Epileptic seizure detection from EEG signals on the basis of channels using multifractal, spectral analysis and long-short-term memory deep learning model

Irregular discharge of electrical activity in the brain is an important disorder. This disorder, known as epilepsy, is diagnosed by experts by examining electroencephalography signals. It is important to support the detection of seizures with artificial intelligence in the intensive work of experts and long electroencephalography shots. For this purpose, the artificial intelligence model proposed in this study and the electroencephalography signals in the CHB-MIT Scalp data set were analyzed and normal and epileptic data were classified. Multifractal, Instantaneous frequency and Spectral entropy methods were applied to electroencephalography signals recorded from 22 channels and divided into 10-second segments for analysis. With the analysis methods, a feature vector consisting of 162 components was obtained. The feature vectors of 175 normal labeled and 172 epileptic seizure electroencephalography recordings were created for the study. Afterwards, the dataset of the study was divided into two as 80% for training and 20% for validation. Long-short-term memory deep learning model was trained with 278 data allocated for training. The deep learning model was validated with the remaining 69 data. As a result of the validation, the highest success metrics on channel basis were obtained in the Fp1-F7 and CZ-PZ channels. The success metrics in these channels are Accuracy 85.29%, Originality 87.88%, Sensitivity 82.86%, False Positive Rate 12.12%, Sensitivity 82.86%, Negative Prediction Value 82.86%, F1-score 85%, respectively. .29, Matthews Correlation Coefficient 70.74%, False Negative Rate 17.14% and False Discovery Rate 12.12%. These results show that the proposed method can detect epileptic seizures on a channel basis. As a result, the developed artificial intelligence method is capable of assisting experts working in this field.