Clinical use and radiological yield of magnetic resonance fingerprinting in epilepsy

Abstract

Objective

Magnetic resonance fingerprinting (MRF) is a novel paradigm for magnetic resonance imaging (MRI) that efficiently generates multiparametric quantitative tissue property maps with a single acquisition. Its quantitative nature offers many advantages over conventional MRI. Although interest in MRF for epilepsy research has grown in recent years, systematic evaluation of its radiological and clinical value as a supplementary imaging modality is limited.

Methods

We reviewed the diagnostic utility of MRF imaging for 100 epilepsy patients undergoing presurgical evaluation. MRF quantitative maps (T1, T2, gray matter fraction, and white matter fraction) were generated and reviewed alongside conventional MRI. MRF findings were classified as (1) consistent—MRF findings the same as conventional MRI; (2) additional—MRF revealed findings beyond those seen on conventional MRI; or (3) discrepant—findings on conventional MRI not reproduced on MRF. We further stratified results by radiological diagnosis or pathology if available. Observations were further classified based on whether T1 or T2 MRF maps provided greater diagnostic utility.

Results

MRF revealed additional findings in a total of 46 patients (46/100, 46%). In one patient, findings on conventional MRI were not recapitulated on MRF (1/100, 1%). In the remaining 53 patients, MRF findings were consistent with conventional MRI (53/100, 53%). Additional benefits of MRF included the following: improved lesion conspicuity (n = 16), improved visualization of lesion extent (n = 11) or asymmetry between bilateral mesial temporal regions (n = 9), differentiation between multiple lesions (n = 5), new lesion identified (n = 2), and visualization of connection between lesion and other areas (n = 1). Added value from MRF was particularly common in patients with focal cortical dysplasia (FCD) type IIa (p = .009 vs. FCD IIb) and periventricular nodular heterotopia.

Significance

Our findings demonstrate the additional yield from MRF in the identification and definition of the extent of epileptogenic lesions, underscoring its potential as a valuable adjunct in presurgical epilepsy evaluation.

Key points.

• MRF is a novel paradigm of MRI acquisition that yields multiple quantitative maps (e.g., T1, T2) from a single fast acquisition.

• MRF provided additional value in 46 of 100 epilepsy patients who underwent presurgical evaluation.

• Benefits included greater lesion conspicuity, extent visualization, and asymmetry between bilateral mesial temporal regions.

• Added value from MRF was particularly common in patients with FCD IIa and PVNH.

1. INTRODUCTION

Focal resective surgery provides the most effective treatment in patients whose epilepsy is resistant to pharmacotherapy. An important predictor of outcome after epilepsy surgery is the identification of a lesion on neuroimaging. ¹ Magnetic resonance imaging (MRI) is the fundamental diagnostic tool to identify epileptogenic lesions. However, visual analysis of conventional MRI is largely limited to qualitative review of T1‐ and T2‐weighted and fluid‐attenuated inversion recovery (FLAIR) images for morphometric and signal abnormalities. To augment radiological evaluation, we propose that a reproducible quantitative MRI method could contribute to improved noninvasive imaging of epileptic lesions.

Magnetic resonance fingerprinting (MRF) is a novel MRI paradigm that yields multiparametric quantitative tissue property maps in a single, fast acquisition. ² In MRF, pseudorandom acquisition parameters for flip angles and repetition times are used to generate signal evolutions (i.e. “fingerprints”) that reflect intrinsic tissue properties. After signal acquisition, these evolution patterns are then matched to a dictionary of known patterns unique to various tissue types (Figure 1). Multiple tissue property maps can then be generated simultaneously, including T1 and T2 relaxometry, proton density, as well as gray matter (GM) and white matter (WM) tissue fraction maps. ² , ³ Importantly, these maps are inherently coregistered to each other, because they are reconstructed from the same acquisition. In contrast to the qualitative nature and inherent variability of conventional MRI from scan to scan, the MRF technique yields quantitative information that is reproducible and directly comparable across timepoints, scanners, patient setup, and imaging sites. ³ , ⁴ , ⁵ , ⁶ This makes MRF an ideal framework for quantitative neuroimaging studies of epilepsy.

FIGURE 1.

Overview of magnetic resonance fingerprinting framework, adapted from Bipin Mehta et al. ⁷ with permission.

There has been a growing literature in clinical application of MRF across multiple fields, including cardiology, ⁸ multiple sclerosis, ⁹ neuro‐oncology ¹⁰ and normal pressure hydrocephalus. ¹¹ In epilepsy, MRF has been used as a tool for studying epileptogenicity of periventricular nodular heterotopias (PVNH), ¹² investigating group‐level differences in focal epilepsy patients, ¹³ , ¹⁴ and as an input feature for machine learning models for epilepsy laterality prediction, ¹³ lesion detection, ¹⁵ or classification of focal cortical dysplasia (FCD). ¹⁶

Although these studies highlight the promise of MRF, there is a lack of robust systematic reports on the radiological utility of MRF as a supplementary imaging modality. Here, we describe the added value of MRF compared to conventional 3‐Tesla (3 T) MRI in a well‐characterized cohort of 100 presurgical epilepsy patients, incorporating detailed clinical profiles, pathological subgroups, and surgical outcomes.

2. MATERIALS AND METHODS

2.1. Participants

This was a retrospective single‐center study of patients recruited from the Cleveland Clinic Epilepsy Center between July 2017 and October 2024. The study was approved by the Cleveland Clinic institutional review board (CCF‐IRB, 13‐058). Patients with medically intractable epilepsy were included if they underwent a presurgical evaluation and consented to a research MRF scan. Patients were excluded if they (1) had metal implants incompatible with MRI, (2) were unable to lay still in the scanner, or (3) were unable to provide consent.

2.2. MRI and MRF acquisition

Whole brain three‐dimensional (3D) MRF scans were acquired using a Siemens 3 T Prisma scanner and 20‐channel head coil, with field of view of 300 × 300 × 144 mm, ⁴ resolution of 1 ×1 ×1 mm, ⁴ and scan time of 10 min 24 s. ¹⁴ To correct for B1 inhomogeneity, a B1 mapping sequence with the same resolution as the MRF scan was acquired. Subsequently, T1 and T2 maps as well as GM and WM tissue fraction maps were reconstructed by matching the acquired data to a pregenerated dictionary containing signal evolution patterns from a wide range of all physiologically possible T1 and T2 parameters using the Bloch equations. ³

2.3. Neuroimaging review

MRF maps (T1, T2, GM, and WM) in DICOM format were uploaded into a secure institutional picture archiving and communication system (PACS) using Agfa IMPAX 6 and Agfa EI (Agfa HealthCare) with embedded Agfa IMPAX DICOM 3D coregistration tool. MRF maps (T1, T2, GM, WM) were visualized in conjunction with conventional 3‐T MRI using grayscale and color scales to facilitate discrimination between normal and lesional brain regions. MRF review was performed retrospectively in comparison to conventional 3‐T MRI by joint consensus, with a board‐certified neuroradiologist with 18 years of experience in epilepsy imaging (S.E.J.) and an experienced neuroimaging specialist (Z.I.W.) reviewing images together. Reviewers were not blinded to results of conventional MRI. Findings were categorized based on their additional value relative to conventional MRI as (1) consistent—MRF findings the same as conventional MRI, (2) additional—MRF revealed findings beyond those seen on conventional MRI, or (3) discrepant—findings on conventional MRI not reproduced on MRF. For cases with additional findings, the type of added value was further classified as one or more of the following: improved lesion conspicuity, improved delineation of lesion extent, identification of mesial temporal asymmetry, differentiation of lesion intensity in multilesional pathologies (e.g., tuberous sclerosis complex [TSC] or PVNH), visualization of a lesion's connection to adjacent cortex, or identification of a previously unrecognized lesion. When applicable, observations were further classified based on whether T1 or T2 MRF maps provided greater diagnostic utility. Findings on GM and WM maps were also noted when helpful for the diagnostic review.

2.4. Surgical outcomes and pathology

Seizure outcome at last follow‐up was classified according to International League Against Epilepsy (ILAE) outcome scale (median duration of follow‐up = 16.8 months, range = 1–72). Favorable seizure outcome was defined as either completely seizure‐free (ILAE 1) or only auras (ILAE 2). When available, surgical specimens were microscopically rereviewed by an experienced neuropathologist (I.B.). FCD subtypes were classified using the most recent ILAE classification guidelines. ¹⁷ , ¹⁸

2.5. Statistical analyses

Statistical analyses were performed using R software (version 4.4.1, R Foundation for Statistical Computing). Fisher exact test was used to compare percentages where appropriate.

3. RESULTS

3.1. Clinical characteristics of overall cohort

A total of 100 patients were included in the study, with MRF performed between 2018 and 2024. Table 1 summarizes the clinical and demographic characteristics of the overall cohort stratified by lesional status on conventional MRI. The cohort had a median age of 25 years (range = 6–64), was predominantly right‐handed (80%), and had median epilepsy duration of 11 years (range = 1–61). Sixty‐eight patients (68%) had lesions identified on conventional MRI, most commonly FCD (n = 33), PVNH (n = 12), and mesial temporal sclerosis (MTS; n = 6). The remaining studies (n = 32) were nonlesional on initial radiology report. Overall, 43 had invasive monitoring and 65 underwent surgery (56 resection, nine laser ablation), 49 of whom had a favorable outcome (75.4%). Histopathology was available for 53 patients who had a resection, with the most common findings being FCD IIb (n = 12), FCD IIa (n = 9), mild malformation of cortical development (mMCD; n = 9), and nonspecific gliosis (n = 11).

TABLE 1.

Patient demographic and clinical characteristics.

Characteristic	Nonlesional, n = 32	Lesional, n = 68	Overall, N = 100
Age, years	25.0 [14.0, 56.0]	25.0 [6.0, 64.0]	25.0 [6.0, 64.0]
Handedness
Left	5 (15.6%)	12 (17.6%)	17 (17%)
Right	26 (81.3%)	54 (79.4%)	80 (80%)
Ambidextrous	1 (3.1%)	2 (2.9%)	3 (3%)
Sex
Male	16 (50%)	35 (51.5%)	51 (51%)
Female	16 (50%)	33 (48.5%)	49 (49%)
Seizure onset, years	9.5 [3.0, 49.0]	14.0 [.1, 44.0]	12.8 [.1, 49.0]
Duration of epilepsy, years	10.8 [1.0, 39.0]	11.0 [1.0, 61.0]	11.0 [1.0, 61.0]
Conventional MRI lesion
Encephalocele		3 (4.4%)	3 (3%)
PMG		5 (7.4%)	5 (5%)
‐		‐	‐
FCD		33 (48.5%)	33 (33%)
Gliosis		5 (7.4%)	5 (5%)
PVNH		12 (17.6%)	12 (12%)
MTS		6 (8.8%)	6 (6%)
Other		3 (4.4%)	3 (3%)
TSC		1 (1.5%)	1 (1%)
Underwent invasive monitoring	17 (53.1%)	26 (38.2%)	43 (43%)
Underwent surgery	15 (46.9%)	50 (73.5%)	65 (65%)
Favorable outcome a	10 (75%)	39 (78%)	49 (72%)
Surgical follow‐up, months	21.4 [.6, 70.2]	15.5 [.4, 72.6]	16.8 [.4, 72.6]
Histopathology
Nonspecific/gliosis	2 (6.3%)	9 (13.2%)	11 (11%)
TSC	0 (0%)	1 (1.5%)	1 (1%)
mMCD	5 (15.6%)	4 (5.9%)	9 (9%)
FCD IIb	3 (9.4%)	9 (13.2%)	12 (12%)
FCD IIa	3 (9.4%)	6 (8.8%)	9 (9%)
Ganglioma	0 (0%)	1 (1.5%)	1 (1%)
HS, type 1	0 (0%)	2 (2.9%)	2 (2%)
HS, type 2	0 (0%)	2 (2.9%)	2 (2%)
MOGHE	0 (0%)	4 (5.9%)	4 (4%)
Nodular heterotopia	0 (0%)	2 (2.9%)	2 (2%)

Note: Data are presented as median [minimum, maximum] for continuous variables and n (%) for categorical variables. MRI classification is based on conventional structural imaging. "Other" conventional MRI lesions include one patient each with subcortical band heteroropia, hypothalamic lesion, and amygdalar enlargement.

Abbreviations: FCD, focal cortical dysplasia; HS, hippocampal sclerosis; mMCD, mild MCD; MOGHE, mild malformation of cortical development with oligodendroglial hyperplasia in epilepsy; MRI, magnetic resonance imaging; MTS, mesial temporal sclerosis; PMG, polymicrogyria; PVNH, periventricular nodular heterotopia; TSC, tuberous sclerosis complex.

^a Seizure outcome and surgical follow‐up duration calculated only for patients who underwent epilepsy surgery (i.e., n = 15 for nonlesional, n = 50 for lesional, n = 65 overall).

3.2. Clinical yield for overall cohort

MRF revealed additional findings in a total of 46 patients (46/100, 46%). In one patient with FCD IIa, an ambiguous finding in the left inferior frontal gyrus was not recapitulated on MRF (1/100, 1%); this finding was positive on MRI postprocessing using the Morphometric Analysis Program (MAP) ¹⁹ , ²⁰ and confirmed on conventional MRI at the patient management conference. In the remaining 53 patients, MRF findings were consistent with conventional MRI (53/100, 53%). The additional value offered by MRF included the following: improved lesion conspicuity (n = 16), improved visualization of lesion extent (n = 11) or asymmetry between mesial temporal structures (n = 9), differentiation between multiple lesions (n = 5), new lesion identified (n = 2), and visualization of connection between lesion and other areas (n = 1). There were two patients with ≥2 reasons; one was a patient with PVNH whose MRF maps helped to differentiate between nodules and visualize nodule connection to adjacent cortex. In another patient, also with PVNH, MRF increased the nodule's conspicuity and revealed its connection to the mesial occipital cortex. Figure 2 shows detailed catogories of the additonal findings detected on MRF, stratified by radiological diagnosis on conventional MRI.

FIGURE 2.

Additional value on magnetic resonance fingerprinting (MRF) grouped by radiological diagnosis on conventional magnetic resonance imaging and type of improvement. Each cell represents the number of patients for whom MRF provided a specific type of added value. Two patients had multiple reasons for added value. In one patient with periventricular nodular heterotopias (PVNH), MRF maps helped differentiate lesions and visualize their connection to adjacent cortex. In another, also with PVNH, MRF increased lesion conspicuity and visualized the nodule's connection to adjacent cortex. Cases in which MRF did not provide added value are not included. FCD, focal cortical dysplasia; MTS, mesial temporal sclerosis; PMG, polymicrogyria; TSC, tuberous sclerosis complex.

3.3. MRF yield by lesional status

Among 100 patients, 68 had lesional initial MRI reports and 32 had nonlesional reports (Figure 3). Among the 32 patients whose initial MRI reports were negative, MRF added value in 12 (38%). Rereview of conventional MRI during patient management conference or guided by MAP identified abnormalities in 20 of 32 (62%), whereas 12 (32%) remained truly nonlesional. MRF added value in 11 of the 20 MAP‐positive cases (55%) and identified a new lesion in one of 12 truly nonlesional cases (8%). Among the 68 patients with initially lesional MRI, MRF revealed additional findings in 34 (50%).

FIGURE 3.

Magnetic resonance fingerprinting (MRF) findings for patients whose initial magnetic resonance imaging (MRI) report was negative. Some patients became classified as lesional after rereview during the patient management conference or guided by Morphometric Analysis Program (MAP). The yield of MRF is shown for each category. *In the overall cohort, two patients had de novo abnormalities detected on MRF: one patient from the truly nonlesional group (shown), and one patient from the MRI report lesional group who had an additional distinct lesion identified on MRF (not depicted in flowchart).

Overall, MRF detected two de novo abnormalities in locations not identified or suspected on the original MRI; in one patient with right amygdalar enlargement who became seizure‐free after temporal lobectomy, an additional subtle lesion of transmantle sign was identified near the superior temporal sulcus not previously seen on conventional MRI; in another patient, subtle gray–white junction abnormality was identified in the right superior frontal gyrus with decreased WM and increased GM fractions. This patient has not yet had epilepsy surgery.

3.4. MRF yield varies by pathology type

The diagnostic yield of MRF varied substantially by the underlying pathology. In patients with surgical histopathology available (n = 53; Figure 4A), MRF was particularly helpful in FCD IIa, where additional findings were detected in seven of nine cases (78%), significantly higher than in FCD IIb, where only two of 12 cases (17%) showed additional MRF value (Fisher exact test, p = .009). MRF also added value in all cases of nodular heterotopia with available histopathology (2/2, 100%) and was common in mMCD (5/9, 56%).

FIGURE 4.

Distribution of magnetic resonance fingerprinting (MRF) findings stratified by the underlying pathology. (A) Analysis by surgical histopathology (when available, n = 53) shows MRF's added value is most pronounced in focal cortical dysplasia (FCD) IIa compared to FCD IIb (p = .009). (B) Analysis by conventional magnetic resonance imaging (MRI) diagnosis (for patients without surgical pathology available, n = 47) shows particularly high yield in periventricular nodular heterotopias (PVNH), mesial temporal sclerosis (MTS), and polymicrogyria (PMG). Color coding indicates whether MRF provided additional findings (green), consistent findings (gray), or discrepant findings (blue). “Other” category includes one patient with bilateral band heterotopia and another with nonspecific hypothalamic lesion. HS, hippocampal sclerosis; mMCD, mild malformation of cortical development; MOGHE, mild malformation of cortical development with oligodendroglial hyperplasia in epilepsy; TSC, tuberous sclerosis complex.

For patients without surgical pathology (n = 47), we analyzed MRF's yield by initial radiological diagnosis (Figure 4B). Additional findings were particularly observed in patients with PVNH (7/8, 88%), MTS (3/3, 100%), and polymicrogyria (2/4, 50%). MRF was especially useful in differentiating between lesions in patients with PVNH and one patient with TSC, and delineating lesion extent in patients with suspected FCD.

3.5. Representative clinical examples

Five representative cases illustrate the spectrum of MRF's added diagnostic value (Figure 5). Patient 1 demonstrates MRF's ability to differentiate between lesions; the T1 map reveals heterogeneous signals across PVNH lesions, with the nodule showing most pronounced T1 elevation (upper left, arrow) correlating with seizure onset zone localized by stereoelectroencephalography (SEEG). Patient 2 shows a new lesion identified by MRF in the left frontal region, despite a nonlesional clinical MRI. Patient 3 exemplifies increased lesion conspicuity, the most common category of MRF's added value in our cohort; the left orbitofrontal abnormality was not identified on conventional MRI and equivocal on MAP, but more clearly visible on MRF; SEEG findings and seizure freedom after resection confirmed this as the epileptogenic focus, with pathology revealing mMCD. Patient 4 highlights enhanced lesion extent revealed by MRF in a patient with mild malformation of cortical development and oligodendroglial hyperplasia in epilepsy (MOGHE). Finally, Patient 5 demonstrates mesial temporal asymmetry better seen on MRF, where reduced field inhomogeneity of MRF facilitates detection of subtle mesial temporal asymmetry in a patient with nonlesional clinical MRI and seizure freedom after SEEG‐guided surgery.

FIGURE 5.

Examples of additional value from magnetic resonance fingerprinting (MRF). All images are oriented in radiological convention (left side of figure corresponds to patient's right). Patient 1 (A–D): Differentiation between lesions in 42‐year‐old right‐handed female with periventricular nodular heterotopias and reduction in seizures after laser ablation (International League Against Epilepsy class 3 outcome). (A) Axial T1‐weighted image demonstrates bilateral periventricular nodules. (B) Corresponding MRF T1 map reveals elevated T1 values with heterogenous intensity across nodules. (C) Zoomed‐in view of the T1 map highlights lesion with most pronounced T1 signal elevation, corresponding to ictal onset zone on stereoelectroencephalography (SEEG). (D) Ictal onset location during SEEG evaluation (red); all implanted electrodes are shown in blue; only in‐plane electrodes are shown. Patient 2 (E–H): Identification of new abnormality in the left frontal region on MRF in a 24‐year‐old right‐handed male. Clinical magnetic resonance imaging (MRI) was nonlesional, as demonstrated on axial fluid‐attenuated inversion recovery (FLAIR; E); noninvasive data, including semiology, electroencephalography, single photon emission computed tomography, and positron emission tomography, localized epilepsy to the left frontal region (F). MRF T1 map shows increased T1 signal in the same region (G), with a zoomed‐in view shown in (H). The patient elected not to pursue invasive monitoring or epilepsy surgery at this point. Patient 3 (I–L): Increased lesion conspicuity visualized on MRF in a 34‐year‐old right‐handed female with initially nonlesional clinical MRI (I); subsequent rereview prompted by the Morphometric Analysis Program (MAP) shows subtle gray–white matter blurring in the right frontal region. SEEG evaluation localized the seizure onset zone to this region, the resection of which made the patient seizure‐free. Axial FLAIR image (J) shows subtle signal hyperintensity in the right lateral frontal region, identified only on MAP‐guided rereview. Field inhomogeneity, a common pitfall in FLAIR, made the finding less convincing. MRF T1 (K) and T2 (L) maps show increased lesion conspicuity, with both maps revealing a denser, more compact, thickened cortex and reduced white matter, in the orbitofrontal region. Patient 4 (M–P): Enhanced lesion extent with MRF in a 27‐year‐old female with mild malformation of cortical development with oligodendroglial hyperplasia and epilepsy (MOGHE), seizure‐free after resection. Conventional clinical MRI was initially read as negative, and a subtle lesion in the right posterolateral frontal region was suspected after MAP (M, N: axial T1‐weighted and FLAIR). MRF maps reveal more extensive abnormality on the T1 map with loss of expected arborization of white matter (WM) into gyri, appearing as green color due to increased T1 signal (O); this is recapitulated on the MRF WM fraction map (P). Patient 5 (Q–T): Reduced field inhomogeneity on MRF facilitates the assessment of asymmetry in mesial temporal structures in a 30‐year‐old female with left mesial temporal epilepsy, seizure‐free since surgery (>1 year follow up). Clinical MRI was reported as nonlesional on multiple studies (Q: coronal T2‐weighted, R: coronal FLAIR). Surgery was guided by SEEG, which showed a left mesial temporal onset (S). MRF T1 map (T) shows increased signal in the left mesial temporal region.

3.6. MRF T1 map tends to be more useful than T2

A distinction between the yield of MRF T1 versus T2 maps was noted in 41 patients, with the MRF T1 map proving more useful in 32 cases and T2 map more useful in the remaining nine patients. This was particularly evident in patients with FCD and PVNH, where 34% and 28% of patients demonstrated greater utility from the MRF T1 map, respectively (Table 2).

TABLE 2.

Yield of magnetic resonance fingerprinting T1 versus T2 maps.

Radiological diagnosis per clinical MRI	T1 yield, n = 32	T2 yield, n = 9
PMG	3 (9%)	0 (0%)
FCD	11 (34%)	3 (33%)
Nonlesional	7 (21%)	3 (33%)
Gliosis	0 (0%)	2 (22%)
PVNH	9 (28%)	0 (0%)
MTS	1 (3%)	0 (0%)
Other	1 (3%)	0 (0%)
TSC	0 (0%)	1 (11%)

Note: Data are presented as n (%).

Abbreviations: FCD, focal cortical dysplasia; MRI, magnetic resonance imaging; MTS, mesial temporal sclerosis; PMG, polymicrogyria; PVNH, periventricular nodular heterotopia.

4. DISCUSSION

We report the largest retrospective cohort study (N = 100) evaluating the radiological utility of MRF as a supplement to conventional clinical MRI in the context of patients with medically intractable focal epilepsy undergoing presurgical evaluation. Overall, we found that MRF provided additional value in nearly half of cases studied, particularly for enhanced conspicuity and visualization of lesion extent. In the following sections, we discuss the implications of these findings particularly for surgical management and clinical integration.

4.1. Reduced field inhomogeneity and assessment of asymmetries

Tiny nonuniformities in the magnetic field, also known as inhomogeneity, can cause subtle shading artifacts across image slices/volume, limiting the ability to globally compare regions across the image based on signal intensity. This is a well‐known problem and is often appreciated in epilepsy imaging when viewing FLAIR images, seen as a hemispheric/regional brightness on certain parts of the image. This hinders the comparison of the hippocampi based on increased hyperintensity (i.e., is the hippocampus hyperintense due to a true pathology such as MTS or due to inhomogeneity?). Of note, imaging at higher field strength (e.g., 7 T) does not alleviate this problem, as inhomogeneity artifacts tend to be even more pronounced in areas close to air–tissue boundaries such as the anterior/inferior temporal lobes. ²¹ As illustrated in Figure 5, MRF proved especially useful in mitigating this limitation. Within the 3D MRF protocol used for the current study, a whole brain 3D B1 mapping sequence was acquired with the same field of view and resolution as the MRF scan. This allows for B1 inhomogeneity correction, resulting in “flatter” MRF maps with more uniform tissue contrast. This enables more confident assessment of mesial temporal asymmetry, with minimized confounding from field‐related artifacts. Consequently, regional asymmetries can be more confidently assessed, which is a significant advantage when evaluating TLE patients, for whom lateralizing pathology carries important implications for surgical management.

4.2. Enhanced lesion conspicuity and extent

In the majority of cases with additional findings, MRF provided improved visualization of lesion conspicuity and extent compared to conventional MRI. This added value is likely to stem from the quantitative nature of MRF. As MRF is more directly related to underlying tissue properties and less susceptible to artifacts and variability across scanner settings, sensitivity for imaging subtle tissue property changes can be enhanced. Additionally, we reviewed the MRF images using a colorized scheme, which allowed for more precise windowing over a dynamic range, compared to the grayscale images typical of conventional MRI. This approach can have some drawbacks, as it may create visually busy images, particularly for readers unfamiliar with the appearance of the images. However, after an initial learning period, the colorized scheme effectively enabled finer tuning for detection of subtle features, such as blurring into WM and subtle connections to overlying cortex. In addition to T1 and T2 maps, MRF allows for simultaneous generation of tissue fraction maps such as GM and WM maps that are inherently aligned, eliminating the need for post hoc processing. This allows direct comparison of the same voxel or region between different maps. We found this particularly useful for highlighting brain regions with abnormal gyration or sulcation (e.g., Figure 5, Patient 3 and 4), particularly with windowing of the colorized maps to enhance subtle regional variations.

4.3. Biophysical basis of MRF's yield

Tissue contrast observed on conventional MRI is derived from a complex, nonlinear mixture of signals (e.g., T1 and T2 relaxation times and proton density). In practice, most clinical images are acquired with parameters that emphasize one of these properties (e.g., T1‐weighted or T2‐weighted images). However, the weighted signals are influenced by several external variables such as acquisition parameters, receiver coil sensitivity, and other hardware factors. ²² As a result, the images produced do not reflect purely intrinsic tissue properties, limiting the user's ability to directly infer tissue‐level pathophysiology. In contrast, MRF produces quantitative maps of individual tissue properties, specifically T1 and T2 relaxation times, based on the matching of signal evolution patterns to a predefined dictionary. ² These maps are independent of scanner‐specific settings and are more directly tied to the underlying cellular architecture, myelin and water content, and other microenvironmental properties, thus providing a more physiologically meaningful representation of brain tissue.

We observed that MRF T1 maps were more likely to add diagnostic value than T2 maps, consistent with prior conventional MRI literature. T1 relaxation time represents the time it takes for proton's magnetic moment to realign with the external field. This rate of recovery is influenced by the surrounding environment, including lipids, proteins, and other macromolecules. ²³ T1 is particularly influenced by the concentration and distribution of myelin, one of the most abundant magnetic components in brain tissue. ²⁴ Hypomyelination, which is common in many epilepsy‐associated pathologies, can lengthen T1 relaxation times. ²⁵ Studies that compare MRI metrics with histopathological findings have shown positive correlations between prolonged T1 values and neuronal markers such as MAP2, and negative correlations with myelin stains, reinforcing the idea that MRI can serve as a proxy for certain microscopic tissue features. ²⁶ Several clinical studies have shown that T1 values tend to increase in a variety of epilepsy syndromes. For instance, in temporal lobe epilepsy, increased T1 has been observed in temporopolar, parahippocampal, and orbitofrontal regions by magnetization‐prepared two rapid acquisition gradient echoes‐based quantitative T1 mapping ²⁷ and in the hippocampi of a subset of patients with otherwise normal‐appearing structures using MRF. ²⁸ Additionally, patients with hippocampal sclerosis show increased MRF T1 in adjacent white matter regions, such as the temporal stem and temporopolar cortex. ²⁹ For FCD, elevated T1 has been identified within the lesional area, with both MRF and variable flip angle spoiled gradient echo‐based T1 mapping method. ³ , ¹⁶ , ³⁰ For PVNH, MRF T1 values have been effective in differentiating seizure‐onset nodules from noninvolved nodules at both the individual and group level. ¹² Collectively, these findings indicate that increased T1 is associated with epileptic abnormalities across diverse pathologies, including cases with both overt lesions and subtle changes not apparent on conventional MRI (nonlesional cases).

T2 relaxation time reflects spin–spin relaxation, that is, the dephasing of proton spins due to interactions between water molecules that reduce the overall magnetization within a voxel. MRF T2 maps were noted to be noisier and less consistently useful as compared to T1, in our studies and in prior literature. ¹² , ¹⁴ , ¹⁶ In the context of PVNH, for example, MRF T2 values did not consistently differ between seizure‐onset and nononset nodules. ¹² One reason may be that T2 is more susceptible to imaging artifacts, such as field inhomogeneity, and higher measurement variabilities compared to T1. ⁵ , ⁶ , ³¹ This highlights the necessity for future technical development work to improve T2 map quality within the MRF framework, some of which is already underway. ³²

4.4. Pathology and MRF findings

We found the major added value of MRF is in a subgroup of patients with FCD and PVNH. Of interest, we observed that MRF was more likely to add value in cases of FCD IIa compared to FCD IIb (78% vs. 17%, p = .009). This finding has important clinical implications, as FCD IIa is substantially more difficult to detect on conventional MRI compared to IIb. ¹⁷ , ³³ In this context, MRF appears to address a diagnostic gap, enhancing detection and delineation of the more subtle IIa lesions. Similarly, in mMCD, MRF was useful in lesion conspicuity and delineating lesion extent. For PVNH, which often presents with multiple nodules, MRF's added value lay in the ability to differentiate among them. This is consistent with a prior study focusing on the quantitative analysis of MRF maps for PVNH, which showed nodules at seizure onsets had higher T1 values compared to noninvolved nodules. ¹² The same study also showed that higher MRF T1 and T2 values in PVNH are associated with increased power spectra at ictal onset on SEEG, ¹² supporting the link between tissue properties and electrophysiological activity.

4.5. De novo findings and enhanced diagnostic confidence

In the current study, MRF identified previously unrecognized findings in two patients. Due to the retrospective nature of this study, these de novo findings could not be directly validated by invasive studies. It remains unclear whether these observations represent true epileptogenic lesions or incidental findings. Although such de novo findings were rare, MRF more frequently improved visualization or increased confidence in lesions already suspected on conventional MRI (e.g., those identified by MAP). These findings highlight MRF's potential as an adjunctive tool to refine lesion detection, either by corroborating ambiguous findings or by helping to rule out false positive ones. Greater diagnostic confidence is essential in the presurgical workup, where quantitative imaging plays an increasingly pivotal role in patient selection and surgical planning.

4.6. Clinical adoption of MRF: Promises and challenges

Based on our experience interpreting MRF in this cohort, we offer preliminary observations that may inform future systematic studies and workflow development. In our hands, interpretation was facilitated by tailored colorscales to enhance tissue contrast, and coregistration with conventional MRI sequences. Reviewing MRF maps within the PACS environment allowed use of familiar radiology tools like synchronized scrolling, region of interest measurements, and windowing, although the optimal workflow for clinical integration remains to be formally established. MRF has seen increasing interest from the clinical community and is beginning to enter the clinical market, with US Food and Drug Administration 510(k) clearance granted to an initial implementation in 2023. Several challenges remain to widespread clinical adoption including the need for specialized expertise, large dictionary sizes, and long reconstruction times and initial learning curve for image reviewers. Ongoing efforts aimed at overcoming these technical barriers, along with large‐scale studies across centers validating its clinical utility, will be important in facilitating the clinical adoption of MRF.

Although MRF offers promising capabilities, unique challenges also present that limit its standalone diagnostic utility. In our experience, MRF T2 maps contain lower signal‐to‐noise ratios than T1 maps, resulting in coarser and noisier images that complicate interpretation. The inherent hyperintensity of cerebrospinal fluid on MRF maps can further obscure nearby pathology by “swamping” the reader's visual attention. Unlike conventional MRI, the current MRF paradigms lack FLAIR‐like sequences to suppress fluid‐related signal, which is a crucial area for future development. Furthermore, MRF may be suboptimal for assessment of lesions near the brain–bone interface such as encephaloceles, due to higher signal variability in skull signal. These limitations suggest the continued need for simultaneous coreview of conventional MRI and motivate future advancements in MRF acquisition, reconstruction, and postprocessing frameworks.

4.7. Limitations

This study has several limitations. First, although MRF data were acquired prospectively with a rather broad inclusion criteria, our cohort was limited to patients who consented to participate, who may differ systematically from those who decline (e.g., in disease severity, motivation, or access to care); eligibility criteria for the MRF research protocol may also have excluded certain patient populations (e.g., those with metal implants or who cannot stay still in the scanner). Therefore, our study cohort may not be fully representative of the broader epilepsy population. Second, interpretation of MRF scans was not blinded to the conventional MRI findings or clinical information, which may have introduced confirmation bias. Additionally, MRF images were reviewed concurrently by both readers rather than independently, precluding assessment of interrater agreement. Although the review was intentionally designed to make the review feasible across a large number of cases for information relevant to the epilepsy cases, this prior knowledge may have influenced the interpretation of MRF images, potentially biasing the identification of subtle abnormalities. A prospective, blinded study with independent reviews by multiple radiologists is necessary to rigorously assess the independent diagnostic value of MRF in detecting epileptogenic abnormalities, particularly in cases where conventional MRI is nonlesional, and to establish interrater reliability.

5. CONCLUSIONS

The present study lays the groundwork for the clinical adoption of MRF by demonstrating its radiologic utility in the largest‐to‐date cohort of 100 epilepsy patients undergoing presurgical evaluation. Notably, the highest yield of additional findings in our study was observed in patients with PVNH and FCD type IIa, both being challenging epilepsy pathologies where improved imaging guidance is critical. The demonstrated improvements in lesion conspicuity or delineation of lesion extent may significantly enhance planning for invasive monitoring and epilepsy surgery. Although these findings are encouraging, MRF's greatest advantage lies in its exceptional sensitivity, repeatability, and reproducibility in quantifying tissue‐level changes. To fully harness this capability, these quantitative markers must be rigorously correlated with histopathological findings and electrophysiology, preferably with a multicenter study design. Establishing these relationships will enhance our ability to identify, quantify, and understand epileptogenic lesions, and ultimately, more effectively treat the epilepsies associated with them.

AUTHOR CONTRIBUTIONS

Dan Ma, Stephen E. Jones, and Zhong Irene Wang contributed to the conception and design of the study. Maksim Parfyonov, Ting‐Yu Su, Joon Yul Choi, Ingmar Blümcke, Ken Sakaie, Andreas Alexopoulos, Hiroatsu Murakami, Spencer Morris, Imad Najm, Dan Ma, Stephen E. Jones, and Zhong Irene Wang contributed to the acquisition and analysis of data and editing the manuscript. Maksim Parfyonov, Stephen E. Jones, and Zhong Irene Wang contributed to drafting a significant portion of the manuscript or figures.

CONFLICT OF INTEREST STATEMENT

I.N. is on the Speakers' bureau of SK Life. D.M. has MRF patents licensed by Siemens. S.E.J. has received speaker support from Siemens Healthineers and previously conducted Siemens‐funded research and consulting unrelated to the current study. The other authors have no competing interests to disclose. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.