Abstracts

Rationale: It is now becoming apparent that dense-array EEG (dEEG) is required to fully capture the clinical EEG features, such as inter-ictal spikes. Automated methods for detecting clinically significant events can be useful, particularly as they can ameliorate the review burden imposed by dEEG data, but their performance has not been evaluated with respect to EEG channel density. Understanding the effects of channel density on automated detection of clinical events will inform both algorithmic shortcomings as well as clinical interpretation of automated results. Methods: Five dEEG (256-channels) data sets were processed with an automated spike detection method optimized for dEEG. The dEEG data were also subsampled to sparser montages (10-10 and 10-20) and automatically processed with the same spike detection method. Results for dense- and sparse-array data were verified by a clinician. Results: The number of spike-like events detected decreased as a function of EEG channel count. The number of real spikes detected decrease, as expected, as channel density decreased. However, this was dependent on spike topography. Those spikes that occur at more ventral scalp locations were completely missed with sparse-array data Conclusions: Automated spike detection using dEEG data is practical and accurate. Like expert reading, accuracy is highly dependent on accurate sampling of the EEG potential field. Sparse-array EEG frequently misses clinically significant events, leading to poor performance for automated methods. Funding: No funding.