions.