Abstracts

Rationale: Epilepsy and migraine are known to be comorbid conditions even when there is no preceding head-trauma involved [Leniger, Headache. 2003 Jun;43(6):672-7; Gorgi, Brain Res Rev 2001, 38(1-2):33-60]. In severe cases, CSDs are known to cause neuronal death [Pietrobon, 2014 Nature Reviews Neuroscience; 15; 379?"393], and thus motivate our work that aims at developing engineering and algorithmic tools for reliable detection/diagnosis of CSD as well as deeper understanding of the connection between CSD and epilepsy. A major obstacle to understanding this connection is that CSD diagnosis today requires surgical placement of Electrocorticography (ECoG) arrays on the brain surface, which is an invasive procedure that most epilepsy sufferers do not receive. On the other hand, Electroencephalography (EEG) ?" a noninvasive modality ?" is commonly used as a diagnosis tool in epilepsy. Recent studies show that EEG can indeed detect some CSDs [Hartings, Ann Neurol. 2014 Nov;76(5):681-94]. However, because of the diffusive effect of the layers between brain and the scalp (which act as spatial low-pass filters), EEG detection reliability is substantially smaller than ECoG's. The question we ask is: can high-density EEG, coupled with advanced data processing, improve noninvasive detection of CSD waves? We expected that the diffusive effect of head layers can be compensated to a degree by increasing the EEG electrode count. Methods: In this work we propose a spatial equalization-based approach that appropriately inverts the brain-to-scalp spatial low-pass filter, with improved inversion as the number of electrodes increase. The key difficulty lies in performing this inversion with a finite number of sensors. As a first step, the work relies on simulations on idealized and simplified (spherical) head models. By paying attention to the highest spatial frequency that can be recorded with a given electrode count, we partially invert the spatial low pass filter. This requires assuming that electrodes themselves act as spatial filters [Nunez-Srinivasan 2005; Oxford Univ Prs], and we further idealize this spatial filtering. Results: We expect that the accuracy of recovering the cortical signal from scalp measurements improves with increasing number of electrodes [Grover et al., Allerton 2015], and this is indeed what we see. Consequently, our algorithm (for high-density EEG measurements) enables improved visual detection using EEG in cases when visual inspection might fail (e.g. when the data is noisy or when the spreading depression causes only a mild depression on the brain surface). As an example, the attached figure shows a simulated CSD on a 4-sphere head model where the depression causes loss of 80% energy. Visual reading of unprocessed EEG is hard, but signal projected onto the brain surface brings out the hidden depression! Conclusions: Our simulations strongly suggest that high-density EEG, in conjunction with the proposed algorithms, offers a way to make CSD detection more reliable using EEG sensing, allowing for more correlation and causation studies between CSD and epilepsy and improved epilepsy, migraine, and TBI treatments. Admittedly, the work is preliminary in its use of simplified models which admit simpler signal processing techniques. We do note that the projection on the cortical surface performed here can be performed on realistic brains through solving Poisson's equation with boundary conditions specified by potential on the scalp surface, and visually examining the solution on the cortical surface. Our ongoing work addresses realistic brain models, and future work should address data recorded from sufferers of CSD. Funding: CMU BrainHUB-ProSEED grant; Henry L. Hillman Presidential Fellowship (for PV); NSF CAREER award (for PG)