Modelling of EEG dynamics under ECT using Square-Root Cubature Kalman filter -

Abstract

Electroconvulsive therapy (ECT) is a clinical technique used for treating patients with severe drug resistant depression. The therapy consists of administering brief current pulses through two stimulating electrodes placed on the subject scalp thus creating short-lived therapeutic seizure activity in the underlying brain tissue. Despite being a superior treatment method, ECT efficacy and cognitive side effects remain influenced by many parameters including electrode position and configuration as well as the applied current intensity, duration, and polarity. Over the many years, several innovations were introduced in terms of ECT stimulation parameters to reduce side effects while maintaining the antidepressant quality. Importantly, advancement was experimentally driven with limited understanding of the role of different brain areas in initiating, recruiting and maintaining ECT-induced seizures. In this thesis, we propose the use of a nonlinear interaction model to explain multichannel scalp EEG recordings as the outcome for interacting cortical areas, and thus aid in identifying key cortical players in initiating efficient seizures. The interaction models are built from modified neuronal population activity models whose dynamics can reproduce basic features of ECT-induced seizures within local areas and across distant cortical areas. The dynamic models are then used to identify the strength and directionality of effective connections between 4 areas of the brain in three various EEG states: normal, ictal and post-ictal. The Square-Root Cubature Kalman filter, a recently introduced nonlinear estimation technique, is demonstrated to correctly estimate effective inter-areal connection in simulation models. The method is subsequently applied on real EEG recordings obtained in the AUBMC psychiatry department for patients under the FEAST configuration of treatment.