A Novel ERP Pattern Analysis Method for Revealing Invariant Reference Brain Network Models (pp. 295-317)
Authors: Amit Reches, Dan Kerem, Noga Gal, Ilan Laufer, Revital Shani-Hershkovitch, Dalia Dickman, and Amir B. Geva
Abstract: Background: Objective and reliable neuro-electrophysiological methods for the longitudinal monitoring of a patientís cognitive state are scarce. Since baseline neuro-electrophysiological measurements are generally not performed as standard practice, this type of longitudinal monitoring requires an invariant normal reference to which the individual brain activity may be scored, with sufficiently high within-subject repeatability.
Methods: Group-common functionally connected Reference Brain Network Models (RBNMs) were extracted with the Brain Network Activation (BNA) Technology from multi-channel ERPs of 120 young healthy subjects (Reference Group) who underwent the three-stimulus Auditory Oddball Task. The test-retest repeatability of an age-matched group of 116 subjects (Database Group), whose individual brain activity on the oddball task was scored on the RBNMs, was determined and the Standard Error of Measurement (SEM) computed as a measure of Minimal Important Clinical Difference. The general applicability was cross-validated on 36 healthy patients of a wider and marginally overlapping age-range (Validation Group).
Results: Intraclass correlation values of scores on repeated tests in the Database Group ranged between 0.58-0.81. SEM values ranged from 14.1-17.9, on a score scale of 0-100. Scoring, repeatability and SEM applicability were all successfully validated on subjects from the Validation Group, yielding values that were similar to or higher than those of the Database Group.
Conclusions: The BNA generated brain network models are largely invariant and repeatable and as such may be useful for diagnosis and follow-up of disease progression and treatment management. To further test the algorithmís utility, additional research should be conducted on diseased populations.