Localizing focal brain injury via EEG spectral variance. (July 2021)

• Record Type: Journal Article
• Authors: Khanmohammadi, Sina; Laurido-Soto, Osvaldo; Eisenman, Lawrence N.; Kummer, Terrance T.; Ching, ShiNung
• Is Part Of: Biomedical signal processing and control. Volume 68(2021)
• Publication Date: 2021-07
• Subjects: Brain injury -- Injury location -- Electroencephalography -- Spectral variance; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102746
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A new EEG analysis method is proposed to quantify brain region interactivity. The proposed method is used to localize lesions from scalp EEG data. The results suggest a framework for noninvasive assessment of brain injury. Abstract: In this study, we consider the problem of localizing focal brain injuries from surface electroencephalogram (EEG) recordings. To this end, we introduce a new analysis technique termed frequency-based intrinsic network dynamic reactivity (FINDR), which quantifies the extent to which different brain regions (defined in EEG channel space) are responsive to each other in terms of their frequency-domain activity. The technique generalizes the idea of EEG reactivity, a measure of how well EEG signals react/respond to exogenous stimuli. In the present work we generalize this notion to endogenous 'stimuli, ' defined as short-time window frequency domain motifs that are most predominant on a per channel basis. For each of these predominant motifs, we quantify the variance of the activity in all other channels as a measure of 'intrinsic reactivity', under the hypothesis that channels proximal to injured regions will be systematically disassociated from other brain areas. We use this method as a front-end to a neural network classifier to predict injury location in a cohort of etiologically heterogeneous comatose patients. We achieve a 0.6 correlation between the predicted injury location and the actual brain injury. These results suggest a possibility of precise localization of brain injury using EEG.