Optimization of epileptic seizure detectors through machine learning techniques

Abstract

Among the various research niches in epilepsy, scientists and computer engineers have been contributing with methods for the detection and prediction of seizures, as well as with the location of epileptic foci. A common challenge is represented by the inherent variability of the disease and the scarcity of data containing seizure onset time and length, usually making most optimized solutions found for a particular patient of poor performance when applied to other patients, highlighting the need for developing software solutions with greater inter-individual generalization capabilities and requiring progressively less data to adapt to unseen epileptic patients. Aiming at providing a more adapted initial condition as well as enhancing the initial training progress of seizure detectors for new patients, this proposal will make use of two machine learning strategies: (1) transfer learning, which explores aspects of the learning process from source-patients in order to improve the detection performance on a target-patient; (2) feature selection for the automatic extraction of highly discriminatory features from normal and abnormal brain activity, also including topological attributes derived from models of brain connection, such as synchronization graphs involving active cerebral regions. Therefore, this proposal does not consider seizure forecasting techniques, being focused only on seizure detection. Initially, open databases available for scientific investigation, such as the Physionet platform, will be the main source of data. However, it is not discarded the cooperation with other research groups associated with the CEPID/Fapesp BRAINN project, to which this proposal is linked. (AU)