Mathematical model for epileptic seizures detection on an EEG recording

Purpose of this study - analysis of the possibility of using convolutional neural networks as a model for detecting epileptic seizures on real EEG data Methods. In this paper, wavelet analysis is used for time-frequency analysis. To localize epileptic discharges, the task of detecting them was reduced to the classification task and the ResNet18 architecture of neural network was used. Techniques were used to augment and balance the biomedical data dataset under consideration. Wavelet analysis is used for time-frequency analysis. To localize epileptic discharges, the problem of their detection was reduced to the classification task, and the ResNet18 neural network architecture was used. Techniques were used to augment and balance the considered biomedical dataset. Results. Convolutional neural network can be successfully used to detect epileptic seizures, a method of postprocessing the results of primary detection is proposed to improve the quality of the model. It is shown that the developed model demonstrates high accuracy in comparison with other methods based on classical machine learning algorithms. The value of the F-1-score metric reaches 0.44, which is a high value for classification of the real biological data. Conclusion. The presented model based on a convolutional neural network for detecting epileptic seizures on an EEG recording can become the main one in medical decision support systems for epileptologist.