To delineate the association between interictal EEG discharges and fluctuations of dynamic local brain connectivity in focal epilepsy towards automatic detection of individualized epileptic networks.Evidence from epilepsy studies suggests links between IEDs and functional networks in presence of epilepsy^1,2. and that epilepsy may compromise local properties of brain networks^3,4. In this study, we propose a simultaneous EEG-fMRI data fusion framework in order to extract dynamic relationships between scalp-level interictal EEG spikes and local fMRI connectivity in focal epilepsy. We developed DRePS, a novel technique that measures local dynamic connectivity of task-free fMRI (tf-fMRI) at single repetition time (T_R) resolution⁵ (Fig. 1). The DRePS maps have the same temporal length and spatial resolution with the original fMRI signals. We hypothesized that there is a significant time-dependent link between scalp IED BAFs and DRePS in focal epilepsy.

We selected 7 patients with different types of refractory focal epilepsy (Table 1). They were scanned with 3T Siemens Skyra system while not performing any specific mental task with eyes closed. Functional data (200 volumes) were acquired using an EPI sequence with 44 interleaved 3 mm slices, T_R= 3 s, T_E= 30 ms, flip angle = 85^o, voxel size of 3×3×3 mm³ and an acquisition matrix of 72×72. A subject-specific target EEG electrode (out of 32 electrodes, monopolar montage, reference to FCz, sampled at 250 Hz, artifact corrected) was also selected to either represent dominant IEDs in markup-positive subjects or be close to the suspected epileptic focus for markup-negative subjects (Table 1). Then, a multi-step wavelet coherence analysis was adapted for each subject in order to investigate multi-scale nonstationary similarities between DRePS time series within gray matter and the target EEG channel:

1. Extraction of dynamic local rs-fMRI connectivity (i.e., DRePS) for each subject.

2. Uniform parcellation of grey matter into 4096 same-size regions of interest (ROIs)⁶ and averaging the DRePS time series over each ROI leading to an ROI-based mean-DRePS node-by-time matrix of size 4096x200.

3. Temporal resolution matching between EEG and tf-fMRI by taking the BAF of the target EEG channel⁷ within the most relevant frequency band and downsampling to the TR resolution leading to a target DS-BAF (Fig. 2A). The frequency band was chosen to maximize the correlation between the DS-BAF and the binary IED markups made by the EEG expert. The frequency interval was chosen so that it covers at least 90% of the BAF spectral power. The DS-BAF still retains a considerable amount of ‘Morlet similarity’ information in the wavelet domain (Fig. 2B).

4. Similarity assessment between target DS-BAF and each ROI-based mean-DRePS time series using Morlet wavelet transform coherence (MWTC) analysis^8,9 (Fig. 3A).

5. Significance testing of MWTC magnitudes at 95% confidence using the surrogate data method⁹.

6. Averaging of the suprathresholded MWTC planes within the cone of influence⁹ for each pair of ROI-based mean-DRePS signal and the target DS-BAF.

This procedure leads to a 3D map for each subject where each element is associated with the average significant MWTC magnitude between the corresponding ROI-based mean-DRePS time series and the target DS-BAF.

Statistically significant areas were identified in the time-frequency domain between mean ROI-based DRePS time series and the reference DS-BAF for all subjects (see Fig. 3A as an example). The subject-specific histograms of the mean suprathresholded magnitudes over ROIs represented heavy-tailed distributions with a few ROIs having significantly higher EEG-fMRI wavelet coherence (Fig. 3B). In 4 out of 7 subjects, the detected regions had overlap with the clinically suspected epileptic networks (Table 1). Parts of the default mode network (DMN) were also observed among the detected regions in some subjects (Fig. 4 as an example).Due to the temporal mismatch between EEG and fMRI, tracking local dynamic changes of functional connectivity at the time of epileptiform EEG discharges is not straightforward. Our proposed method offers a statistically rigorous time-frequency framework for investigating this possible link. It suggests that there is a scale-dependent and non-stationary coherence between DRePS and scalp IEDs, mostly occurring at ultra-slow to slow frequencies (~0.003-0.03Hz). This coherence tends to be greatest within parts of the suspected epilepsy networks and the DMN. It is in line with the previous intracerebral EEG-fMRI findings on possible relationship of IEDs and the DMN. This finding also implies that DRePS fluctuates in association with interictal epileptic brain activity. Thus evaluation of dynamic local phase synchrony, particularly using data-fusion with power envelope of scalp IEDs, may provide subject-specific markers of localized and transient brain connectivity disturbance in epilepsy.