Search

OBJECTIVE: Localization of focal epilepsy is critical for surgical treatment of refractory seizures. There remains a great need for noninvasive techniques to localize seizures for surgical decision-making. We investigate the use of deep learning using resting state functional magnetic resonance imaging (RS-fMRI) to identify the hemisphere of seizure onset in temporal lobe epilepsy (TLE) patients. METHODS: A total of 2132 healthy controls and 32 preoperative TLE patients were studied. All participants underwent structural MRI and RS-fMRI. Healthy control data were used to generate training samples for a three-dimensional convolutional neural network (3DCNN). RS-fMRI was synthetically altered in randomly lateralized regions in the healthy control participants. The model was then trained to classify the hemisphere containing synthetic noise. Finally, the model was tested on TLE patients to assess its performance for detecting biological seizure onset zones, and gradient-weighted class activation mapping (Grad-CAM) identified the strongest predictive regions. RESULTS: The 3DCNN classified healthy control hemispheres known to contain synthetic noise with 96% accuracy, and TLE hemispheres clinically identified to be seizure onset zones with 90.6% accuracy. Grad-CAM identified a range of temporal, frontal, parietal, and subcortical regions that were strong anatomical predictors of the seizure onset zone, and the resting state networks that colocalized with Grad-CAM results included default mode, medial temporal, and dorsal attention networks. Lastly, in an analysis of a subset of patients with postsurgical outcomes, the 3DCNN leveraged a more focal set of regions to achieve classification in patients with Engel Class >I compared to Engel Class I. SIGNIFICANCE: Noninvasive techniques capable of localizing the seizure onset zone could improve presurgical planning in patients with intractable epilepsy. We have demonstrated the ability of deep learning to identify the correct hemisphere of the seizure onset zone in TLE patients using RS-fMRI with high accuracy. This approach represents a novel technique of seizure lateralization that could improve preoperative surgical planning.

SummaryObjectives Specific changes in the functional connectivity of brain networks occur in patients with epilepsy. Yet whether such changes reflect a stable disease effect or one that is a function of active seizure burden remains unclear. Here, we longitudinally assessed the connectivity of canonical cognitive functional networks in patients with intractable temporal lobe epilepsy (TLE), both before and after patients underwent epilepsy surgery and achieved seizure freedom. Methods Seventeen patients with intractable TLE who underwent epilepsy surgery with Engel class I outcome and 17 matched healthy controls took part in the study. The functional connectivity of a set of cognitive functional networks derived from typical cognitive tasks was assessed in patients, preoperatively and postoperatively, as well as in controls, using stringent methods of artifact reduction. Results Preoperatively, functional networks in TLE patients differed significantly from healthy controls, with differences that largely, but not exclusively, involved the default mode and temporal/auditory subnetworks. However, undergoing epilepsy surgery and achieving seizure freedom did not lead to significant changes in network connectivity, with postoperative functional network abnormalities closely mirroring the preoperative state. Significance This result argues for a stable chronic effect of the disease on brain connectivity, with changes that are largely “burned in” by the time a patient with intractable TLE undergoes epilepsy surgery, which typically occurs years after the initial diagnosis. The result has potential implications for the treatment of intractable epilepsy, suggesting that delaying surgical intervention that may achieve seizure freedom may lead to functional network changes that are no longer reversible by the time of epilepsy surgery.

BACKGROUND AND PURPOSE To demonstrate the accuracy across different acquisition and analysis methods, we evaluated the variability in hippocampal volumetric and surface displacement measurements resulting from two different MRI (magnetic resonance imaging) acquisition protocols. METHODS Nine epilepsy patients underwent two independent T1-weighted magnetization prepared spoiled gradient sequences during a single 3T MRI session. Using high-dimension mapping-large deformation (HDM-LD) segmentation, we calculated volumetric estimates and generated a vector-based 3-dimensional surface model of each subject's hippocampi, and evaluated volume and surface changes, the latter using a cluster-based noise estimation model. RESULTS Mean hippocampal volumes and standard deviations for the left hippocampi were 2,750 (826) mm3 and 2,782 (859) mm3 (P = .13), and for the right hippocampi were 2,558 (750) mm3 and 2,547 (692) mm3 (P = .76), respectively for the MPR1 and MPR2 sequences. Average Dice coefficient comparing overlap for segmentations was 86%. There was no significant effect of MRI sequence on volume estimates and no significant hippocampal surface change between sequences. CONCLUSION Statistical comparison of hippocampal volumes and statistically thresholded HDM-LD surfaces in TLE patients showed no differences between the segmentations obtained in the two MRI acquisition sequences. This validates the robustness across MRI sequences of the HDM-LD technique for estimating volume and surface changes in subjects with epilepsy.

Summary Objective Hippocampal atrophy in temporal lobe epilepsy (TLE) can indicate mesial temporal sclerosis and predict surgical success. Yet many patients with TLE do not have significant atrophy (magnetic resonance imaging (MRI) negative), which presents a diagnostic challenge. We used a new variant of high-dimensional large-deformation mapping to assess whether patients with apparently normal hippocampi have local shape changes that mirror those of patients with significant hippocampal atrophy. Methods Forty-seven patients with unilateral TLE and 32 controls underwent structural brain MRI. High-dimensional large-deformation mapping provided hippocampal surface and volume estimates for each participant, dividing patients into low versus high hippocampal atrophy groups. A vertex-level generalized linear model compared local shape changes between groups. Results Patients with low-atrophy TLE (MRI negative) had significant local hippocampal shape changes compared to controls, similar to those in the contralateral hippocampus of high-atrophy patients. These changes primarily involved the subicular and hilar/dentate regions, instead of the classically affected CA1 region. Disease duration instead co-varied with lateral hippocampal atrophy, co-localizing with the CA1 subfield. Significance These findings show that patients with “MRI-negative” TLE have regions of hippocampal atrophy that cluster medially, sparing the lateral regions (CA1) involved in high-atrophy patients. This suggests an overall effect of temporal lobe seizures manifesting as bilateral medial hippocampal atrophy, and a more selective effect of hippocampal seizures leading to disease-proportional CA1 atrophy, potentially reflecting epileptogenesis.

Many cognitive paradigms require self-paced responses or examine events that occur at unpredictable times. To explore whether functional MRI (fMRI) can accommodate such paradigms, a method allowing rapid, unpredictable trial pacing was developed and tested on 17 subjects using activation of the motor network as a model. Trial onset was determined solely by the subjects' self-paced responses and trials occurred, on average, less than 2 s apart. The hemodynamic response was estimated both in relation to stimulus onset (stimulus-locked) and in relation to behavioral response time (response-locked). Results yielded robust activation maps and hemodynamic response estimates. Specifically, significant activation in motor cortex, supplementary motor area (SMA), and cerebellum was observed both at the group and at the individual-subject level, confirming predicted patterns of brain activity. Moreover, the self-paced design resulted in even temporal sampling of the hemodynamic response across the image acquisition, allowing estimation of response parameters. Stimulus-locked analysis demonstrated strong correlation between hemodynamic- and behavioral-response timing both within and across subjects. Conversely, response-locked analysis showed minimal correlation with behavioral timing, suggesting effective resynchronization of the timing parameters. These results demonstrate fMRI procedures that can accommodate rapid, arbitrarily timed events and, in doing so, provide precise temporal estimates of the hemodynamic response.

Eighteen participants were imaged using fMRI to explore whether brain regions predicting successful verbal memory encoding during semantic decisions would continue to predict encoding during structural (non-semantic) decisions. Consistent with prior studies, left inferior frontal and fusiform regions were more active during semantic than structural decisions, and activity was greater for remembered than forgotten words during semantic decisions. Critically, structural decisions yielded significantly greater activity for remembered than forgotten words in these regions providing evidence that a common frontal-temporal network supports verbal memory encoding irrespective of orienting task. Further analysis revealed activity associated with successful encoding in the right precentral gyrus, suggesting other regions may also play a role in verbal encoding during non-semantic processing.

Rapid-presentation event-related functional MRI (ER-fMRI) allows neuroimaging methods based on hemodynamics to employ behavioral task paradigms typical of cognitive settings. However, the sluggishness of the hemodynamic response and its variance provide constraints on how ER-fMRI can be applied. In a series of two studies, estimates of the hemodynamic response in or near the primary visual and motor cortices were compared across various paradigms and sampling procedures to determine the limits of ER-fMRI procedures and, more generally, to describe the behavior of the hemodynamic response. The temporal profile of the hemodynamic response was estimated across overlapping events by solving a set of linear equations within the general linear model. No assumptions about the shape were made in solving the equations. Following estimation of the temporal profile, the amplitude and timing were modeled using a gamma function. Results indicated that (1) within a region, for a given subject, estimation of the hemodynamic response is extremely stable for both amplitude (r(2) = 0.98) and time to peak (r(2) = 0.95), from one series of measurements to the next, and slightly less stable for estimation of time to onset (r(2) = 0.60). (2) As the trial presentation rate changed (from those spaced 20 s apart to temporally overlapping trials), the hemodynamic response amplitude showed a small, but significant, decrease. Trial onsets spaced (on average) 5 s apart showed a 17-25% reduction in amplitude compared to those spaced 20 s apart. Power analysis indicated that the increased number of trials at fast rates outweighs this decrease in amplitude if statistically reliable response detection is the goal. (3) Knowledge of the amplitude and timing of the hemodynamic response in one region failed to predict those properties in another region, even for within-subject comparisons. (4) Across subjects, the amplitude of the response showed no significant correlation with timing of the response, for either time-to-onset or time-to-peak estimates. (5) The within-region stability of the response was sufficient to allow offsets in the timing of the response to be detected that were under a second, placing event-related fMRI methods in a position to answer questions about the change in relative timing between regions.

Summary Purpose Posttraumatic epilepsy (PTE) occurs in a proportion of traumatic brain injury (TBI) cases, significantly compounding the disability, and risk of injury and death for sufferers. To date, predictive biomarkers for PTE have not been identified. This study used the lateral fluid percussion injury (LFPI) rat model of TBI to investigate whether structural, functional, and behavioral changes post-TBI relate to the later development of PTE. Methods Adult male Wistar rats underwent LFPI or sham injury. Serial magnetic resonance (MR) and positron emission tomography (PET) imaging, and behavioral analyses were performed over 6 months postinjury. Rats were then implanted with recording electrodes and monitored for two consecutive weeks using video–electroencephalography (EEG) to assess for PTE. Of the LFPI rats, 52% (n = 12) displayed spontaneous recurring seizures and/or epileptic discharges on the video-EEG recordings. Key Findings MRI volumetric and signal analysis of changes in cortex, hippocampus, thalamus, and amygdala, 18F-fluorodeoxyglucose (FDG)–PET analysis of metabolic function, and behavioral analysis of cognitive and emotional changes, at 1 week, and 1, 3, and 6 months post-LFPI, all failed to identify significant differences on univariate analysis between the epileptic and nonepileptic groups. However, hippocampal surface shape analysis using large-deformation high-dimensional mapping identified significant changes in the ipsilateral hippocampus at 1 week postinjury relative to baseline that differed between rats that would go onto become epileptic versus those who did not. Furthermore, a multivariate logistic regression model that incorporated the 1 week, and 1 and 3 month 18F-FDG PET parameters from the ipsilateral hippocampus was able to correctly predict the epileptic outcome in all of the LFPI cases. As such, these subtle changes in the ipsilateral hippocampus at acute phases after LFPI may be related to PTE and require further examination. Significance These findings suggest that PTE may be independent of major structural, functional, and behavioral changes induced by TBI, and suggest that more subtle abnormalities are likely involved. However, there are limitations associated with studying acquired epilepsies in animal models that must be considered when interpreting these results, in particular the failure to detect differences between the groups may be related to the limitations of properly identifying/separating the epileptic and nonepileptic animals into the correct group.

Stimulus repetition associates with neural activity reductions during tasks that elicit behavioral priming. Here we present direct evidence for a quantitative relation between neural activity reductions and behavioral priming. Fifty-four subjects performed a word classification task while being scanned with functional MRI. Activity reductions were found in multiple high-level cortical regions including those within the prefrontal cortex. Importantly, activity within several of these regions, including the prefrontal cortex, correlated with behavior such that greater activity reductions associated with faster performance. Whole-brain correlational analyses confirmed the observation of anatomic overlap between regions showing activity reductions and those showing direct brain–behavioral correlations. The finding of a quantitative relation between neural and behavioral effects in frontal regions suggests that repetition reduces frontally mediated processing in a manner that ultimately facilitates behavior.

Two methods of quantify- ing hemispheric language dominance (HLD) in neurosurgical patients are compared: (1) an average magni- tudes (AM) method, which is a cal- culation of the average signal intensity variation in regions of interest for each patient that were predefined in a group analysis for each task, and (2) a lateralization indices (LI) method, which is based on the number of activated pixels in regions of interest predefined in each individual patient. Four language tasks (a living/nonliving (LNL) judgment, word stem completion (WSC), semantic associate (SA) and a phonological associate (PA) task) were compared with ''gold stan- dard'' measures such as the Wada test or electrocortical stimulation. Results showed that the LI method was more accurate (73% agreement with gold standard methods) than the AM method (only 40% agree- ment) across tasks and subjects. Furthermore, by varying the threshold used for determining lat- erality, the ability of functional magnetic resonance imaging (fMRI) to predict HLD was influenced for the AM method, whereas the LI method was relatively unaffected by changing the threshold. Using the LI method, the SA task was the most accurate for quantifying HLD (100% agreement with gold stan- dard methods) with respect to the other three language tasks (80% accuracy for WSC, 65% for the LNL and 63% for phonological task). Depending on the method and the task, fMRI may be a promising tool for assessing HLD in neuro- surgical patients.

Recent advances in functional neuroimaging allow comparisons between individuals with schizophrenia and control groups. Previous studies of schizophrenia have used blocked task paradigms and, more recently, “rapid event-related” designs in which stimuli of different types are presented close together in an intermixed fashion. The validity of between-group comparisons in both of these types of paradigms depends on excluding the possibility that observed functional response differences are attributable to altered hemodynamic responses in individuals with schizophrenia. The goal of the current study was to begin a systematic examination of the hemodynamic response in schizophrenia. We administered a flashing checkerboard paradigm with a motor response to 17 individuals with schizophrenia and 24 healthy controls. Both groups showed robust activation of visual, motor, somatosensory, and supplementary motor regions. For the most part, the individuals with schizophrenia demonstrated intact peak amplitude, variance, latency, and linear summation properties in regions activated by this task. We did find some evidence for increased variability in the amplitude and latency of the hemodynamic responses in the visual and somatosensory cortices, although the magnitudes of these group differences were relatively small. These results begin to validate the interpretation of functional neuroimaging studies of schizophrenia in terms of neuronal as opposed to vascular mechanisms.

Preclinical and clinical studies have demonstrated the significance of inflammation and autoantibodies in epilepsy, and the use of immunotherapies in certain situations has become an established practice. Temporal lobe epilepsy can follow paraneoplastic or nonparaneoplastic limbic encephalitis associated with antibodies directed against brain antigens. Here, we focus on a patient with worsening confusion and temporal lobe seizures despite treatment with antiepileptic medications. Serial brain MRIs did not conclusively reveal structural abnormalities, so the patient underwent brain PET/MRI to simultaneously evaluate brain structure and function, revealing bitemporal abnormalities. The patient was diagnosed with voltage-gated potassium channel antibody-related limbic encephalitis based on clinical presentation, imaging findings, and antibody testing. Treatment included the addition of a second antiepileptic agent and oral steroids. His seizures and cognitive deficits improved and stabilized.