Re-evaluating human MTL in working memory: insights from intracranial recordings.

The study of human working memory (WM) holds significant importance in neuroscience; yet, exploring the role of the medial temporal lobe (MTL) in WM has been limited by the technological constraints of noninvasive methods. Recent advancements in human intracranial neural recordings have indicated the involvement of the MTL in WM processes. These recordings show that different regions of the MTL are involved in distinct aspects of WM processing and also dynamically interact with each other and the broader brain network. These findings support incorporating the MTL into models of the neural basis of WM. This integration can better reflect the complex neural mechanisms underlying WM and enhance our understanding of WM's flexibility, adaptability, and precision.

Deep learning based decoding of single local field potential events.

How is information processed in the cerebral cortex? In most cases, recorded brain activity is averaged over many (stimulus) repetitions, which erases the fine-structure of the neural signal. However, the brain is obviously a single-trial processor. Thus, we here demonstrate that an unsupervised machine learning approach can be used to extract meaningful information from electro-physiological recordings on a single-trial basis. We use an auto-encoder network to reduce the dimensions of single local field potential (LFP) events to create interpretable clusters of different neural activity patterns. Strikingly, certain LFP shapes correspond to latency differences in different recording channels. Hence, LFP shapes can be used to determine the direction of information flux in the cerebral cortex. Furthermore, after clustering, we decoded the cluster centroids to reverse-engineer the underlying prototypical LFP event shapes. To evaluate our approach, we applied it to both extra-cellular neural recordings in rodents, and intra-cranial EEG recordings in humans. Finally, we find that single channel LFP event shapes during spontaneous activity sample from the realm of possible stimulus evoked event shapes. A finding which so far has only been demonstrated for multi-channel population coding.

Neurophysiological mechanisms of cognition in the developing brain: Insights from intracranial EEG studies.

The quest to understand how the development of the brain supports the development of complex cognitive functions is fueled by advances in cognitive neuroscience methods. Intracranial EEG (iEEG) recorded directly from the developing human brain provides unprecedented spatial and temporal resolution for mapping the neurophysiological mechanisms supporting cognitive development. In this paper, we focus on episodic memory, the ability to remember detailed information about past experiences, which improves from childhood into adulthood. We review memory effects based on broadband spectral power and emphasize the importance of isolating narrowband oscillations from broadband activity to determine mechanisms of neural coordination within and between brain regions. We then review evidence of developmental variability in neural oscillations and present emerging evidence linking the development of neural oscillations to the development of memory. We conclude by proposing that the development of oscillations increases the precision of neural coordination and may be an essential factor underlying memory development. More broadly, we demonstrate how recording neural activity directly from the developing brain holds immense potential to advance our understanding of cognitive development.

Electrical Source Imaging of Somatosensory Evoked Potentials from Intracranial EEG Signals.

Stereoelectroencephalography (SEEG) records electrical brain activity with intracerebral electrodes. However, it has an inherently limited spatial coverage. Electrical source imaging (ESI) infers the position of the neural generators from the recorded electric potentials, and thus, could overcome this spatial undersampling problem. Here, we aimed to quantify the accuracy of SEEG ESI under clinical conditions. We measured the somatosensory evoked potential (SEP) in SEEG and in high-density EEG (HD-EEG) in 20 epilepsy surgery patients. To localize the source of the SEP, we employed standardized low resolution brain electromagnetic tomography (sLORETA) and equivalent current dipole (ECD) algorithms. Both sLORETA and ECD converged to similar solutions. Reflecting the large differences in the SEEG implantations, the localization error also varied in a wide range from 0.4 to 10 cm. The SEEG ESI localization error was linearly correlated with the distance from the putative neural source to the most activated contact. We show that it is possible to obtain reliable source reconstructions from SEEG under realistic clinical conditions, provided that the high signal fidelity recording contacts are sufficiently close to the source of the brain activity.

Anatomical registration of intracranial electrodes. Robust model-based localization and deformable smooth brain-shift compensation methods.

Precise electrode localization is important for maximizing the utility of intracranial EEG data. Electrodes are typically localized from post-implantation CT artifacts, but algorithms can fail due to low signal-to-noise ratio, unrelated artifacts, or high-density electrode arrays. Minimizing these errors usually requires time-consuming visual localization and can still result in inaccurate localizations. In addition, surgical implantation of grids and strips typically introduces non-linear brain deformations, which result in anatomical registration errors when post-implantation CT images are fused with the pre-implantation MRI images. Several projection methods are currently available, but they either fail to produce smooth solutions or do not account for brain deformations. To address these shortcomings, we propose two novel algorithms for the anatomical registration of intracranial electrodes that are almost fully automatic and provide highly accurate results. We first present GridFit, an algorithm that simultaneously localizes all contacts in grids, strips, or depth arrays by fitting flexible models to the electrodes' CT artifacts. We observed localization errors of less than one millimeter (below 8% relative to the inter-electrode distance) and robust performance under the presence of noise, unrelated artifacts, and high-density implants when we ran ~6000 simulated scenarios. Furthermore, we validated the method with real data from 20 intracranial patients. As a second registration step, we introduce CEPA, a brain-shift compensation algorithm that combines orthogonal-based projections, spring-mesh models, and spatial regularization constraints. When tested with real data from 15 patients, anatomical registration errors were smaller than those obtained for well-established alternatives. Additionally, CEPA accounted simultaneously for simple mechanical deformation principles, which is not possible with other available methods. Inter-electrode distances of projected coordinates smoothly changed across neighbor electrodes, while changes in inter-electrode distances linearly increased with projection distance. Moreover, in an additional validation procedure, we found that modeling resting-state high-frequency activity (75-145 Hz ) in five patients further supported our new algorithm. Together, GridFit and CEPA constitute a versatile set of tools for the registration of subdural grid, strip, and depth electrode coordinates that provide highly accurate results even in the most challenging implantation scenarios. The methods presented here are implemented in the iElectrodes open-source toolbox, making their use simple, accessible, and straightforward to integrate with other popular toolboxes used for analyzing electrophysiological data.

Acute effects of deep brain stimulation on brain function in obsessive-compulsive disorder.

OBJECTIVE: Deep brain stimulation (DBS) is an effective treatment for refractory obsessive-compulsive disorder (OCD) yet neural markers of optimized stimulation parameters are largely unknown. We aimed to describe (sub-)cortical electrophysiological responses to acute DBS at various voltages in OCD. METHODS: We explored how DBS doses between 3-5 V delivered to the ventral anterior limb of the internal capsule of five OCD patients affected electroencephalograms and intracranial local field potentials (LFPs). We focused on theta power/ phase-stability, given their previously established role in DBS for OCD. RESULTS: Cortical theta power and theta phase-stability did not increase significantly with DBS voltage. DBS-induced theta power peaks were seen at the previously defined individualized therapeutic voltage. Although LFP power generally increased with DBS voltages, this occurred mostly in frequency peaks that overlapped with stimulation artifacts limiting its interpretability. Though highly idiosyncratic, three subjects showed significant acute DBS effects on electroencephalogram theta power and four subjects showed significant carry-over effects (pre-vs post DBS, unstimulated) on LFP and electroencephalogram theta power. CONCLUSIONS: Our findings challenge the presence of a consistent dose-response relationship between stimulation voltage and brain activity. SIGNIFICANCE: Theta power may be investigated further as a neurophysiological marker to aid personalized DBS voltage optimization in OCD.

An application of dynamical directed connectivity of ictal intracranial EEG recordings in seizure onset zone localization.

BACKGROUND: Identification of the seizure onset zone (SOZ) is a challenging task in epilepsy surgery. Patients with epilepsy have an altered brain network, allowing connectivity-based analyses to have a great potential in SOZ identification. We investigated a dynamical directed connectivity analysis utilizing ictal intracranial electroencephalographic (iEEG) recordings and proposed an algorithm for SOZ identification based on grouping iEEG contacts. NEW METHODS: Granger Causality was used for directed connectivity analysis in this study. The intracranial contacts were grouped into visually detected contacts (VDCs), which were identified as SOZ by epileptologists, and non-resected contacts (NRCs). The intragroup and intergroup directed connectivity for VDCs and NRCs were calculated around seizure onset. We then proposed an algorithm for SOZ identification based on the cross-correlation of intragroup outflow and inflow of SOZ candidate contacts. RESULTS: Our results revealed that the intragroup connectivity of VDCs (VDC→VDC) was significantly larger than the intragroup connectivity of NRCs (NRC→NRC) and the intergroup connectivity between NRCs and VDCs (NRC→VDC) around seizure onset. We found that the proposed algorithm had 90.1 % accuracy for SOZ identification in the seizure-free patients. COMPARISON WITH EXISTING METHODS: The existing connectivity-based methods for SOZ identification often use either outflow or inflow. In this study, SOZ contacts were identified by integrating outflow and inflow based on the cross correlation between these two measures. CONCLUSIONS: The proposed group-based dynamical connectivity analysis in this study can aid our understanding of underlying seizure network and may be used to assist in identifying the SOZ contacts before epilepsy surgery.

A comparison of uni- and multi-variate methods for identifying brain networks activated by cognitive tasks using intracranial EEG.

Cognitive tasks are commonly used to identify brain networks involved in the underlying cognitive process. However, inferring the brain networks from intracranial EEG data presents several challenges related to the sparse spatial sampling of the brain and the high variability of the EEG trace due to concurrent brain processes. In this manuscript, we use a well-known facial emotion recognition task to compare three different ways of analyzing the contrasts between task conditions: permutation cluster tests, machine learning (ML) classifiers, and a searchlight implementation of multivariate pattern analysis (MVPA) for intracranial sparse data recorded from 13 patients undergoing presurgical evaluation for drug-resistant epilepsy. Using all three methods, we aim at highlighting the brain structures with significant contrast between conditions. In the absence of ground truth, we use the scientific literature to validate our results. The comparison of the three methods' results shows moderate agreement, measured by the Jaccard coefficient, between the permutation cluster tests and the machine learning [0.33 and 0.52 for the left (LH) and right (RH) hemispheres], and 0.44 and 0.37 for the LH and RH between the permutation cluster tests and MVPA. The agreement between ML and MVPA is higher: 0.65 for the LH and 0.62 for the RH. To put these results in context, we performed a brief review of the literature and we discuss how each brain structure's involvement in the facial emotion recognition task.

Decoding Intracranial EEG With Machine Learning: A Systematic Review.

Advances in intracranial electroencephalography (iEEG) and neurophysiology have enabled the study of previously inaccessible brain regions with high fidelity temporal and spatial resolution. Studies of iEEG have revealed a rich neural code subserving healthy brain function and which fails in disease states. Machine learning (ML), a form of artificial intelligence, is a modern tool that may be able to better decode complex neural signals and enhance interpretation of these data. To date, a number of publications have applied ML to iEEG, but clinician awareness of these techniques and their relevance to neurosurgery, has been limited. The present work presents a review of existing applications of ML techniques in iEEG data, discusses the relative merits and limitations of the various approaches, and examines potential avenues for clinical translation in neurosurgery. One-hundred-seven articles examining artificial intelligence applications to iEEG were identified from 3 databases. Clinical applications of ML from these articles were categorized into 4 domains: i) seizure analysis, ii) motor tasks, iii) cognitive assessment, and iv) sleep staging. The review revealed that supervised algorithms were most commonly used across studies and often leveraged publicly available timeseries datasets. We conclude with recommendations for future work and potential clinical applications.

Interpretable functional specialization emerges in deep convolutional networks trained on brain signals.

Objective.Functional specialization is fundamental to neural information processing. Here, we study whether and how functional specialization emerges in artificial deep convolutional neural networks (CNNs) during a brain-computer interfacing (BCI) task.Approach.We trained CNNs to predict hand movement speed from intracranial electroencephalography (iEEG) and delineated how units across the different CNN hidden layers learned to represent the iEEG signal.Main results.We show that distinct, functionally interpretable neural populations emerged as a result of the training process. While some units became sensitive to either iEEG amplitude or phase, others showed bimodal behavior with significant sensitivity to both features. Pruning of highly sensitive units resulted in a steep drop of decoding accuracy not observed for pruning of less sensitive units, highlighting the functional relevance of the amplitude- and phase-specialized populations.Significance.We anticipate that emergent functional specialization as uncovered here will become a key concept in research towards interpretable deep learning for neuroscience and BCI applications.

Quantification of Epileptogenic Network From Stereo EEG Recordings Using Epileptogenicity Ranking Method.

Introduction: Precise localization of the epileptogenic zone is very essential for the success of epilepsy surgery. Epileptogenicity index (EI) computationally estimates epileptogenicity of brain structures based on the temporal domain parameters and magnitude of ictal discharges. This method works well in cases of mesial temporal lobe epilepsy but it showed reduced accuracy in neocortical epilepsy. To overcome this scenario, in this study, we propose Epileptogenicity Rank (ER), a modified method of EI for quantifying epileptogenicity, that is based on spatio-temporal properties of Stereo EEG (SEEG). Methods: Energy ratio during ictal discharges, the time of involvement and Euclidean distance between brain structures were used to compute the ER. Retrospectively, we localized the EZ for 33 patients (9 for mesial-temporal lobe epilepsy and 24 for neocortical epilepsy) using post op MRI and Engel 1 surgical outcome at a mean of 40.9 months and then optimized the ER in this group. Results: Epileptic network estimation based on ER successfully differentiated brain regions involved in the seizure onset from the propagation network. ER was calculated at multiple thresholds leading to an optimum value that differentiated the seizure onset from the propagation network. We observed that ER < 7.1 could localize the EZ in neocortical epilepsy with a sensitivity of 94.6% and specificity of 98.3% and ER < 7.3 in mesial temporal lobe epilepsy with a sensitivity of 95% and specificity of 98%. In non-seizure-free patients, the EZ localization based on ER pointed to brain area beyond the cortical resections. Significance: Methods like ER can improve the accuracy of EZ localization for brain resection and increase the precision of minimally invasive surgery techniques (radio-frequency or laser ablation) by identifying the epileptic hubs where the lesion is extensive or in nonlesional cases. For inclusivity with other clinical applications, this ER method has to be studied in more patients.

Computer-Aided Intracranial EEG Signal Identification Method Based on a Multi-Branch Deep Learning Fusion Model and Clinical Validation.

Surgical intervention or the control of drug-refractory epilepsy requires accurate analysis of invasive inspection intracranial EEG (iEEG) data. A multi-branch deep learning fusion model is proposed to identify epileptogenic signals from the epileptogenic area of the brain. The classical approach extracts multi-domain signal wave features to construct a time-series feature sequence and then abstracts it through the bi-directional long short-term memory attention machine (Bi-LSTM-AM) classifier. The deep learning approach uses raw time-series signals to build a one-dimensional convolutional neural network (1D-CNN) to achieve end-to-end deep feature extraction and signal detection. These two branches are integrated to obtain deep fusion features and results. Resampling is employed to split the imbalanced epileptogenic and non-epileptogenic samples into balanced subsets for clinical validation. The model is validated over two publicly available benchmark iEEG databases to verify its effectiveness on a private, large-scale, clinical stereo EEG database. The model achieves high sensitivity (97.78%), accuracy (97.60%), and specificity (97.42%) on the Bern-Barcelona database, surpassing the performance of existing state-of-the-art techniques. It is then demonstrated on a clinical dataset with an average intra-subject accuracy of 92.53% and cross-subject accuracy of 88.03%. The results suggest that the proposed method is a valuable and extremely robust approach to help researchers and clinicians develop an automated method to identify the source of iEEG signals.

Identification of epilepsy from intracranial EEG signals by using different neural network models.

In this work, a framework is provided for identifying intracranial electroencephalography (iEEG) seizures based on discrete wavelet transform (DWT) analysis of iEEG signals using forward propagation and feedback neural networks. The performance of 5 different data sets combination classifications is studied using the probabilistic neural network (PNN), learning vector quantization neural network (LVQ) and Elman neural network (ENN). Different feature combinations serve as the input vectors of the classifiers to obtain the best outcomes. It has been found that PNN has less running time and provides better classification accuracy (CA) than ENN and LVQ classifiers for all 5 classification problems. It is worth noticing that the CA for the C-D classification task, which shows the status of pre-ictal versus post-ictal, has been greatly improved, and reached 83.13%. Hence, the epilepsy iEEG signals pattern recognition based on DWT statistical features using the PNN classifier is more suitable for forming a reliable, automatic classification system in order to assist doctors in diagnosis.

Mismatch negativity (MMN) as a tool for translational investigations into early psychosis: A review.

Mismatch negativity (MMN) reduction is one of the most robust findings among several neurophysiological and neurocognitive measures in patients with schizophrenia. MMN is a promising biomarker for schizophrenia because of the following properties: 1) its relationship with early psychosis, including clinical high-risk (CHR); 2) its relationship with the functional abilities of patients; and 3) its translatability into basic research using animal models. Specifically, the utility of the passive auditory oddball paradigm that does not require subjects to make behavioral responses enables identical physiological activities to be obtained from both experimental animals and patients. This advantage has contributed to clarifying the generating mechanism of MMN in various animal studies. We reviewed clinical reports focused on early psychosis; specifically differential effects of deviance type and relationships to clinical and functional outcome. For the utility of MMN as a tool for translational research, we next reviewed recent MMN studies in rodents and nonhuman primates (NHP) as well as studies using intracranial recordings in humans, a rare opportunity to detect neural signals in vivo in humans. Neural computations of MMN, such as adaptation, deviance detection, and predictive coding, have been recent topics for understanding MMN generating mechanisms. Finally, several significant research questions were provided for future directions. MMN research could contribute to innovative, novel, therapeutic strategies in the future by becoming a bridge between basic and clinical research.

Intracerebral EEG Artifact Identification Using Convolutional Neural Networks.

Manual and semi-automatic identification of artifacts and unwanted physiological signals in large intracerebral electroencephalographic (iEEG) recordings is time consuming and inaccurate. To date, unsupervised methods to accurately detect iEEG artifacts are not available. This study introduces a novel machine-learning approach for detection of artifacts in iEEG signals in clinically controlled conditions using convolutional neural networks (CNN) and benchmarks the method's performance against expert annotations. The method was trained and tested on data obtained from St Anne's University Hospital (Brno, Czech Republic) and validated on data from Mayo Clinic (Rochester, Minnesota, U.S.A). We show that the proposed technique can be used as a generalized model for iEEG artifact detection. Moreover, a transfer learning process might be used for retraining of the generalized version to form a data-specific model. The generalized model can be efficiently retrained for use with different EEG acquisition systems and noise environments. The generalized and specialized model F1 scores on the testing dataset were 0.81 and 0.96, respectively. The CNN model provides faster, more objective, and more reproducible iEEG artifact detection compared to manual approaches.

Kurtosis-Based Detection of Intracranial High-Frequency Oscillations for the Identification of the Seizure Onset Zone.

Pathological High-Frequency Oscillations (HFOs) have been recently proposed as potential biomarker of the seizure onset zone (SOZ) and have shown superior accuracy to interictal epileptiform discharges in delineating its anatomical boundaries. Characterization of HFOs is still in its infancy and this is reflected in the heterogeneity of analysis and reporting methods across studies and in clinical practice. The clinical approach to HFOs identification and quantification usually still relies on visual inspection of EEG data. In this study, we developed a pipeline for the detection and analysis of HFOs. This includes preliminary selection of the most informative channels exploiting statistical properties of the pre-ictal and ictal intracranial EEG (iEEG) time series based on spectral kurtosis, followed by wavelet-based characterization of the time-frequency properties of the signal. We performed a preliminary validation analyzing EEG data in the ripple frequency band (80-250 Hz) from six patients with drug-resistant epilepsy who underwent pre-surgical evaluation with stereo-EEG (SEEG) followed by surgical resection of pathologic brain areas, who had at least two-year positive post-surgical outcome. In this series, kurtosis-driven selection and wavelet-based detection of HFOs had average sensitivity of 81.94% and average specificity of 96.03% in identifying the HFO area which overlapped with the SOZ as defined by clinical presurgical workup. Furthermore, the kurtosis-based channel selection resulted in an average reduction in computational time of 66.60%.

Seizure Forecasting and the Preictal State in Canine Epilepsy.

The ability to predict seizures may enable patients with epilepsy to better manage their medications and activities, potentially reducing side effects and improving quality of life. Forecasting epileptic seizures remains a challenging problem, but machine learning methods using intracranial electroencephalographic (iEEG) measures have shown promise. A machine-learning-based pipeline was developed to process iEEG recordings and generate seizure warnings. Results support the ability to forecast seizures at rates greater than a Poisson random predictor for all feature sets and machine learning algorithms tested. In addition, subject-specific neurophysiological changes in multiple features are reported preceding lead seizures, providing evidence supporting the existence of a distinct and identifiable preictal state.

High frequency gamma activity in the left hippocampus predicts visual object naming performance.

Access to an object's name requires the retrieval of an arbitrary association between it's identity and a word-label. The hippocampus is essential in retrieving arbitrary associations, and thus could be involved in retrieving the link between an object and its name. To test this hypothesis we recorded the iEEG signal from epileptic patients, directly implanted in the hippocampus, while they performed a picture naming task. High-frequency broadband gamma (50-150 Hz) responses were computed as an index of population-level spiking activity. Our results show, for the first time, single-trial hippocampal dynamics between visual confrontation and naming. Remarkably, the latency of the hippocampal response predicts naming latency, while inefficient hippocampal activation is associated with "tip-of-the-tongue" states (a failure to retrieve the name of a recognized object) suggesting that the hippocampus is an active component of the naming network and that its dynamics are closely related to efficient word production.

Frequency domain beamforming of magnetoencephalographic beta band activity in epilepsy patients with focal cortical dysplasia.

INTRODUCTION: Spike-based magnetoencephalography (MEG) source localization is an established method in the presurgical evaluation of epilepsy patients. Focal cortical dysplasias (FCDs) are associated with focal epileptic discharges of variable morphologies in the beta frequency band in addition to single epileptic spikes. Therefore, we investigated the potential diagnostic value of MEG-based localization of spike-independent beta band (12-30Hz) activity generated by epileptogenic lesions. METHODS: Five patients with FCD IIB underwent MEG. In one patient, invasive EEG (iEEG) was recorded simultaneously with MEG. In two patients, iEEG succeeded MEG, and two patients had MEG only. MEG and iEEG were evaluated for epileptic spikes. Two minutes of iEEG data and MEG epochs with no spikes as well as MEG epochs with epileptic spikes were analyzed in the frequency domain. MEG oscillatory beta band activity was localized using Dynamic Imaging of Coherent Sources. RESULTS: Intralesional beta band activity was coherent between simultaneous MEG and iEEG recordings. Continuous 14Hz beta band power correlated with the rate of interictal epileptic discharges detected in iEEG. In cases where visual MEG evaluation revealed epileptic spikes, the sources of beta band activity localized within <2cm of the epileptogenic lesion as shown on magnetic resonance imaging. This result held even when visually marked epileptic spikes were deselected. When epileptic spikes were detectable in iEEG but not MEG, MEG beta band activity source localization failed. DISCUSSION: Source localization of beta band activity has the potential to contribute to the identification of epileptic foci in addition to source localization of visually marked epileptic spikes. Thus, this technique may assist in the localization of epileptic foci in patients with suspected FCD.