Abstract de conférence
Résumé : Objective: Brain connectivity is a promising approach to electroencephalography analysis for diagnosing neurological disorders. However, numerous methods have been designed to estimate it, and no standard has been adopted [1, 2]. This work aims to fill this gap by defining a framework for methods selection. Methods: In this framework, major brain connectivity methods (see Figure 1) were compared based on their ability to discriminate sub-populations in the patient pool of the given study. The comparison was conducted in two stages. First, a statistical test was performed between subpopulations’ connectivity to filter out non-significant results. Then, the remaining methods were used to create classification models. The method with the highest classification performance was then selected. Such procedure was applied to epileptic seizures, continuous slow-wave sleep syndrome, and coma. Results: Best discrimination powers were obtained for the methods correcting for the volume conduction EEG artifact, in the Delta frequency range [1-4 Hz]. Nevertheless, similar trends in the differences between groups are seen among methods. Conclusions: This work proposes a selection framework for brain connectivity estimation. While one method can be ranked above the others, connectivity is reassuringly consistently estimated across techniques.

Documents en relation

DI-fusion