The detection rate of subtle epileptic lesions is also affected by the magnetic field strength. Three Tesla (3 T) MRI is thought to be more sensitive to subtle epileptic lesions in some patients who are MRI-negative at 1.5 T [21-23], especially when phased array surface coils are used [24]. Furthermore, the use of ultra-highfield (7 T) MRI has increased for in vivo brain imaging, offering a higher signal-to-noise ratio (SNR) and contrast-to-noise ratio, which can result in higher spatial resolution. Compared with conventional field-strength MRI (1.5–3 T), the diagnostic value of 7 T MRI has been demonstrated for polymicrogyria [25], moyamoya disease [26], hippocampal sclerosis [27], brain tumors [28], stroke [29], and multiple sclerosis (Fig. 1) [30]. In a prospective study, 7 T MRI was performed for 21 consecutive patients (17 adults, 4 children) with a single seizure onset zone who were previously considered MRI-negative. GRE and FLAIR images revealed a distinct lesion in 29% of patients (1 child), with the histopathological diagnosis of FCD in all patients who underwent surgical resection [31]. Nevertheless, there are few reports of the clinical application of 7 T MRI in pediatric epilepsy [31,32]. High cost, increased susceptibility artifacts, image inhomogeneity, and patient comfort and safety problems including increased radiofrequency power deposition, dizziness, nausea, and metallic taste must be considered when using ultra-high-field MRI in clinical practice for pediatric epilepsy [33].

Diffusion tensor imaging (DTI) has been reported to better visualize the white matter fiber tracts by estimating the direction, orientation, and anisotropy of the white matter fiber tract based on the differential mobility of water molecules along versus perpendicular to the fiber bundles. DTI tractography can delineate 3D-specific white matter tracts from the preferential selection of anisotropic water diffusion by connecting fibers that pass through anatomically placed regions of interest. This technique can be useful for evaluating white matter tract alterations associated with epilepsy [34,35]. It can also be used as a presurgical evaluation tool to avoid eloquent pathways such as the corticospinal tract for motor function, optic radiation for conveying visual information, or arcuate fasciculus for language, particularly in children for whom invasive presurgical evaluation is challenging (Fig. 2) [36-38]. DTI has also been used to visualize the functional connectivity of seizure tracts by developing a topographical map of seizure network nodes [39]. However, tractography has its limitations, i.e., it tends to have high variability and low reliability depending on technical factors such as data acquisition, fiber modeling, fiber reconstruction, and expert interpretation including choices of seed regions and stopping thresholds [40]. In addition, in tractography, each voxel is considered as a single diffusion compartment even though it could be a combination of white matter, gray matter, and cerebrospinal fluid (CSF). The neurite orientation dispersion and density imaging (NODDI) technique were recently proposed to overcome this limitation. This technique assumes that each voxel is a combination of intracellular, extracellular, and CSF components. This model allows the determination of individual contributions of free water (CSF), neurite density (axons and dendrites), and orientation dispersion (myelination) for fractional anisotropy, one of the parameters of DTI. In a preliminary study, NODDI was used to detect subtle FCD by identifying reduced intracellular volume fraction, a marker of neural density, in the FCD lesion (Fig. 3) [41].

Although a visual inspection of high-resolution MRI can improve the detection of subtle epileptogenic lesions such as hippocampal sclerosis and FCD, a conventional visual analysis has limited ability to recognize the presence and extension of subtle lesions. Advanced MRI postprocessing techniques have been developed to identify subtle structural abnormalities by quantifying the anatomic features of the brain without the need for time-consuming manual measurements or subjective visual assessments.

Voxel-based morphometry (VBM), the most popular morphometric technique to date, can be used for T1-weighted images to extract gray matter and white matter maps for comparison with a normal control database. The use of VBM has indicated that lesions with increased gray matter concentration are in agreement with FCD lesions visible on MRI in 63%–87% of cases [42-44]. VBM also allows the enhanced visualization of FCD with a junction map to highlight blurring of the gray-white matter boundary and an extension map to delineate abnormally deep sulci (Fig. 4) [45]. Surface-based morphometry techniques reconstruct the cortical surface and allow the measurement of morphologic features such as cortical thickness, curvature, and depth and texture features (blurring of the gray-white matter boundary) to detect the abnormal gyral and sulcal patterns of FCD that could not be detected on VBM [46]. T2 relaxometry refers to the quantitative measurement of T2 relaxation times in the brain. This method is useful for detecting signal changes in hippocampal sclerosis in combination with hippocampal atrophy detected by automated volumetry in temporal lobe epilepsy [47,48].

Various methods of structural postprocessing analysis have been described in detail in the literature [16,49]. Hong et al. [50] recently reported that an automated classifier relying on surfacebased features of FCD morphology and intensity detected FCD type II in 14 of 19 MRI-negative patients with 74% sensitivity and 100% specificity. In the future, the automated detection of subtle lesions by morphometry and machine learning methods with structural imaging data showed a promising role in presurgical decision-making about the epileptogenic zone in patients previously regarded as MRI-negative [51].

In patients without definite abnormalities on conventional MRI, functional neuroimaging may aid our understanding of the mechanisms underlying the epileptic process, leading to a more accurate localization of abnormal neuronal activities by measuring in vivo perfusion, metabolic changes, or neurotransmission abnormalities associated with the seizure. In fact, 18F-fluorodeoxyglucose (FDG)-positron emission tomography (PET) is the most widely used modality for detecting focal areas of relative hypometabolism, which may reflect functional disturbances in cerebral activity associated with the epileptogenic zone. The regional cerebral hypometabolism identified by FDG-PET often has a wider distribution than that of the seizure focus, which can represent the focus and the projection areas of seizure activity [52]. This could determine the extent of resection difficulty. FDGPET is also a useful tool for predicting seizure control after epilepsy surgery; the extent of resection of the hypometabolism on FDG-PET may be associated with better a postoperative seizure outcome [53-55].

Single-photon emission computed tomography (SPECT) is a nuclear imaging modality that can show the cerebral perfusion before, during, and after a seizure using a radioactive tracer. To increase its sensitivity, the subtraction of interictal from ictal SPECT coregistered to MRI techniques is used to complement visual interpretation to identify subtle seizure-associated changes. However, the delayed injection of the tracer and a short seizure duration can result in a variable pattern of perfusion changes according to the seizure propagation rather than the seizure onset zone. Therefore, the use of PET or SPECT in the peri-ictal period is often difficult in the pediatric population due to the requirement for a separate evaluation and administration of radiopharmaceuticals.

Arterial spin labeling (ASL) is a perfusion MRI technique that assesses cerebral blood flow by magnetically labeling inflowing blood water as an endogenous tracer [56]. By comparing the signal difference between labeled and control images, ASL can identify peri-ictal perfusion abnormalities and provide complementary information about the seizure onset zone. The perfusion map obtained from ASL corresponds well to PET imaging and electroencephalography (EEG) results of intractable epilepsy [57,58]. Although ASL suffers from a relatively low SNR, advantages such as noninvasiveness, easy accessibility, high reproducibility, and the absence of intravenous contrast agents make it particularly suitable for pediatric epilepsy. In our study, 24 of 36 (66.7%) children with focal seizures showed ASL abnormalities during the post- or interictal periods; of them, 19 (52.8%) showed concordance between ASL and clinical seizure focus (Fig. 5) [59].

Functional MRI (fMRI) is based on the principle that metabolically active regions of the brain have increased blood flow, known as activation-flow coupling. It is used to identify blood oxygenation level-dependent (BOLD) signals, which are determined by the local concentration of deoxygenated and oxygenated hemoglobin in response to neural activities. Recently, fMRI has become a promising noninvasive alternative technique to the Wada test (intracarotid amobarbital test) in the adult population to guide the resection plan by identifying critical areas of the brain near the seizure focus and minimize the risk of new deficits [60]. The patient is asked to perform a series of tasks during a task-based fMRI study. Although strong preliminary evidence shows that fMRI can be used to lateralize and localize language and motor function in pediatric epilepsy surgery candidates, functional mapping via task-based fMRI is challenging in children due to a lack of cooperation, developmental immaturity, and age-dependent cerebral plasticity [61]. Furthermore, it remains unclear which language and memory tasks produce optimal activation and how it is statistically quantified. Further studies specifically targeting pediatric epilepsy populations are needed. Resting-state fMRI has been used to localize the eloquent cortex by identifying specific brain resting-state networks for presurgical planning in patients who may not be able to cooperate in a taskbased paradigm [62,63].