Automatic recognition of chewing noises in epileptic EEG based on period segmentation

Abstract

Automatic detection of Interictal Epileptiform Discharges (IED) has the great significance in diagnosis of epilepsy and relieves the heavy workload of inspecting electroencephalogram (EEG). The artifacts in epileptic EEG strongly affect the detection results, especially in the form of chewing noises. This paper proposes a novel time-domain approach to process chewing noises in epileptic EEG signals based on period segmentation. Firstly, merger of increasing and decreasing sequences (MIDS) is employed to segment EEG periods. This period segmentation approach considers information of waveform rather than a single sample point and applies human vision principle. Experimental results show that the performance of merger and period segmentation is close to clinical visual detection. Secondly, chewing noises are recognized in epilepsy patients׳ EEG following period segmentation. To reduce the false recognition rate, classification results of four channels F3, F4, F7 and F8, which acquire high sensitivity and low false recognition rate, are fused to determine fragments׳ classification by weighting. With this method, EEG recordings of 20 epilepsy patients were analyzed. The results showed that a sensitivity of 98.10% and a false recognition rate of 0.01375/s were achieved. It demonstrates that the proposed approach performs well in automatic recognition of chewing noises in epileptic EEG, which is of crucial importance to automatic detection of IED.