Detecting Radiologically Occult Lesions in Focal Epilepsy

24 concepts19 papers

For patients with drug-resistant focal epilepsy, surgical resection of the epileptogenic zone (EZ)—the specific region of the brain where seizures originate—offers a powerful chance for a complete cure. However, the success of this intervention hinges entirely on the ability to precisely locate the pathology. In a substantial subset of patients, standard clinical MRI scans reveal no visible abnormalities. These cases are classified as "MRI-negative," "cryptogenic," or "non-lesional."

When human radiologists review these scans, the brain appears entirely unremarkable. This diagnostic gap creates a clinical impasse: without a clear visible target, specialists are hesitant to operate, leaving patients without access to potentially life-changing surgery.

The underlying pathology, such as Focal Cortical Dysplasia (FCD), is not actually absent; rather, it is radiologically occult. It manifests as microscopic cellular disorganization or subtle metabolic shifts rather than gross anatomical distortions. To bridge this gap, a multidisciplinary research direction has emerged. By leveraging advanced machine learning, molecular imaging, and electrophysiological network analysis, researchers are developing methods to expose these hidden signatures.

Amplifying Subtle Structural Footprints

Human vision is highly optimized for detecting sharp boundaries and distinct structural changes, such as a tumor or major tissue loss. It is poorly suited, however, for evaluating continuous mathematical gradients across a three-dimensional volume. Subtle lesions often cause minute alterations in cortical thickness or a slight blurring of the boundary between gray and white matter.

To capture these imperceptible changes, researchers use automated computational pipelines to analyze the entire brain as a continuous geometric surface. Early work by Keihaninejad et al. in Classification and Lateralization of Temporal Lobe Epilepsies with and without Hippocampal Atrophy Based on Whole-Brain Automatic MRI Segmentation demonstrated that even when a patient's scan is deemed non-lesional by expert radiologists, the underlying disease leaves a distinct, statistically detectable morphometric footprint across the wider brain network.

To systematically isolate these footprints, machine learning models can ingest multi-channel structural data to flag anomalies. For instance, Brodley et al. developed a semi-supervised framework using Hierarchical Conditional Random Fields in Decrypting "Cryptogenic" Epilepsy: Semi-supervised Hierarchical Conditional Random Fields For Detecting Cortical Lesions In MRI-Negative Patients, achieving a 75% detection rate for subtle FCD lesions by evaluating contextual relationships between neighboring cortical regions.

Standard MRI Scan ───► Human Vision ───► "MRI-Negative" (Occult)
       │
       ▼
High-Dimensional ────► Deep Learning ───► Statistical Anomalies Exposed
Feature Extraction     (2D/3D CNNs)       (e.g., Gray-White Matter Blurring)

The transition to deep learning has further amplified these subtle structural signals. Because deep neural networks evaluate complex, non-linear combinations of intensity gradients and spatial features, they can discriminate disease states without relying on an obvious focal lesion.

In MRI-based deep learning can discriminate between temporal lobe epilepsy, Alzheimer’s disease, and healthy controls, Chang et al. demonstrated that a 2D convolutional neural network (CNN) could successfully identify temporal lobe epilepsy (TLE) signatures that are invisible to the human eye.

Expanding this into the third dimension, Kaestner et al. showed in Adding the third dimension: 3D convolutional neural network diagnosis of temporal lobe epilepsy that 3D CNNs maintain high diagnostic accuracy even in strictly MRI-negative cohorts, provided they are trained on large, multi-site datasets.

When the target pathology is highly specific, such as FCD Type II, targeted segmentation architectures like the self-configuring nnU-Net deployed by Joshi et al. in A nnU-Net-based automatic segmentation of FCD type II lesions in 3D FLAIR MRI images can automate the identification of these subtle structural boundaries, standardizing what was once a highly subjective and time-consuming manual review.

Mapping Metabolic and Molecular Imbalances

When anatomy appears entirely normal, the underlying physiology often betrays the location of the epileptogenic zone. Seizure-generating tissue is characterized by chronic metabolic imbalances and localized immune responses. Functional and molecular imaging techniques exploit these cellular changes to reveal the focus.

Positron Emission Tomography (PET) using the radiotracer $^{18}$F-FDG is widely used to map glucose metabolism, as epileptic zones typically show reduced metabolic activity (hypometabolism) between seizures. However, standard PET protocols often fail to resolve subtle asymmetries.

The authors of Delayed 18F-FDG PET imaging provides better metabolic asymmetry in potential epileptogenic zone in temporal lobe epilepsy, led by Hong et al., discovered that extending the imaging window from the traditional 40 minutes to 2–3 hours post-injection significantly improves detection. This works because the delayed window allows the tracer to clear from healthy tissue while remaining trapped in the pathologically altered zone, widening the contrast gap and exposing the focus.

Standard Window (40 min):  [ Healthy Tissue: High FDG ] vs [ Epileptic Focus: Med FDG ]  --> Low Contrast
Delayed Window (2-3 hrs):  [ Healthy Tissue: Cleared  ] vs [ Epileptic Focus: Trapped ]  --> High Contrast

Beyond glucose metabolism, chronic seizures trigger localized neuroinflammation, characterized by the activation of microglia. In Clinical evaluation of neuroinflammation in child-onset focal epilepsy: a translocator protein PET study, Kagitani‐Shimono et al. utilized PET imaging to target the translocator protein (TSPO), a biomarker for activated immune cells. They successfully localized epileptogenic zones in MRI-negative children, demonstrating that inflammatory molecular markers can provide a more sensitive diagnostic signal than structural anatomy.

Similarly, vascular alterations can serve as a proxy for metabolic stress. Verma et al. demonstrated in Focal and Generalized Patterns of Cerebral Cortical Veins Due to Non-Convulsive Status Epilepticus or Prolonged Seizure Episode after Convulsive Status Epilepticus – A MRI Study Using Susceptibility Weighted Imaging that Susceptibility Weighted Imaging (SWI) can track deoxygenated blood in cortical veins, acting as a contrast-free window into the brain's local metabolic consumption during or after prolonged seizure episodes.

Rather than relying on a single modality, modern frameworks achieve higher diagnostic specificity by fusing structural and functional data. By combining automated Morphometric Analysis Programs (MAP) with quantitative PET data in a simultaneous PET/MRI workflow, the authors of Morphometric Analysis Program and Quantitative Positron Emission Tomography in Presurgical Localization in MRI-negative Epilepsies: a Simultaneous PET/MRI Study (Guo et al.) demonstrated a highly reliable method for pinpointing invisible lesions.

This multimodal strategy is echoed by Kanber et al. in Detection of covert lesions in focal epilepsy using computational analysis of multimodal magnetic resonance imaging data and Mo et al. in Clinical Value of Machine Learning in the Automated Detection of Focal Cortical Dysplasia Using Quantitative Multimodal Surface-Based Features, confirming that aggregating morphological, intensity, and metabolic features into a unified computational pipeline yields significantly higher localization accuracy than analyzing any single modality in isolation.

Uncovering Pathological Network Dynamics

Epilepsy is fundamentally a disease of abnormal electrical synchrony. Even when the brain's macro-structure appears perfectly intact, its underlying electrical grid is disrupted. This disruption is continuous, persisting even during "normal," spike-free intervals between seizures (interictal periods).

By shifting the diagnostic focus from static anatomy to dynamic network behavior, researchers can locate the source of the disease through its functional impact. Choi et al. demonstrated in Electroencephalographic Resting-State Functional Connectivity of Benign Epilepsy with Centrotemporal Spikes that focal epilepsy causes measurable drops in global network efficiency and connectivity during resting-state wakefulness, proving that the disease continuously alters the brain's functional state.

To turn these network disruptions into localized targets, mathematical metrics are applied to standard electroencephalography (EEG) data. In Frontotemporal phase lag index correlates with seizure severity in patients with temporal lobe epilepsy, the authors (led by Mao et al.) utilized the Phase Lag Index (PLI)—a metric that quantifies phase synchronization between different brain regions while discarding volume conduction artifacts—to provide an objective functional indicator of epilepsy severity in patients lacking visible structural lesions.

Scalp EEG Channels ───► Phase Lag Index (PLI) ───► Source Imaging ───► Localized Network Hub
                        & Graph Theory             Framework           (Epileptogenic Zone)

To map these scalp-level signals back to their physical origin in the brain, researchers use advanced source imaging. Huang et al. introduced a network-based EEG source imaging framework in A network-based EEG source imaging framework for noninvasive localization of epileptogenic zones in MRI-negative focal drug-resistant epilepsy, mathematically reconstructing deep brain electrical sources from scalp recordings to pinpoint the EZ non-invasively.

To train and validate these non-invasive network models, researchers rely on gold-standard invasive recordings, such as intracranial EEG (iEEG) and stereoelectroencephalography (SEEG). These invasive modalities capture the precise electrical onset patterns of a seizure directly from the brain tissue.

In Multiple ictal onset patterns underlie seizure generation in seizure-free patients with temporal lobe epilepsy surgery: an SEEG study, Xu et al. mapped distinct electrical onset signatures across normal-appearing structures in patients who achieved complete seizure freedom post-surgery, providing the baseline functional maps needed to refine surgical strategies when structural targets are missing.

Finally, these electrophysiological signals can be fed back into structural models to guide them. In Multi-task Learning with Weak Class Labels: Leveraging iEEG to Detect Cortical Lesions in Cryptogenic Epilepsy, Ahmed et al. designed a multi-task learning framework that uses functional data from iEEG as auxiliary "weak labels" to train machine learning models. By forcing the algorithm to learn structural features that correlate directly with known electrical abnormalities, the system successfully uncovers subtle, radiologically occult cortical lesions that would otherwise escape detection.

History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_mri_negative_diagnostic_gap_62da2efd,
  title = {MRI-negative diagnostic gap},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/mri-negative-diagnostic-gap/rdr_62da2efdee6a4602a46ac46252fd1adc},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_mri_negative_diagnostic_gap_62da2efd_v1,
  title = {MRI-negative diagnostic gap},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/mri-negative-diagnostic-gap/rdr_62da2efdee6a4602a46ac46252fd1adc/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}

See something wrong? Report a mistake