Titulo: Epileptic seizure prediction by machine learning

Abstract: Some patients who suffer from epilepsy are resistant to medication, and surgery is not an option. These patients are subjected, for example, to accidents because seizures can happen without any warning, and having support technology to manage drug-resistant epilepsy (DRE) will be of utmost importance. Consequently, seizure prediction algorithms, with high sensitivity and high specificity, embedded in wearable devices would help these patients in their daily lives.
Since the seventies of the last century, several algorithms for seizure predictions have been developed, mainly based on the electroencephalogram (EEG). The first algorithms were based on single EEG features, and decisions were made by defining thresholds. More recently, machine learning (ML) algorithms were applied, with the computational power to accommodate several features and raw data and to decide by defining linear and non-linear decision surfaces.
This seminar starts by presenting a brief historical perspective of the field. Then, it provides an overview of the ML pipeline, highlighting essential developments at different stages (preprocessing, feature extraction, dimensionality reduction, and classification). I highlight advances proposed by other authors but, more importantly, contributions by my team. The seminar ends with a balance of past, actual, and expected future developments.

Les esperamos