A quantitative EEG method for detecting post clamp changes during carotid endarterectomy
Meenakshi Mishra1, Muhammed Banday, Reza Derakhshani
1Department of Computer Science and Electrical Engineering, School of Computing and Engineering, University of Missouri at Kansas City, MO 64110-2499, USA.
Objective:
To apply univariate and multivariate analyses of EEG spectral features in order to detect subtle post-clamp changes during carotid endarterectomy, and to devise a data-driven, multivariate classification method for rapid, real-time detection of small post-clamp EEG changes.
Methods:
We used intraoperative EEG records of 27 patients who underwent carotid endarterectomy (CEA) and did not undergo a shunt per surgeon's and technologist's visual inspection of the electro-encephalogram (EEG). Furthermore, these records could not be flagged with r-sBSI (revised spatial brain symmetry index) or r-tBSI (revised temporal brain symmetry index) qEEG (quantitative EEG) metrics. Spectral energies of 10 s EEG snippets were recorded at 5 min intervals from 10 min pre-clamp to 20 min post-clamp. Using receiver operating characteristic curves, the power of standard EEG spectral bands in detection of post-clamp changes was determined. Next, we employed Fisher Linear Discriminant Analysis for classifier-guided multivariate feature selection and classification.
Results:
In univariate analysis, delta waves in non-frontal areas, irrespective of the laterality of the affected hemisphere, were found to be generally more powerful indicators of clamp-induced EEG changes. In multivariate analysis, cross-validation results for Fisher LDA produced subject-independent equal error rates as low as 32% when separating post-clamp signals from pre-clamp controls. Within our subject group, the intricate clamp-induced EEG signatures were predominantly anterior, bilateral, and had a strong delta rhythm presence. The overall post-clamp saliencies, both in univariate and multivariate analyses, were time-dependent.
Conclusion:
By applying automatic data-driven feature extraction and classification to short EEG records, it is possible to construct subject-independent computational models that can detect subtle post clamp changes possibly caused by small perturbations in cerebral blood flow. These subtle changes are missed by visual inspection of the EEG and by other quantitative EEG techniques such as r-sBSI and r-tBSI. Within our subject group, the intricate post-clamp EEG signatures were predominantly anterior, bilateral, and had a strong delta rhythm presence.
