| Abstract [eng] |
Epilepsy is one of the most prevalent neurological disorder. The International League against Epilepsy defines this phenomenon as transient symptoms or signs resulting from abnormal, excessive, or synchronous activity of brain neurons. Quality of life can be improved by personalizing treatment, requiring an objective and quantitative assessment of the onset of the seizures and the body's response to treatment. Devices capable of detecting epilepsy would enable quantitative assessment of seizures, allowing physicians more objectively tailor and evaluate treatment efficiency. For such reasons, there is a need for a personalized system that is specific to the patient's symptoms and could detect epilepsy episodes. An expanding field of implantable devices and the evolving hardware open opportunities to consider the implementation of such a system and require new energy-efficient methods. This work introduces a personalized method for the detection of epilepsy from electroencephalographic (EEG) signals, potentially suitable for use in implantable devices. The method consists of an echo-state artificial neural network and a methodology to reduce the number of channels based on EEG signal power, line length, and nonlinear energy. This computationally time-efficient method for epileptic seizure detection shows diagnostic accuracy of 92,02 %, sensitivity of 88,68 % and specificity of 95,40 %. |
