Abstracts

Rationale: Functional near infrared spectroscopy (fNIRS) can be combined with electroencephalography (EEG) to continuously monitor the hemodynamic signal evoked by epileptic events such as seizures or inter-ictal epileptiform discharges (IEDs). Currently, the majority of analytical methods used for EEG-fNIRS analysis make assumptions on the hemodynamic response function (HRF). Evidence has accumulated showing that the HRF to neuronal activity can vary across different subjects as well as across different regions. Hence, studies conducted using a canonical HRF may fail to achieve the optimal sensitivity of the technique. Although many methods have been proposed to model specific HRF, few of them were applied to the analysis of EEG-fNIRS data. Methods: EEG-fNIRS data from 6 patients with refractory focal epilepsy were selected from a database of 41 patients, based on showing many clear IEDs and unambiguous focus localization. A linear model was introduced to explain the hemodynamic changes associated with IEDs in the data. If IEDs were highly frequent in the recording, a quadratic term was added to quantify nonlinear effects in the response. A deconvolution algorithm was applied to reconstruct channel-specific HRFs for each patient. The reconstructed HRFs were fitted with the difference of two gamma density functions, and were compared with the canonical HRF. A general linear model (GLM) with regressors formulated by channel-specific HRFs was applied to produce contrast topographical maps of IEDs. Comparison was then made between contrast maps obtained with specific HRFs and maps generated under a fixed canonical HRF framework. The sensitivity of EEG-fNIRS in localizing the epileptic focus region on each patient was tabulated for both techniques. Results: We observed that: (1) the channel-specific HRF deconvolved from EEG-fNIRS data can vary greatly from the canonical one, see table.1. In all our six cases, the hemodynamic response to an IED started before the IED could be captured on a scalp EEG, which confirms the presence of early phases in the response to IEDs. (2) For three of four patients who had very frequent IEDs during their recordings (> 15/min), the addition of a nonlinear term in the channel-specific model improved the goodness of fits. (3) The contrast maps obtained with the channel-specific HRF generally provided better detection of activated sites related to IEDs than those with a fixed canonical HRF. As a result, the sensitivity in the localization of epileptic focus regions was also improved. Conclusions: This work highlights the importance of modeling channel-specific HRF in the analysis of EEG-fNIRS data of epileptic patients. Specific HRF associated with IEDs were deconvolved directly from combined EEG-fNIRS data thereby improving detection, while adding a quadratic term to account for the nonlinear effects in the response was reasonable when IEDs were highly frequent. We conclude that deconvolved channel-specific HRF may be more adequate than a canonical HRF in EEG-fNIRS data analysis, as a higher sensitivity in the localization of epileptic focus region can be achieved.