Today, epilepsy is one of the most common diseases. According to the World Health Organization, approximately 65 million people in the world suffer from epilepsy. This disease can occur in all age groups; Therefore, its timely diagnosis can prevent the consequences of the disease. Since the use of electroencephalogram signals to diagnose diseases is a non-invasive method, this method is used to diagnose epilepsy and many other diseases such as Alzheimer's and chronic fatigue syndrome. Many factors affect the creation of defects during the recording of electroencephalogram data. For example, blinking and head and neck movements can damage the signals. In addition, reading the recorded signals is a specialized task and requires a trained person, so the use of automatic methods for diagnosing the disease and its stages can be helpful. So far, several methods have been presented to distinguish the stages of epilepsy from seizures, among which we can mention methods based on graphs and complex networks. These methods do not preserve all the properties of the time series, and a new approach is needed that, while preserving all the properties of the time series, also provides a suitable representation of it. In this research, a new approach to distinguish the stages of epilepsy from seizures is presented. This approach is based on building a multiplex network of visibility graphs. First, the length of the signals is reduced by using the scale factor to reduce the computational complexity of the problem. Then a visibility graph is made for each signal from each data set. These visibility graphs form a multiplex network. The coherence of the multiplex network is checked by the average overlap of the edges. A mapping of 100 layers is performed on one layer, which gives the complete and weighted graph as output. Removing minima helps the matrix to be sparse and as a result, suitable features will be selected. In the end, the characteristics of the network are given as input to the classifier. The eva‎luation of the proposed performance resulted in 100% precision, recall, and accuracy for the network obtained from the horizontal and natural visibility graphs. Also, the average accuracy of 95% was obtained for the network made of the circular visibility graphs; Therefore, the proposed approach preserves the properties of the time series for the signals of different stages of epilepsy and performs better in terms of precision, accuracy and recall than the previous works.

ناصر قديري مدرس

زينب مالكي

جلال ذهبي، فرزانه شايق