MR Fingerprinting Shows Potential to Reshape FCD Detection and Epileptogenicity Mapping

For clinicians managing medically refractory focal epilepsy, a leading challenge frequently lies in what have been termed “nonlesional” brain MRI scans. Subtle focal cortical dysplasia (FCD) often eludes visual assessment on standard MRI, resulting in forgone or deferred epilepsy surgeries or unsuccessful surgical outcomes.

Two recent Cleveland Clinic studies demonstrate that magnetic resonance fingerprinting (MRF) — a rapid, quantitative imaging technique — can significantly improve the ability to detect the elusive lesions of FCD and determine which ones are actively driving a patient’s seizures.

The studies collectively signal progress in defining optimal clinical applicability of MRF when traditional MRI comes up short in epilepsy imaging. “We have shown that, by integrating MRF with machine learning and surface-based analysis, clinicians can now achieve high detection sensitivities for subtle FCD while simultaneously reducing the noise of false positives,” says the studies’ senior author, Irene Wang, PhD, Research Director, Cleveland Clinic Epilepsy Center. “We also found that in complex cases involving multiple malformations, MRF offers a noninvasive method for prioritizing which lesions require invasive electroencephalography (SEEG) exploration.”

Study 1: Automated detection and subtyping via machine learning

The first study (Epilepsia. 2025 Epub 9 Oct), led by Dr. Wang and Ting-Yu Su, PhD, a postdoctoral fellow in her lab, focused on developing a robust framework for whole-brain FCD detection by combining MRF with machine learning and surface-based morphometry, a widely adopted MRI postprocessing technique.

“Manual identification of FCD is highly variable and depends on specialized clinical expertise,” Dr. Su explains. “Subtle imaging features are easily missed. We sought to create an automated pipeline that leverages the high-resolution, quantitative data of MRF to improve upon existing surface-based detection methods.”

Study design and results

The team retrospectively analyzed 44 patients with confirmed FCD who underwent an MRF research scan at Cleveland Clinic. They also recruited 70 age- and gender-matched healthy individuals to undergo MRF scanning to serve as a control group.

All MRF imaging was performed using a high-resolution 3T MRF sequence (approximately 10 minutes scan time) to generate T1 and T2 maps. These maps were integrated with structural T1-weighted and 3D FLAIR images collected clinically. Quantitative features were generated based on all these data.

For FCD detection the researchers employed a two-stage machine learning approach using the following:

• A vertexwise neural network classifier to identify potentially abnormal cortical vertices
• A clusterwise Random Undersampling Boosting classifier designed specifically to prevent false positives by analyzing cluster size and feature statistics

Key findings included the following:

• Enhanced sensitivity. Combining T1-weighted images with MRF and FLAIR resulted in 71.4% sensitivity for FCD detection. In comparison, clinical MRI detected only 57% of lesions based on official radiology reports.
• Superior control of false positives. The addition of MRF data significantly reduced false-positive clusters in patients and controls, in some cases by more than 50%.
• Accurate subtyping. The framework distinguished between type II FCD and non-type II malformations with 80.8% accuracy, and it outperformed traditional radiological markers such as the transmantle sign in classifying FCD IIb subtypes.
• Interpretability. True-positive clusters showed consistently higher detection probabilities than false-positive clusters across all feature sets. This separation suggests that probability values may serve as a practical confidence measure, the researchers note, strengthening the case for clinical adoption.
• Outcome correlations. The framework showed higher sensitivity for FCD detection in patients who became seizure-free after surgery (72.7%) compared with patients who did not (50.0%), suggesting that its output may serve as an indicator of seizure outcome.

Image content: This image is available to view online.

View image online (https://assets.clevelandclinic.org/transform/d6259df8-1610-474f-b0bb-c789d6b2d346/brain-scan-images-dr-wang-inset-1)

Figure 1. Example images from patients in whom the MRF machine learning models identified clusters (yellow) that had excellent overlap with the manual lesion labels (red) in cases where the clinical MRI was positive (left) and negative (right). LEFT: Patient with FCD IIb in the right mesial occipital lobe. RIGHT: Patient with mild malformation of cortical development (mMCD) in the left basal temporal region.

Implications

“This research is the first to integrate surface-based morphometry with MRF-generated quantitative tissue property maps,” Dr. Wang notes. “It establishes that MRF-driven machine learning frameworks can match or exceed expert human performance in lesion detection. For clinicians, the framework’s prediction probabilities may potentially serve as a confidence index in distinguishing true lesions from anatomical variants or imaging artifacts.”

Study 2: Differentiating epileptogenic from silent malformations

The second study (J Neurol Sci. 2025;477:12351) addressed a different clinical dilemma: In a patient with multiple cortical malformations, which one is the source of the seizures?

Multiple FCD-like abnormalities or widespread polymicrogyria (PMG) are present in a single patient, but these lesions are not always equally epileptogenic. Determining which lesion(s) to target often requires invasive stereoelectroencephalography (SEEG). This pilot study investigated whether MRF could noninvasively indicate the active seizure onset zone in this setting.

Study design and results

Dr. Wang and colleagues retrospectively analyzed 69 individuals who underwent a 3D whole-brain MRF research scan at Cleveland Clinic. This included 21 patients with refractory focal epilepsy and 48 healthy controls included for comparative analysis. Of the patients, four had complex cortical malformations (FCD or PMG) for which they underwent SEEG and/or surgery and 17 had histopathologically verified FCD II.

Among the four complex cases, the researchers compared MRF signatures within the same patient (comparing active vs. silent lesions) and across patients (comparing lesions with those in the 17 patients with FCD II).

Key results included the following:

• Gray matter T1 as a signature for seizures. In all four complex cases, the epileptogenic malformations showed significantly higher gray matter T1 values relative to nonepileptogenic regions. This finding held true within individual patients and across patients.
• Ability to unmask an MRI-negative focus. In one of the complex cases, conventional MRI identified a lesion that was ultimately proven nonepileptogenic. Conversely, MRF detected elevated T1 and T2 values in a different, MRI-negative region that was later confirmed by SEEG as the true seizure onset zone.

Image content: This image is available to view online.

View image online (https://assets.clevelandclinic.org/transform/f1f7e0f6-09fb-4a38-8bba-bbf90cac9600/brain-scan-images-dr-wang-inset-2)

Figure 2. Example patients in whom the MRF T1 shows significant differences in the seizure-causing lesion compared with the electrically “silent” lesion. Knowing this information before surgery could help design better implantation/surgical plans. Reprinted from Kochi et al., J Neurol Sci. 2025;477:12351, ©2025 The Authors, under the CC BY-NC license.

Implications

Dr. Wang and co-authors conclude that quantitative gray matter T1 as measured by MRF could serve as a sensitive marker for epileptogenicity. “MRF shows promise as a noninvasive imaging probe for in vivo epileptogenicity,” Dr. Wang says. “What we need to understand more is which quantitative metrics consistently and strongly correlate with epileptogenic tissue. If our pilot findings are confirmed in larger studies, MRF will have a role in guiding SEEG implantation and surgical planning in complex cases, including those with multiple cortical malformations.”

Takeaways about MRF in FCD

According to the Cleveland Clinic Epilepsy Center researchers, these two studies collectively offer several insights for the management of intractable epilepsy:

• Value beyond visual detection. Clinicians should view MRF not just as a tool for finding lesions but as a way to characterize tissue pathology. Its ability to provide absolute values makes it possible to compare a patient’s brain against normative pathology libraries, aiding in the diagnosis of MRI-negative cases.
• A role in optimizing SEEG implantation. When multiple lesions are present, elevated T1 gray matter values on MRF can help guide electrode placement. A lesion with normal MRF values, despite appearing abnormal on structural scans, may be less likely to be the primary seizure source.
• Potential to reduce false positives in automated detection. One of the hurdles of AI-based lesion detection is the high rate of false positives, which can be a burden for clinical review. These studies show that the quantitative nature of MRF can be invaluable for filtering out artifacts that look like FCD but which lack the underlying tissue property changes associated with the disease.
• Prognostic value. The correlation between model detection and seizure-free outcomes suggests that MRF-based measures may offer valuable noninvasive biomarkers for outcome indication.

“As MRF moves toward broader multicenter validation, the fact that it can now be acquired in just a few minutes makes it increasingly practical for routine use,” Dr. Wang concludes. “It has the potential to become an important part of presurgical evaluation and surgical planning for patients with refractory focal epilepsy.”