Development of diagnosis of epilepsy and electroencefalography analysis with machine learning algorithms

With the development of artificial intelligence technology, methods based on machine learning are preferred in the field of health to interpret of signals received from the human body. Electroencephalography (EEG), which measures the electrical activity of the brain, is used as the most important diagnostic method in epilepsy. Epilepsy as a neurological disease is common in children. Due to the mobility of children during EEG monitoring and the incomplete brain development makes it difficult to diagnose the disease. In this study, datasets obtained from 22 patients aged between 7 and 12 years, with a sampling frequency rate of 256 Hz, provided by Children Hospital Boston, were used. In this study, firstly, the amplitude-frequency characteristics of the frequency spectrum of the patients during the seizure and in the non-seizure state were extracted by determining the time intervals of the seizure-containing imaging in the datasets, using the discrete Fourier transform and data classification. When the artificial neural networks (ANN) techniques were used for machine learning classification ,92.01% accuracy rate was achieved. As a next step, EEG signals were resampled by 64Hz. By resampling, less storage space usage and higher processing speed can be achieved by minimizing the disadvantages of limiting factors such as storage space and processor power. For 64Hz by applying artificial neural networks machine learning and classification methods, 95.89% overall accuracy rate was achieved. When support vector machines (SVM) are used as a classification method, 97.27% accuracy for sampling rate 256Hz and 96.80% accuracy for sampling rate 64Hz was achieved.