One-Dimensional Convolutional Neural Networks Combined with Channel Selection Strategy for Seizure Prediction Using Long-Term Intracranial EEG

Automatic Detection of Scalp High-Frequency Oscillations Based on Deep Learning

Scalp high-frequency oscillations (sHFOs) are a promising non-invasive biomarker of epilepsy. However, the visual marking of sHFOs is a time-consuming and subjective process, existing automatic detectors based on single-dimensional analysis have difficulty with accurately eliminating artifacts and thus do not provide sufficient reliability to meet clinical needs. Therefore, we propose a high-performance sHFOs detector based on a deep learning algorithm. An initial detection module was designed to extract candidate high-frequency oscillations. Then, one-dimensional (1D) and two-dimensional (2D) deep learning models were designed, respectively. Finally, the weighted voting method is used to combine the outputs of the two model. In experiments, the precision, recall, specificity and F1-score were 83.44%, 83.60%, 96.61% and 83.42%, respectively, on average and the kappa coefficient was 80.02%. In addition, the proposed detector showed a stable performance on multi-centre datasets. Our sHFOs detector demonstrated high robustness and generalisation ability, which indicates its potential applicability as a clinical assistance tool. The proposed sHFOs detector achieves an accurate and robust method via deep learning algorithm.

Adaptive Time-Frequency Segment Optimization for Motor Imagery Classification

Motor imagery (MI)-based brain-computer interface (BCI) has emerged as a crucial method for rehabilitating stroke patients. However, the variability in the time-frequency distribution of MI-electroencephalography (EEG) among individuals limits the generalizability of algorithms that rely on non-customized time-frequency segments. In this study, we propose a novel method for optimizing time-frequency segments of MI-EEG using the sparrow search algorithm (SSA). Additionally, we apply a correlation-based channel selection (CCS) method that considers the correlation coefficient of features between each pair of EEG channels. Subsequently, we utilize a regularized common spatial pattern method to extract effective features. Finally, a support vector machine is employed for signal classification. The results on three BCI datasets confirmed that our algorithm achieved better accuracy (99.11% vs. 94.00% for BCI Competition III Dataset IIIa, 87.70% vs. 81.10% for Chinese Academy of Medical Sciences dataset, and 87.94% vs. 81.97% for BCI Competition IV Dataset 1) compared to algorithms with non-customized time-frequency segments. Our proposed algorithm enables adaptive optimization of EEG time-frequency segments, which is crucial for the development of clinically effective motor rehabilitation.

A Few-Shot Transfer Learning Approach for Motion Intention Decoding from Electroencephalographic Signals

In this study, a few-shot transfer learning approach was introduced to decode movement intention from electroencephalographic (EEG) signals, allowing to recognize new tasks with minimal adaptation. To this end, a dataset of EEG signals recorded during the preparation of complex sub-movements was created from a publicly available data collection. The dataset was divided into two parts: the source domain dataset (including 5 classes) and the support (target domain) dataset, (including 2 classes) with no overlap between the two datasets in terms of classes. The proposed methodology consists in projecting EEG signals into the space-frequency-time domain, in processing such projections (rearranged in channels × frequency frames) by means of a custom EEG-based deep neural network (denoted as EEGframeNET5), and then adapting the system to recognize new tasks through a few-shot transfer learning approach. The proposed method achieved an average accuracy of 72.45 ± 4.19% in the 5-way classification of samples from the source domain dataset, outperforming comparable studies in the literature. In the second phase of the study, a few-shot transfer learning approach was proposed to adapt the neural system and make it able to recognize new tasks in the support dataset. The results demonstrated the system's ability to adapt and recognize new tasks with an average accuracy of 80 ± 0.12% in discriminating hand opening/closing preparation and outperforming reported results in the literature. This study suggests the effectiveness of EEG in capturing information related to the motor preparation of complex movements, potentially paving the way for BCI systems based on motion planning decoding. The proposed methodology could be straightforwardly extended to advanced EEG signal processing in other scenarios, such as motor imagery or neural disorder classification.

Hybrid Network for Patient-Specific Seizure Prediction from EEG Data

Seizure prediction can improve the quality of life for patients with drug-resistant epilepsy. With the rapid development of deep learning, lots of seizure prediction methods have been proposed. However, seizure prediction based on single convolution models is limited by the inherent defects of convolution itself. Convolution pays attention to the local features while underestimates the global features. The long-term dependence of the electroencephalogram (EEG) data cannot be captured. In view of these defects, a hybrid model called STCNN based on Swin transformer (ST) and 2D convolutional neural network (2DCNN) is proposed. Time-frequency features extracted by short-term Fourier transform (STFT) are taken as the input of STCNN. ST blocks are used in STCNN to capture the global information and long-term dependencies of EEGs. Meanwhile, the 2DCNN blocks are adopted to capture the local information and short-term dependent features. The combination of the two blocks can fully exploit the seizure-related information thus improve the prediction performance. Comprehensive experiments are performed on the CHB-MIT scalp EEG dataset. The average seizure prediction sensitivity, the area under the ROC curve (AUC) and the false positive rate (FPR) are 92.94%, 95.56% and 0.073, respectively.

Epileptic Seizure Prediction Using Attention Augmented Convolutional Network

Early seizure prediction is crucial for epilepsy patients to reduce accidental injuries and improve their quality of life. Identifying pre-ictal EEG from the inter-ictal state is particularly challenging due to their nonictal nature and remarkable similarities. In this study, a novel epileptic seizure prediction method is proposed based on multi-head attention (MHA) augmented convolutional neural network (CNN) to address the issue of CNN's limit of capturing global information of input signals. First, data enhancement is performed on original EEG recordings to balance the pre-ictal and inter-ictal EEG data, and the EEG recordings are sliced into 6-second-long EEG segments. Subsequently, EEG time-frequency distribution is obtained using Stockwell transform (ST), and the attention augmented convolutional network is employed for feature extraction and classification. Finally, post-processing is utilized to reduce the false prediction rate (FPR). The CHB-MIT EEG database was used to evaluate the system. The validation results showed a segment-based sensitivity of 98.24% and an event-based sensitivity of 94.78% with a FPR of 0.05/h were yielded, respectively. The satisfying results of the proposed method demonstrate its possible potential for clinical applications.

Localization of Epileptic Brain Responses to Single-Pulse Electrical Stimulation by Developing an Adaptive Iterative Linearly Constrained Minimum Variance Beamformer

Delayed responses (DRs) to single pulse electrical stimulation (SPES) in patients with severe refractory epilepsy, from their intracranial recordings, can help to identify regions associated with epileptogenicity. Automatic DR localization is a large step in speeding up the identification of epileptogenic focus. Here, for the first time, an adaptive iterative linearly constrained minimum variance beamformer (AI-LCMV) is developed and employed to localize the DR sources from intracranial electroencephalogram (EEG) recorded using subdural electrodes. The prime objective here is to accurately localize the regions for the corresponding DRs using an adaptive localization method that exploits the morphology of DRs as the desired sources. The traditional closed-form linearly constrained minimum variance (CF-LCMV) solution is meant for tracking the sources with dominating power. Here, by incorporating the morphology of DRs, as a constraint, to an iterative linearly constrained minimum variance (LCMV) solution, the array of subdural electrodes is used to localize the low-power DRs, some not even visible in any of the electrode signals. The results from the cases included in this study also indicate more distinctive locations compared to those achievable by conventional beamformers. Most importantly, the proposed AI-LCMV is able to localize the DRs invisible over other electrodes.

A Class-Imbalance Aware and Explainable Spatio-Temporal Graph Attention Network for Neonatal Seizure Detection

Seizures are the most prevalent clinical indication of neurological disorders in neonates. In this study, a class-imbalance aware and explainable deep learning approach based on Convolutional Neural Networks (CNNs) and Graph Attention Networks (GATs) is proposed for the accurate automated detection of neonatal seizures. The proposed model integrates the temporal information of EEG signals with the spatial information on the EEG channels through the graph representation of the multi-channel EEG segments. One-dimensional CNNs are used to automatically develop a feature set that accurately represents the differences between seizure and nonseizure epochs in the time domain. By employing GAT, the attention mechanism is utilized to emphasize the critical channel pairs and information flow among brain regions. GAT coefficients were then used to empirically visualize the important regions during the seizure and nonseizure epochs, which can provide valuable insight into the location of seizures in the neonatal brain. Additionally, to tackle the severe class imbalance in the neonatal seizure dataset using under-sampling and focal loss techniques are used. Overall, the final Spatio-Temporal Graph Attention Network (ST-GAT) outperformed previous benchmarked methods with a mean AUC of 96.6% and Kappa of 0.88, demonstrating its high accuracy and potential for clinical applications.

Epileptic seizures detection and the analysis of optimal seizure prediction horizon based on frequency and phase analysis

Changes in the frequency composition of the human electroencephalogram are associated with the transitions to epileptic seizures. Cross-frequency coupling (CFC) is a measure of neural oscillations in different frequency bands and brain areas, and specifically phase-amplitude coupling (PAC), a form of CFC, can be used to characterize these dynamic transitions. In this study, we propose a method for seizure detection and prediction based on frequency domain analysis and PAC combined with machine learning. We analyzed two databases, the Siena Scalp EEG database and the CHB-MIT database, and used the frequency features and modulation index (MI) for time-dependent quantification. The extracted features were fed to a random forest classifier for classification and prediction. The seizure prediction horizon (SPH) was also analyzed based on the highest-performing band to maximize the time for intervention and treatment while ensuring the accuracy of the prediction. Under comprehensive consideration, the results demonstrate that better performance could be achieved at an interval length of 5 min with an average accuracy of 85.71% and 95.87% for the Siena Scalp EEG database and the CHB-MIT database, respectively. As for the adult database, the combination of PAC analysis and classification can be of significant help for seizure detection and prediction. It suggests that the rarely used SPH also has a major impact on seizure detection and prediction and further explorations for the application of PAC are needed.

Electrical Stimulation Induced Current Distribution in Peripheral Nerves Varies Significantly with the Extent of Nerve Damage: A Computational Study Utilizing Convolutional Neural Network and Realistic Nerve Models

Electrical stimulation of the peripheral nervous system is a promising therapeutic option for several conditions; however, its effects on tissue and the safety of the stimulation remain poorly understood. In order to devise stimulation protocols that enhance therapeutic efficacy without the risk of causing tissue damage, we constructed computational models of peripheral nerve and stimulation cuffs based on extremely high-resolution cross-sectional images of the nerves using the most recent advances in computing power and machine learning techniques. We developed nerve models using nonstimulated (healthy) and over-stimulated (damaged) rat sciatic nerves to explore how nerve damage affects the induced current density distribution. Using our in-house computational, quasi-static, platform, and the Admittance Method (AM), we estimated the induced current distribution within the nerves and compared it for healthy and damaged nerves. We also estimated the extent of localized cell damage in both healthy and damaged nerve samples. When the nerve is damaged, as demonstrated principally by the decreased nerve fiber packing, the current penetrates deeper into the over-stimulated nerve than in the healthy sample. As safety limits for electrical stimulation of peripheral nerves still refer to the Shannon criterion to distinguish between safe and unsafe stimulation, the capability this work demonstrated is an important step toward the development of safety criteria that are specific to peripheral nerve and make use of the latest advances in computational bioelectromagnetics and machine learning, such as Python-based AM and CNN-based nerve image segmentation.

Clinical neuroscience and neurotechnology: An amazing symbiosis

In the last decades, clinical neuroscience found a novel ally in neurotechnologies, devices able to record and stimulate electrical activity in the nervous system. These technologies improved the ability to diagnose and treat neural disorders. Neurotechnologies are concurrently enabling a deeper understanding of healthy and pathological dynamics of the nervous system through stimulation and recordings during brain implants. On the other hand, clinical neurosciences are not only driving neuroengineering toward the most relevant clinical issues, but are also shaping the neurotechnologies thanks to clinical advancements. For instance, understanding the etiology of a disease informs the location of a therapeutic stimulation, but also the way stimulation patterns should be designed to be more effective/naturalistic. Here, we describe cases of fruitful integration such as Deep Brain Stimulation and cortical interfaces to highlight how this symbiosis between clinical neuroscience and neurotechnology is closer to a novel integrated framework than to a simple interdisciplinary interaction.