Identifying juvenile myoclonic epilepsy via diffusion tensor imaging using machine learning analysis

Identifying responders to vagus nerve stimulation based on microstructural features of thalamocortical tracts in drug-resistant epilepsy

The mechanisms of action of Vagus Nerve Stimulation (VNS) and the biological prerequisites to respond to the treatment are currently under investigation. It is hypothesized that thalamocortical tracts play a central role in the antiseizure effects of VNS by disrupting the genesis of pathological activity in the brain. This pilot study explored whether in vivo microstructural features of thalamocortical tracts may differentiate Drug-Resistant Epilepsy (DRE) patients responding and not responding to VNS treatment. Eighteen patients with DRE (37.11 ​± ​10.13 years, 10 females), including 11 responders or partial responders and 7 non-responders to VNS, were recruited for this high-gradient multi-shell diffusion Magnetic Resonance Imaging (MRI) study. Using Diffusion Tensor Imaging (DTI) and multi-compartment models - Neurite Orientation Dispersion and Density Imaging (NODDI) and Microstructure Fingerprinting (MF), we extracted microstructural features in 12 subsegments of thalamocortical tracts. These characteristics were compared between responders/partial responders and non-responders. Subsequently, a Support Vector Machine (SVM) classifier was built, incorporating microstructural features and 12 clinical covariates (including age, sex, duration of VNS therapy, number of antiseizure medications, benzodiazepine intake, epilepsy duration, epilepsy onset age, epilepsy type - focal or generalized, presence of an epileptic syndrome - no syndrome or Lennox-Gastaut syndrome, etiology of epilepsy - structural, genetic, viral, or unknown, history of brain surgery, and presence of a brain lesion detected on structural MRI images). Multiple diffusion metrics consistently demonstrated significantly higher white matter fiber integrity in patients with a better response to VNS (pFDR < 0.05) in different subsegments of thalamocortical tracts. The SVM model achieved a classification accuracy of 94.12%. The inclusion of clinical covariates did not improve the classification performance. The results suggest that the structural integrity of thalamocortical tracts may be linked to therapeutic effectiveness of VNS. This study reveals the great potential of diffusion MRI in improving our understanding of the biological factors associated with the response to VNS therapy.

MRI-Based Radiomics Approach for Differentiating Juvenile Myoclonic Epilepsy from Epilepsy with Generalized Tonic-Clonic Seizures Alone

BACKGROUND

The clinical presentation of juvenile myoclonic epilepsy (JME) and epilepsy with generalized tonic-clonic seizures alone (GTCA) is similar, and MRI scans are often perceptually normal in both conditions making them challenging to differentiate.

PURPOSE

To develop and validate an MRI-based radiomics model to accurately diagnose JME and GTCA, as well as to classify prognostic groups.

STUDY TYPE

Retrospective.

POPULATION

164 patients (127 with JME and 37 with GTCA) patients (age 24.0 ± 9.6; 50% male), divided into training (n = 114) and test (n = 50) sets in a 7:3 ratio with the same proportion of JME and GTCA patients kept in both sets.

FIELD STRENGTH/SEQUENCE

3T; 3D T1-weighted spoiled gradient-echo.

ASSESSMENT

A total of 17 region-of-interest in the brain were identified as having clinical evidence of association with JME and GTCA, from where 1581 radiomics features were extracted for each subject. Forty-eight machine-learning combinations of oversampling, feature selection, and classification algorithms were explored to develop an optimal radiomics model. The performance of the best radiomics models for diagnosis and for classification of the favorable outcome group were evaluated in the test set.

STATISTICAL TESTS

Model performance measured using area under the curve (AUC) of receiver operating characteristic (ROC) curve. Shapley additive explanations (SHAP) analysis to estimate the contribution of each radiomics feature.

RESULTS

The AUC (95% confidence interval) of the best radiomics models for diagnosis and for classification of favorable outcome group were 0.767 (0.591-0.943) and 0.717 (0.563-0.871), respectively. SHAP analysis revealed that the first-order and textural features of the caudate, cerebral white matter, thalamus proper, and putamen had the highest importance in the best radiomics model.

CONCLUSION

The proposed MRI-based radiomics model demonstrated the potential to diagnose JME and GTCA, as well as to classify prognostic groups. MRI regions associated with JME, such as the basal ganglia, thalamus, and cerebral white matter, appeared to be important for constructing radiomics models.

LEVEL OF EVIDENCE

3 TECHNICAL EFFICACY: Stage 3.

Brain connectivity in status epilepticus as a predictor of outcome: A diffusion tensor imaging study

BACKGROUND AND PURPOSE

We aimed to explore structural connectivity in status epilepticus.

METHODS

We enrolled participants who underwent diffusion tensor imaging. We applied graph theory to investigate structural connectivity. We compared the structural connectivity measures between patients and healthy controls and between patients with poor (modified Rankin Scale [mRS] >3) and good (mRS ≤3) admission outcomes.

RESULTS

We enrolled 28 patients and 31 healthy controls (age 65.5 vs.62.0 years, p = .438). Of these patients, 16 and 12 showed poor and good admission outcome (age 65.5 vs.62.0 years, p = .438). The assortative coefficient (-0.113 vs. -0.121, p = .021), mean clustering coefficient (0.007 vs.0.006, p = .009), global efficiency (0.023 vs.0.020, p = .009), transitivity (0.007 vs.0.006, p = .009), and small-worldness index (0.006 vs.0.005, p = .021) were higher in patients with status epilepticus than in healthy controls. The assortative coefficient (-0.108 vs. -0.119, p = .042), mean clustering coefficient (0.007 vs.0.006, p = .042), and transitivity (0.008 vs.0.007, p = .042) were higher in patients with poor admission outcome than in those with good admission outcome. MRS score was positively correlated with structural connectivity measures, including the assortative coefficient (r = 0.615, p = .003), mean clustering coefficient (r = 0.544, p = .005), global efficiency (r = 0.515, p = .007), transitivity (r = 0.547, p = .007), and small-worldness index (r = 0.435, p = .024).

CONCLUSION

We revealed alterations in structural connectivity, showing increased integration and segregation in status epilepticus, which might be related with neuronal synchronization. This effect was more pronounced in patients with a poor admission outcome, potentially reshaping our understanding for comprehension of status epilepticus mechanisms and the development of more targeted treatments.

Neurobehavioral and Clinical Comorbidities in Epilepsy: The Role of White Matter Network Disruption

Epilepsy is a common neurological disorder associated with alterations in cortical and subcortical brain networks. Despite a historical focus on gray matter regions involved in seizure generation and propagation, the role of white matter (WM) network disruption in epilepsy and its comorbidities has sparked recent attention. In this review, we describe patterns of WM alterations observed in focal and generalized epilepsy syndromes and highlight studies linking WM disruption to cognitive and psychiatric comorbidities, drug resistance, and poor surgical outcomes. Both tract-based and connectome-based approaches implicate the importance of extratemporal and temporo-limbic WM disconnection across a range of comorbidities, and an evolving literature reveals the utility of WM patterns for predicting outcomes following epilepsy surgery. We encourage new research employing advanced analytic techniques (e.g., machine learning) that will further shape our understanding of epilepsy as a network disorder and guide individualized treatment decisions. We also address the need for research that examines how neuromodulation and other treatments (e.g., laser ablation) affect WM networks, as well as research that leverages larger and more diverse samples, longitudinal designs, and improved magnetic resonance imaging acquisitions. These steps will be critical to ensuring generalizability of current research and determining the extent to which neuroplasticity within WM networks can influence patient outcomes.

Machine Learning and Artificial Intelligence Applications to Epilepsy: a Review for the Practicing Epileptologist

PURPOSE OF REVIEW

Machine Learning (ML) and Artificial Intelligence (AI) are data-driven techniques to translate raw data into applicable and interpretable insights that can assist in clinical decision making. Some of these tools have extremely promising initial results, earning both great excitement and creating hype. This non-technical article reviews recent developments in ML/AI in epilepsy to assist the current practicing epileptologist in understanding both the benefits and limitations of integrating ML/AI tools into their clinical practice.

RECENT FINDINGS

ML/AI tools have been developed to assist clinicians in almost every clinical decision including (1) predicting future epilepsy in people at risk, (2) detecting and monitoring for seizures, (3) differentiating epilepsy from mimics, (4) using data to improve neuroanatomic localization and lateralization, and (5) tracking and predicting response to medical and surgical treatments. We also discuss practical, ethical, and equity considerations in the development and application of ML/AI tools including chatbots based on Large Language Models (e.g., ChatGPT). ML/AI tools will change how clinical medicine is practiced, but, with rare exceptions, the transferability to other centers, effectiveness, and safety of these approaches have not yet been established rigorously. In the future, ML/AI will not replace epileptologists, but epileptologists with ML/AI will replace epileptologists without ML/AI.

Brain networks and epilepsy development in patients with Alzheimer disease

INTRODUCTION

This study aimed to investigate the association between brain networks and epilepsy development in patients with Alzheimer disease (AD).

METHODS

We enrolled patients newly diagnosed with AD at our hospital who underwent three-dimensional T1-weighted magnetic resonance imaging at the time of AD diagnosis and included healthy controls. We obtained the cortical, subcortical, and thalamic nuclei structural volumes using FreeSurfer and applied graph theory to obtain the global brain network and intrinsic thalamic network based on the structural volumes using BRAPH.

RESULTS

We enrolled 25 and 56 patients with AD with and without epilepsy development, respectively. We also included 45 healthy controls. The global brain network differed between the patients with AD and healthy controls. The local efficiency (2.026 vs. 3.185, p = .048) and mean clustering coefficient (0.449 vs. 1.321, p = .024) were lower, whereas the characteristic path length (0.449 vs. 1.321, p = .048) was higher in patients with AD than in healthy controls. Both global and intrinsic thalamic networks were significantly different between AD patients with and without epilepsy development. In the global brain network, local efficiency (1.340 vs. 2.401, p = .045), mean clustering coefficient (0.314 vs. 0.491, p = .045), average degree (27.442 vs. 41.173, p = .045), and assortative coefficient (-0.041 vs. -0.011, p = .045) were lower, whereas the characteristic path length (2.930 vs. 2.118, p = .045) was higher in patients with AD with epilepsy development than in those without. In the intrinsic thalamic network, the mean clustering coefficient (0.646 vs. 0.460, p = .048) was higher, whereas the characteristic path length (1.645 vs. 2.232, p = .048) was lower in patients with AD with epilepsy development than in those without.

CONCLUSION

We found that the global brain network differs between patients with AD and healthy controls. In addition, we demonstrated significant associations between brain networks (both global brain and intrinsic thalamic networks) and epilepsy development in patients with AD.

Structural brain network analysis in occipital lobe epilepsy

BACKGROUND

This study aimed to analyze the structural brain network in patients with occipital lobe epilepsy (OLE) and investigate the differences in structural brain networks between patients with OLE and healthy controls.

METHODS

Patients with OLE and healthy controls with normal brain MRI findings were enrolled. They underwent diffusion tensor imaging using a 3.0T MRI scanner, and we computed the network measures of global and local structural networks in patients with OLE and healthy controls using the DSI studio program. We compared network measures between the groups.

RESULTS

We enrolled 23 patients with OLE and 42 healthy controls. There were significant differences in the global structural network between patients with OLE and healthy controls. The assortativity coefficient (-0.0864 vs. -0.0814, p = 0.0214), mean clustering coefficient (0.0061 vs. 0.0064, p = 0.0203), global efficiency (0.0315 vs. 0.0353, p = 0.0086), and small-worldness index (0.0001 vs. 0.0001, p = 0.0175) were lower, whereas the characteristic path length (59.2724 vs. 53.4684, p = 0.0120) was higher in patients with OLE than those in the healthy controls. There were several nodes beyond the occipital lobe that showed significant differences in the local structural network between the groups. In addition, the assortativity coefficient was negatively correlated with the duration of epilepsy (r=-0.676, p = 0.001).

Glymphatic system dysfunction in patients with occipital lobe epilepsy

BACKGROUND AND PURPOSE

We aimed to investigate the glymphatic system function in patients with occipital lobe epilepsy (OLE) and healthy controls using diffusion tensor image analysis along the perivascular space (DTI-ALPS) index.

METHODS

We retrospectively included 23 patients with OLE and 30 healthy controls. The participants underwent brain MRI, which was normal, and diffusion tensor imaging. We used the DSI Studio for data preprocessing, obtained the fiber orientation and diffusivities, and calculated the DTI-ALPS index from the diffusivity values associated with the projection and association fibers in the left hemisphere.

RESULTS

There were no differences in mean age (31.6 years [range: 13-58] vs. 31.3 years [range: 20-57], p = .912) and male sex ratio (10/23 [43.5%] vs. 15/30 [50.0%]) between the groups. Compared to healthy controls, the diffusivities in patients with OLE were higher along the Y-axis in the projection fiber and along the Z-axis in the association fiber and lower along the Y-axis in the association fiber. The DTI-ALPS index in patients with OLE was lower than that in the healthy controls (1.421 ± 0.171 vs. 1.667 ± 0.271, p < .001, 95% confidence interval of difference = 0.117-0.376, Test statistic t = 3.823). We found no association between the DTI-ALPS index and clinical characteristics in OLE.

CONCLUSION

The DTI-ALPS index in patients with OLE was significantly lower than that in healthy controls, suggesting glymphatic system dysfunction in OLE. The DTI-ALPS index could help assess the glymphatic system function in patients with epilepsy.

Can Artificial Intelligence Diagnose Transient Global Amnesia Using Electroencephalography Data?

BACKGROUND AND PURPOSE

This study aimed to determine the ability of deep learning using convolutional neural networks (CNNs) to diagnose transient global amnesia (TGA) based on electroencephalography (EEG) data, and to differentiate between patients with recurrent TGA events and those with a single TGA event.

METHODS

We retrospectively enrolled newly diagnosed patients with TGA and healthy controls. All patients with TGA and the healthy controls underwent EEG. The EEG signals were converted into images using time-frequency analysis with short-time Fourier transforms. We employed two CNN models (AlexNet and VGG19) to classify the patients with TGA and the healthy controls, and for further classification of patients with recurrent TGA events and those with a single TGA event.

RESULTS

We enrolled 171 patients with TGA and 68 healthy controls. The accuracy and area under the curve (AUC) of the AlexNet and VGG19 models in classifying patients with TGA and healthy controls were 70.4% and 71.8%, and 0.718 and 0.743, respectively. In addition, the accuracy and AUC of the AlexNet and VGG19 models in classifying patients with recurrent TGA events and those with a single TGA event were 71.1% and 88.4%, and 0.773 and 0.873, respectively.

CONCLUSIONS

We have successfully demonstrated the feasibility of deep learning in diagnosing TGA based on EEG data, and used two different CNN models to distinguish between patients with recurrent TGA events and those with a single TGA event.

Artificial Intelligence Applications in the Imaging of Epilepsy and Its Comorbidities: Present and Future

Artificial intelligence (AI) is increasingly used in medical image analysis and has accelerated scientific discoveries across fields of medicine. In this review, we highlight how AI has been applied to neuroimaging in patients with epilepsy to enhance classification of clinical diagnosis, prediction of treatment outcomes, and the understanding of cognitive comorbidities. We outline the strengths and shortcomings of current AI research and the need for future studies using large datasets that test the reproducibility and generalizability of current findings, as well as studies that test the clinical utility of AI approaches.