Examining the Potential of Sequence Patterns from EEG Data as Alternative Case Representation for Seizure Detection

The management of EEG signals for disease diagnosis has been traditionally addressed by extracting features from the bio-signal data, either in the time domain, frequency domain, or time-frequency domain; and recently, deep learning has been used to find patterns that are later used for a classification method. In this work, we propose the use of sequence pattern mining algorithms combined with case-based reasoning (CBR). First, patterns are mined from EEG data. Next, cases are built based on binary features representing the patterns. A CBR system uses the cases to detect epileptic seizures. Experimentation is carried out with the CHB-MIT scalp EEG database of Physionet concerning epileptic seizures. The study reveals distinctive patterns related to the preictal phase of EEG signals, indicating the potential for accurate prediction of epileptic seizures.