2.1 Dataset, preprocessing, and brain network extraction

The dataset contains two resting-state, RS1 and RS2, sessions acquired from 10 ACA patients and 14 controls. Prior to RS2, a silent word generation task was performed for 7 min. Dynamics features that show significant differences between the two populations in RS1 as well as in RS2, would show robustness of the features as biomarkers. Indeed, since the task in-between resting-states was cognitively demanding, such features would have discriminant power regardless of the brain fatigue the participants experienced. On the contrary, the difference in findings between RS1 and RS2 could shed light on the cognitive reserve capacity of the ACA patients, which can be of interest too. Table 1 shows the demographics of the participants, the type of epilepsy, and the deterioration scores, i.e. cognitive decline, for Full-Scale IQ, Perceptual Reasoning Index, and Verbal Comprehension Index (respectively FSIQ, PRI, and VCI). Deterioration scores were calculated by subtracting the estimated premorbid IQ scores from the actual scores, i.e. WAIS-IV (actual) IQ-scores—OPIE-IV (premorbid) IQ-scores [44, 45].

Note: * = p < 0.01 sign. difference between groups. WMI: Working Memory Index. PSI: Processing Speed Index.

^a Dominant seizure type is determined for the two years preceding neuropsychological assessment.

^b The prescribed daily dose of antiepileptic medication divided by the defined daily dose.

^c The deterioration scores = [WAIS-IV (actual) IQ-scores—OPIE-IV (premorbid) IQ-scores]

The T1-weighted anatomical images were acquired using a 3.0 T imaging system (Philips Achieva) with a 3D Fast Field Echo (FFE) sequence: echotime (TE) = 3.8 ms, repetition time (TR) = 8.3 ms; Inversion time (TI) = 1035 ms; field of view (FOV) = 240x240 mm², with 180 sagittal slices; flip-angle (FA) = 8 deg.; and voxel size = 1x1x1 mm³; with SENSE factor of 1.5, and scan duration of 6:02 min. Functional MRI-data were acquired using multi-echo echo-planar imaging (ME-EPI) sequences with 3 echoes: TEs = 12, 35, 58 ms, TR = 2 s; FA = 90 deg.; SENSE factor = 2.7; 208 dynamics for a total of 7 minutes; 27 axial slices (with no gap), 64x64 matrix FOV, with a 3.5x3.5x4.5 mm³ voxel size. The raw multi-echo fMRI data were first preprocessed following the pipeline of [46], using the python script meica.py (available at www.bitbucket.org/prantikk/me-ica). Multi-echo- (ME)-ICA cleaning was applied in order to denoise each individual fMRI scan [47]. ME-ICA cleaning has been proven to be the most robust denoising method for resting-state fMRI, and tend to improve substantially effect sizes and statistical power [48, 49]. Multi-echo cleaned data were further processed using FEAT (FMRI Expert Analysis Tool) Version 6.00, part of FSL (FMRIB’s Software Library, www.fmrib.ox.ac.uk/fsl). The following pre-statistics processing was applied; spatial smoothing using a Gaussian kernel of FWHM 5.0mm; grand-mean intensity normalization of the entire 4D dataset by a single multiplicative factor; high-pass temporal filtering (Gaussian-weighted least-squares straight line fitting, with sigma = 50.0 s). The ME-ICA cleaning preprocessing led to an average, among all participant/scans, of 18 activity-related independent components (ICs or networks). Therefore, we chose 18 degrees of freedom for our group-ICA decomposition. The 18 spatial maps corresponding to functional resting-state networks (RSNs) were extracted using group-ICA followed by dual regression as implemented in FSL [50, 51]. After the dual-regression steps, we obtained all subject/session-specific RSN maps and their associated time series. This brain network extraction step is depicted in Fig 2.

After visual inspection and a goodness-of-fit function implementation [52], to match these maps with the RSN template from [11], 4 noise-related RSNs were discarded, and 14 relevant functional brain networks were kept for further analyses. The time series associated with these 14 RSNs were then used to extract our brain networks features as explained in the next section. The 14 networks, or Independent Components (ICs), and their labelling are depicted in Fig 3.

LVISU–lateral visual network; MVISU–medial visual network; DMN–default mode network; AUD–auditory network; VAN–ventral attention network; LSM—lateral sensory-motor network; DAN–dorsal attention network; OVIS–occipital visual network; SSM–superior sensory-motor network; CB–cerebellar network; FPL–left fronto-parietal network; FPR–right fronto-parietal network; SN–salience network; CING–the cingulate network.