Research And Design Of Automatic Epilepsy Diagnosis System Based On Interictal EEG

Electroencephalogram(EEG) contains lots of pathological and physiological information and is an effective tool for clinicians to detect and diagnose lesions and damage of the brain neural system. Since epilepsy is a neural system disease caused by abnormal electrical activity of ne urons, EEG is a basic method for epilepsy diagnosis. Traditional epilepsy diagnosis relies on well- trained doctors to screen lengthy EEG recording, identify epilepsy-related abnormal waveform. This method is tedious and consuming. Although there are many a utomated seizure detection systems to aid clinicians for epilepsy diagnosis, long-range EEG recording is still required to be taken for the presence of enough seizure activities. Recoding long-range EEG, howerer, is quite difficult in underdeveloped areas lacking of medical resources. Therefore, extracting and analyzing features from EEG signals and design automatic epilepsy diagnosis system will be of great significance, especially if the system can just rely on interictal EEG signal to achieve epilepsy diagnosis.In this paper, we aim to study EEG feature extraction, design a Probabilistic Neural Network-based classifier and realize an automatic epilepsy diagnosis system. Automatic epilepsy diagnosis is composed of feature extraction module and classifier module. After brief summarization of EEG signa l processing methods, we chooses power spectrum-based spectral characteristics and time sequence-based dynamic characteristic, which constructed 24-dimension feature vectorsas the input of the classifier. The e xperimental results show that the system can effectively discriminate the difference among EEG segments from the healthy, interictal EEG segments from the patientsand ictal EEG segments from the patients, as well as provides reference information for location of the epileptic foci. In addition, the possible system optimization is discussed from the perspective of feature and spread selection.In this paper, we also use scalp EEG data with more artifacts and noise to detect the automatic diagnosis system. Through the combination of Probabilistic Neural Network and a majority voting mechanism, the system can successfully distinguish between EEG from the healthy and interictal scalp EEG from the patients and achieve an 99.38% accuracy. This result proves the feasibility of using interictal scalp EEG to automatically diagnose whether a person is epileptic.