Abstracts

Rationale: Simultaneous EEG-fMRI recordings have been very important for surgical planning in temporal lobe epilepsy (TLE). Interictal Epileptic Discharges (IED) are used as triggers for an event-related BOLD fMRI analysis. This usually yields several positive and negative BOLD responses (PBR and NBR) in the cortex. The origins of the PBR are understood as either the local response at the epileptic foci or as the response of distant areas belonging to a network affected by the IED. However, the physiological mechanisms underlying the NBR are still controversial and their clinical relevance in TLE have not been determined yet. Some theories propose NBR reflects decrease in neuronal activity [1], [2]. In TLE, this would correspond to a “shutting down” of the resting state networks in the brain during the IDEs. Another, based on a pure vascular/hemodynamic mechanism, is the co-called “blood stealing”, which advocates for blood reallocation from less to more demanding areas [3]. We believe both situations are not clinically relevant since they are not likely to not localize exclusively in the IED foci. We propose another mechanism: a pathological elevated tissue oxygen metabolic consumption, which is clinically important. Therefore, a quantitative discrimination between these three mechanisms is needed to increase the accuracy of IED-related BOLD imaging for surgical planning in TLE. In this work, we propose a parametric characterization of the temporal behavior of the NBR signal predicted by observable and parsimonious biophysical models. We show this parameterization can be used to distinguish among these three NBR signals. Methods: The three mechanisms were modeled with: (1) the balloon model [4] with a delayed IED-triggered fast suppression of neural activity followed by a slower recovery; (2) two windkessels models [5] connected in parallel, incorporating blood inertia (the equivalent of induction in circuits) in the main shared supplying artery to emulate blood stealing; and (3) the oxygen transport to tissue model (OTT) [6] with pathological elevated metabolic demand. We quantitively characterized them using the coefficient of linear autoregressive models (ARX). Results: The temporal profiles of the NBR exhibited significant differences among models. Figure 1 shows simulated random samples of the signal under real experimental clinical conditions. Other state variables relevant to the models are shown. The probabilistic mass of ARX coefficients for each model dwelled in regions that can be separated, as depicted in Figure 2. Conclusions: It is possible to discriminate between the three models based on the observed NBR. With this, we provide the means to quickly classify different putative IED-related NBR. In short, we will implement a fast method to estimate and classify these models in real data. This will contribute to an increase in the accuracy of EEG-fMRI methods aiding/guiding pre-surgical planning in TLE.References:[1] Shmuel et al (2006) Nat Neurosci 9, 569[2] Stefanovic et al (2004) Neuroimage 22, 771[3] Harel et al (2002) J Cereb Blood Flow Metab 22, 908[4] Buxton et al (2004) Neuroimage23 SUPPL. 1, 220[5] Mandeville et al (1999) J Cereb Blood Flow Metab 19, 679[6] Zheng et al (2002) Neuroimage 16, 617 Funding: This work is funded by the grant NIH 1R56NS094784-01A1