Abstracts

Rationale: EEG-fMRI localizes epileptic activity (EA) with the high spatial resolution of fMRI. However, EEG recording in the scanner is cumbersome and the method cannot detect BOLD changes caused by EA not apparent on scalp EEG. 2D temporal clustering analysis (2D-TCA) is a relatively new fMRI-based EA localization technique that breaks BOLD activity into components based on timing, finding BOLD changes without the help of the EEG. This study is an investigation into the ability of 2D-TCA to detect epileptic activity of various frequency, extent and location. Methods: fMRI scans containing simulated EA were created by adding a BOLD signal, simulated using values based on current knowledge of BOLD responses to EA, in specific regions of interest (ROIs) in scans from control subject. Simulated spikes consisted of all combinations of the following characteristics: 1, 5, or 10 spikes per 6 minute scan; hemodynamic response function (HRF) amplitudes of 0.5-2% above baseline, in 0.25% increments; ROI sizes of 12, 33, 36, 43, 80, and 53 voxels (each voxel being 5x5x5 mm). In addition to spikes, a single short seizure was simulated by a 5 s event with the above HRF amplitudes in ROIs of 43, 53, and 63 voxels. A total of 756 simulated scans were created, each with 4 ROIs for a total of 3024 forms of simulated BOLD responses. Using these simulated scans, the limits of 2D-TCA, in terms of responses that it can detect, were investigated. A slightly modified version of the 2D-TCA algorithm as developed by Morgan and Gore (Hum Brain Mapp 2009; 30: 3393-405) was used. Results: 2D-TCA creates a number of components for a given scan. Fig. 1 shows, for various HRF amplitudes, the true positive rate (TPR) associated with that component which best describes the given form of EA, i.e. that component whose activation map has the highest TPR and lowest false positive rate (generally on the order of 0.01). For reasonable detection, 1 spike/scan is insufficient, while 5 spikes/scan requires an HRF amplitude of 1.5%, 10 spikes/scan 1.25%, and a 5 s event at least 1%. Although 2D-TCA can create components that precisely describe EA with these characteristics, it is important to note that it also creates many other components not associated with the EA, some of which will also create significant activation maps. Fig. 2 is a box plot showing, for scans simulated with an HRF amplitude of 1% or larger, the number of components whose corresponding activation maps contain clusters larger than a range of sizes. Even with a cluster size threshold of 215 voxels (just under the largest simulated ROI), 2 to 9 components still create significant activation that in some cases could be interpreted as arising from EA. Conclusions: We have demonstrated that 2D-TCA is able to effectively detect some forms of epileptic activity (large enough in amplitude and extent), but it can only be effectively used to validate localization by other means or to create hypotheses as to where this activity may be occurring because it also detects some responses not caused by epileptic discharges. Supported by NSERC CGSM, CIHR MOP-38079.