Theoretical and Computational Neuroscience
Author: Carlos Andrés Mateos | Email: mateos.andres@gmail.com
Carlos Andrés Mateos1°, Juan Manuel Miramont2°, Victoria Peterson1°
1° Instituto de Matemática Aplicada del Litoral (IMAL), CONICET-UNL, Santa Fe
2° Postdoctoral Researcher at Centre de Recherche en Informatique, Signal et Automatique de Lille (CRIStAL), France
Individuals with drug-resistant epilepsy are usually treated with invasive therapies like resective surgery or neurostimulation. Responsive neurostimulation (RNS) is a closed loop interface that monitors the electrical brain activity and applies local electrical current when seizure-like patterns are detected. This treatment is designed to work directly at the seizure foci like a heart defibrillator does in cardiac arrhythmia. However, evidence reveals that this may not be its primary mechanism of action. The slow time course of seizure reduction with RNS therapy provides some evidence for a long-term neuromodulatory effect on brain networks that generate seizures. Only effects at some latency after stimulation, called indirect electrographic seizure pattern modulation (iESPM), were associated with clinical improvements. These iESPMs are characterized by changes at the time-frequency domain and were described by visual inspection of expert epileptologists. Due to the large amount of data and escarse of experts’ time, there is a need to develop unsupervised methods to identify iESPM with high precision. Here, we explore Scattering Transform (ST), a tool that constructs invariant, stable, and informative signal representations by cascading wavelet modulus decomposition followed by a low pass filter. One-class support vector machines are then used upon the ST features to build an unsupervised model for estimating changes in seizure patterns along the RNS therapy.