Optimization of Deep Architectures for EEG Signal Classification: An AutoML Approach Using Evolutionary Algorithms.

Electroencephalography (EEG) signal classification is a challenging task due to the low signal-to-noise ratio and the usual presence of artifacts from different sources. Different classification techniques, which are usually based on a predefined set of features extracted from the EEG band power distribution profile, have been previously proposed. However, the classification of EEG still remains a challenge, depending on the experimental conditions and the responses to be captured. In this context, the use of deep neural networks offers new opportunities to improve the classification performance without the use of a predefined set of features. Nevertheless, Deep Learning architectures include a vast number of hyperparameters on which the performance of the model relies. In this paper, we propose a method for optimizing Deep Learning models, not only the hyperparameters, but also their structure, which is able to propose solutions that consist of different architectures due to different layer combinations. The experimental results corroborate that deep architectures optimized by our method outperform the baseline approaches and result in computationally efficient models. Moreover, we demonstrate that optimized architectures improve the energy efficiency with respect to the baseline models.

EEG source-space synchrostate transitions and Markov modeling in the math-gifted brain during a long-chain reasoning task.

To reveal transition dynamics of global neuronal networks of math-gifted adolescents in handling long-chain reasoning, this study explores momentary phase-synchronized patterns, that is, electroencephalogram (EEG) synchrostates, of intracerebral sources sustained in successive 50 ms time windows during a reasoning task and non-task idle process. Through agglomerative hierarchical clustering for functional connectivity graphs and nested iterative cosine similarity tests, this study identifies seven general and one reasoning-specific prototypical functional connectivity patterns from all synchrostates. Markov modeling is performed for the time-sequential synchrostates of each trial to characterize the interstate transitions. The analysis reveals that default mode network, central executive network (CEN), dorsal attention network, cingulo-opercular network, left/right ventral frontoparietal network, and ventral visual network aperiodically recur over non-task or reasoning process, exhibiting high predictability in interactively reachable transitions. Compared to non-gifted subjects, math-gifted adolescents show higher fractional occupancy and mean duration in CEN and reasoning-triggered transient right frontotemporal network (rFTN) in the time course of the reasoning process. Statistical modeling of Markov chains reveals that there are more self-loops in CEN and rFTN of the math-gifted brain, suggesting robust state durability in temporally maintaining the topological structures. Besides, math-gifted subjects show higher probabilities in switching from the other types of synchrostates to CEN and rFTN, which represents more adaptive reconfiguration of connectivity pattern in the large-scale cortical network for focused task-related information processing, which underlies superior executive functions in controlling goal-directed persistence and high predictability of implementing imagination and creative thinking during long-chain reasoning.

Motor imagery and mental fatigue: inter-relationship and EEG based estimation.

Even though it has long been felt that psychological state influences the performance of brain-computer interfaces (BCI), formal analysis to support this hypothesis has been scant. This study investigates the inter-relationship between motor imagery (MI) and mental fatigue using EEG: a. whether prolonged sequences of MI produce mental fatigue and b. whether mental fatigue affects MI EEG class separability. Eleven participants participated in the MI experiment, 5 of which quit in the middle because of experiencing high fatigue. The growth of fatigue was monitored using the Kernel Partial Least Square (KPLS) algorithm on the remaining 6 participants which shows that MI induces substantial mental fatigue. Statistical analysis of the effect of fatigue on motor imagery performance shows that high fatigue level significantly decreases MI EEG separability. Collectively, these results portray an MI-fatigue inter-connection, emphasizing the necessity of developing adaptive MI BCI by tracking mental fatigue.

Collective sparse symmetric non-negative matrix factorization for identifying overlapping communities in resting-state brain functional networks.

Resting-state functional magnetic resonance imaging (rs-fMRI) provides a valuable tool to study spontaneous brain activity. Using rs-fMRI, researchers have extensively studied the organization of the brain functional network and found several consistent communities consisting of functionally connected but spatially separated brain regions across subjects. However, increasing evidence in many disciplines has shown that most realistic complex networks have overlapping community structure. Only recently has the overlapping community structure drawn increasing interest in the domain of brain network studies. Another issue is that the inter-subject variability is often not directly reflected in the process of community detection at the group level. In this paper, we propose a novel method called collective sparse symmetric non-negative matrix factorization (cssNMF) to address these issues. The cssNMF approach identifies the group-level overlapping communities across subjects and in the meantime preserves the information of individual variation in brain functional network organization. To comprehensively validate cssNMF, a simulated fMRI dataset with ground-truth, a real rs-fMRI dataset with two repeated sessions and another different real rs-fMRI dataset have been used for performance comparison in the experiment. Experimental results show that the proposed cssNMF method accurately and stably identifies group-level overlapping communities across subjects as well as individual differences in network organization with neurophysiologically meaningful interpretations. This research extends our understanding of the common underlying community structures and individual differences in community strengths in brain functional network organization.

Spatiotemporal Phase Synchronization in Adaptive Reconfiguration from Action Observation Network to Mentalizing Network for Understanding Other's Action Intention.

In action intention understanding, the mirror system is involved in perception-action matching process and the mentalizing system underlies higher-level intention inference. By analyzing the dynamic functional connectivity in α (8-12 Hz) and ß (12-30 Hz) frequency bands over a "hand-cup interaction" observation task, this study investigates the topological transition from the action observation network (AON) to the mentalizing network (MZN), and estimates their functional relevance for intention identification from other's different action kinematics. Sequential brain microstates were extracted based on event-related potentials (ERPs), in which significantly differing neuronal responses were found in N170-P200 related to perceptually matching kinematic profiles and P400-700 involved in goal inference. Inter-electrode weighted phase lag index analysis on the ERP microstates revealed a shift of hub centrality salient in α frequency band, from the AON dominated by left-lateral frontal-premotor-temporal and temporal-parietooccipital synchronizations to the MZN consisting of more bilateral frontal-parietal and temporal-parietal synchronizations. As compared with usual actions, intention identification of unintelligible actions induces weaker synchronizations in the AON but dramatically increased connectivity in right frontal-temporal-parietal regions of the MZN, indicating a spatiotemporally complementary effect between the functional network configurations involved in mirror and mentalizing processes. Perceptual processing in observing usual/unintelligible actions decreases/increases requirements for intention inference, which would induce less/greater functional network reorganization on the way to mentalization. From the comparison, our study suggests that the adaptive topological changes from the AON to the MZN indicate implicit causal association between the mirror and mentalizing systems for decoding others' intentionality.

Classification of motor imagery tasks for BCI with multiresolution analysis and multiobjective feature selection.

BACKGROUND: Brain-computer interfacing (BCI) applications based on the classification of electroencephalographic (EEG) signals require solving high-dimensional pattern classification problems with such a relatively small number of training patterns that curse of dimensionality problems usually arise. Multiresolution analysis (MRA) has useful properties for signal analysis in both temporal and spectral analysis, and has been broadly used in the BCI field. However, MRA usually increases the dimensionality of the input data. Therefore, some approaches to feature selection or feature dimensionality reduction should be considered for improving the performance of the MRA based BCI. METHODS: This paper investigates feature selection in the MRA-based frameworks for BCI. Several wrapper approaches to evolutionary multiobjective feature selection are proposed with different structures of classifiers. They are evaluated by comparing with baseline methods using sparse representation of features or without feature selection. RESULTS AND CONCLUSION: The statistical analysis, by applying the Kolmogorov-Smirnoff and Kruskal-Wallis tests to the means of the Kappa values evaluated by using the test patterns in each approach, has demonstrated some advantages of the proposed approaches. In comparison with the baseline MRA approach used in previous studies, the proposed evolutionary multiobjective feature selection approaches provide similar or even better classification performances, with significant reduction in the number of features that need to be computed.

Brain connectivity patterns derived from aging-related alterations in dynamic brain functional networks and their potential as features for brain age classification.

Objective.Recent studies have demonstrated that the analysis of brain functional networks (BFNs) is a powerful tool for exploring brain aging and age-related neurodegenerative diseases. However, investigating the mechanism of brain aging associated with dynamic BFN is still limited. The purpose of this study is to develop a novel scheme to explore brain aging patterns by constructing dynamic BFN using resting-state functional magnetic resonance imaging data.Approach.A dynamic sliding-windowed non-negative block-diagonal representation (dNBDR) method is proposed for constructing dynamic BFN, based on which a collection of dynamic BFN measures are suggested for examining age-related differences at the group level and used as features for brain age classification at the individual level.Results.The experimental results reveal that the dNBDR method is superior to the sliding time window with Pearson correlation method in terms of dynamic network structure quality. Additionally, significant alterations in dynamic BFN structures exist across the human lifespan. Specifically, average node flexibility and integration coefficient increase with age, while the recruitment coefficient shows a decreased trend. The proposed feature extraction scheme based on dynamic BFN achieved the highest accuracy of 78.7% in classifying three brain age groups.Significance. These findings suggest that dynamic BFN measures, dynamic community structure metrics in particular, play an important role in quantitatively assessing brain aging.

Reliable and fast automatic artifact rejection of Long-Term EEG recordings based on Isolation Forest.

Long-term electroencephalogram (Long-Term EEG) has the capacity to monitor over a long period, making it a valuable tool in medical institutions. However, due to the large volume of patient data, selecting clean data segments from raw Long-Term EEG for further analysis is an extremely time-consuming and labor-intensive task. Furthermore, the various actions of patients during recording make it difficult to use algorithms to denoise part of the EEG data, and thus lead to the rejection of these data. Therefore, tools for the quick rejection of heavily corrupted epochs in Long-Term EEG records are highly beneficial. In this paper, a new reliable and fast automatic artifact rejection method for Long-Term EEG based on Isolation Forest (IF) is proposed. Specifically, the IF algorithm is repetitively applied to detect outliers in the EEG data, and the boundary of inliers is promptly adjusted by using a statistical indicator to make the algorithm proceed in an iterative manner. The iteration is terminated when the distance metric between clean epochs and artifact-corrupted epochs remains unchanged. Six statistical indicators (i.e., min, max, median, mean, kurtosis, and skewness) are evaluated by setting them as centroid to adjust the boundary during iteration, and the proposed method is compared with several state-of-the-art methods on a retrospectively collected dataset. The experimental results indicate that utilizing the min value of data as the centroid yields the most optimal performance, and the proposed method is highly efficacious and reliable in the automatic artifact rejection of Long-Term EEG, as it significantly improves the overall data quality. Furthermore, the proposed method surpasses compared methods on most data segments with poor data quality, demonstrating its superior capacity to enhance the data quality of the heavily corrupted data. Besides, owing to the linear time complexity of IF, the proposed method is much faster than other methods, thus providing an advantage when dealing with extensive datasets.

Motor Imagery EEG Classification Based on a Weighted Multi-Branch Structure Suitable for Multisubject Data.

OBJECTIVE: Electroencephalogram (EEG) signal recognition based on deep learning technology requires the support of sufficient data. However, training data scarcity usually occurs in subject-specific motor imagery tasks unless multisubject data can be used to enlarge training data. Unfortunately, because of the large discrepancies between data distributions from different subjects, model performance could only be improved marginally or even worsened by simply training on multisubject data. METHOD: This article proposes a novel weighted multi-branch (WMB) structure for handling multisubject data to solve the problem, in which each branch is responsible for fitting a pair of source-target subject data and adaptive weights are used to integrate all branches or select branches with the largest weights to make the final decision. The proposed WMB structure was applied to six well-known deep learning models (EEGNet, Shallow ConvNet, Deep ConvNet, ResNet, MSFBCNN, and EEG_TCNet) and comprehensive experiments were conducted on EEG datasets BCICIV-2a, BCICIV-2b, high gamma dataset (HGD) and two supplementary datasets. RESULT: Superior results against the state-of-the-art models have demonstrated the efficacy of the proposed method in subject-specific motor imagery EEG classification. For example, the proposed WMB_EEGNet achieved classification accuracies of 84.14%, 90.23%, and 97.81% on BCICIV-2a, BCICIV-2b and HGD, respectively. CONCLUSION: It is clear that the proposed WMB structure is capable to make good use of multisubject data with large distribution discrepancies for subject-specific EEG classification.

Recognition of unilateral lower limb movement based on EEG signals with ERP-PCA analysis.

It has been confirmed that motor imagery (MI) and motor execution (ME) share a subset of mechanisms underlying motor cognition. In contrast to the well-studied laterality of upper limb movement, the laterality hypothesis of lower limb movement also exists, but it needs to be characterized by further investigation. This study used electroencephalographic (EEG) recordings of 27 subjects to compare the effects of bilateral lower limb movement in the MI and ME paradigms. Event-related potential (ERP) recorded was decomposed into meaningful and useful representatives of the electrophysiological components, such as N100 and P300. Principal components analysis (PCA) was used to trace the characteristics of ERP components temporally and spatially, respectively. The hypothesis of this study is that the functional opposition of unilateral lower limbs of MI and ME should be reflected in the different alterations of the spatial distribution of lateralized activity. Meanwhile, the significant ERP-PCA components of the EEG signals as identifiable feature sets were applied with support vector machine to identify left and right lower limb movement tasks. The average classification accuracy over all subjects is up to 61.85% for MI and 62.94% for ME. The proportion of subjects with significant results are 51.85% for MI and 59.26% for ME, respectively. Therefore, a potential new classification model for lower limb movement can be applied on brain computer interface (BCI) systems in the future.

Biomarkers Derived from Alterations in Overlapping Community Structure of Resting-state Brain Functional Networks for Detecting Alzheimer's Disease.

Recent studies show that overlapping community structure is an important feature of the brain functional network. However, alterations in such overlapping community structure in Alzheimer's disease (AD) patients have not been examined yet. In this study, we investigate the overlapping community structure in AD by using resting-state functional magnetic resonance imaging (rs-fMRI) data. The collective sparse symmetric non-negative matrix factorization (cssNMF) is adopted to detect the overlapping community structure. Experimental results on 28 AD patients and 32 normal controls (NCs) from the ADNI2 dataset show that the two groups have remarkable differences in terms of the optimal number of communities, the hierarchy of communities detected at different scales, network functional segregation, and nodal functional diversity. In particular, the frontal-parietal and basal ganglia networks exhibit significant differences between the two groups. A machine learning framework proposed in this paper for AD detection achieved an accuracy of 76.7% when using the detected community strengths of the frontal-parietal and basal ganglia networks only as input features. These findings provide novel insights into the understanding of pathological changes in the brain functional network organization of AD and show the potential of the community structure-related features for AD detection.

Frail patients having vascular surgery during the early COVID-19 pandemic experienced high rates of adverse perioperative events and amputation.

BACKGROUND: Frailty predicts adverse perioperative outcomes and increased mortality in patients having vascular surgery. Frailty assessment is a potential tool to inform resource allocation, and shared decision-making about vascular surgery in the resource constrained COVID-19 pandemic environment. This cohort study describes the prevalence of frailty in patients having vascular surgery and the association between frailty, mortality and perioperative outcomes. METHODS: The COVID-19 Vascular Service in Australia (COVER-AU) prospective cohort study evaluates 30-day and six-month outcomes for consecutive patients having vascular surgery in 11 Australian vascular units, March-July 2020. The primary outcome was mortality, with secondary outcomes procedure-related outcomes and hospital utilization. Frailty was assessed using the nine-point visual Clinical Frailty Score, scores of 5 or more considered frail. RESULTS: Of the 917 patients enrolled, 203 were frail (22.1%). The 30 day and 6 month mortality was 2.0% (n = 20) and 5.9% (n = 35) respectively with no significant difference between frail and non-frail patients (OR 1.68, 95%CI 0.79-3.54). However, frail patients stayed longer in hospital, had more perioperative complications, and were more likely to be readmitted or have a reoperation when compared to non-frail patients. At 6 months, frail patients had twice the odds of major amputation compared to non-frail patients, after adjustment (OR 2.01; 95% CI 1.17-3.78), driven by a high rate of amputation during the period of reduced surgical activity. CONCLUSION: Our findings highlight that older, frail patients, experience potentially preventable adverse outcomes and there is a need for targeted interventions to optimize care, especially in times of healthcare stress.

Small facial image dataset augmentation using conditional GANs based on incomplete edge feature input.

Image data collection and labelling is costly or difficult in many real applications. Generating diverse and controllable images using conditional generative adversarial networks (GANs) for data augmentation from a small dataset is promising but challenging as deep convolutional neural networks need a large training dataset to achieve reasonable performance in general. However, unlabeled and incomplete features (e.g., unintegral edges, simplified lines, hand-drawn sketches, discontinuous geometry shapes, etc.) can be conveniently obtained through pre-processing the training images and can be used for image data augmentation. This paper proposes a conditional GAN framework for facial image augmentation using a very small training dataset and incomplete or modified edge features as conditional input for diversity. The proposed method defines a new domain or space for refining interim images to prevent overfitting caused by using a very small training dataset and enhance the tolerance of distortions caused by incomplete edge features, which effectively improves the quality of facial image augmentation with diversity. Experimental results have shown that the proposed method can generate high-quality images of good diversity when the GANs are trained using very sparse edges and a small number of training samples. Compared to the state-of-the-art edge-to-image translation methods that directly convert sparse edges to images, when using a small training dataset, the proposed conditional GAN framework can generate facial images with desirable diversity and acceptable distortions for dataset augmentation and significantly outperform the existing methods in terms of the quality of synthesised images, evaluated by Fréchet Inception Distance (FID) and Kernel Inception Distance (KID) scores.

Traumatic arteriovenous fistula: a 25-year delay in presentation.

A 64-year-old lady presented with a 6-month history of worsening unilateral leg swelling, with an audible bruit of the popliteal artery. Arterial duplex ultrasound confirmed the presence of an arteriovenous fistula (AVF) between the posterior tibial artery and vein. Upon thorough history, it was discovered that the patient had sustained a stab wound to this region some 25 years prior. The fistula was successfully managed endovascularly by means of a covered stent. This case highlights an unusual delayed presentation of an AVF and demonstrates the effectiveness of endovascular treatment of this condition.

Adaptive feature extraction in EEG-based motor imagery BCI: tracking mental fatigue.

OBJECTIVE: Electroencephalogram (EEG) signals are non-stationary. This could be due to internal fluctuation of brain states such as fatigue, frustration, etc. This necessitates the development of adaptive brain-computer interfaces (BCI) whose performance does not deteriorate significantly with the adversary change in the cognitive state. In this paper, we put forward an unsupervised adaptive scheme to adapt the feature extractor of motor imagery (MI) BCIs by tracking the fatigue level of the user. APPROACH: Eleven subjects participated in the study during which they accomplished MI tasks while self-reporting their perceived levels of mental fatigue. Out of the 11 subjects, only six completed the whole experiment, while the others quit in the middle because of experiencing high fatigue. The adaptive feature extractor is attained through the adaptation of the common spatial patterns (CSP), one of the most popular feature extraction algorithms in EEG-based BCIs. The proposed method was analyzed in two ways: offline and in near real-time. The separability of the MI EEG features extracted by the proposed adaptive CSP (ADCSP) has been compared with that by the conventional CSP (C-CSP) and another CSP based adaptive method (ACSP) in terms of: Davies Bouldin index (DBI), Fisher score (FS) and Dunn's index (DI). MAIN RESULTS: Experimental results show significant improvement in the separability of MI EEG features extracted by ADCSP as compared to that by C-CSP and ACSP. SIGNIFICANCE: Collectively, the results of the experiments in this study suggest that adapting CSP based on mental fatigue can improve the class separability of MI EEG features.

Deep learning for EEG-based Motor Imagery classification: Accuracy-cost trade-off.

Electroencephalography (EEG) datasets are often small and high dimensional, owing to cumbersome recording processes. In these conditions, powerful machine learning techniques are essential to deal with the large amount of information and overcome the curse of dimensionality. Artificial Neural Networks (ANNs) have achieved promising performance in EEG-based Brain-Computer Interface (BCI) applications, but they involve computationally intensive training algorithms and hyperparameter optimization methods. Thus, an awareness of the quality-cost trade-off, although usually overlooked, is highly beneficial. In this paper, we apply a hyperparameter optimization procedure based on Genetic Algorithms to Convolutional Neural Networks (CNNs), Feed-Forward Neural Networks (FFNNs), and Recurrent Neural Networks (RNNs), all of them purposely shallow. We compare their relative quality and energy-time cost, but we also analyze the variability in the structural complexity of networks of the same type with similar accuracies. The experimental results show that the optimization procedure improves accuracy in all models, and that CNN models with only one hidden convolutional layer can equal or slightly outperform a 6-layer Deep Belief Network. FFNN and RNN were not able to reach the same quality, although the cost was significantly lower. The results also highlight the fact that size within the same type of network is not necessarily correlated with accuracy, as smaller models can and do match, or even surpass, bigger ones in performance. In this regard, overfitting is likely a contributing factor since deep learning approaches struggle with limited training examples.

Differential recruitment of brain networks in single-digit addition and multiplication: Evidence from EEG oscillations in theta and lower alpha bands.

Previous neuroimaging research investigating dissociation between single-digit addition and multiplication has suggested that the former placed more reliance on the visuo-spatial processing whereas the latter on the verbal processing. However, there has been little exploration into the disassociation in spatio-temporal dynamics of the oscillatory brain activity in specific frequency bands during the two arithmetic operations. To address this issue, the electroencephalogram (EEG) data were recorded from 19 participants engaged in a delayed verification arithmetic task. By analyzing oscillatory EEG activity in theta (5-7 Hz) and lower alpha frequency (9-10 Hz) bands, we found different patterns of oscillatory brain activity between single-digit addition and multiplication during the early processing stage (0-400 ms post-operand onset). Experiment results in this study showed a larger phasic increase of theta-band power for addition than for multiplication in the midline and the right frontal and central regions during the operator and operands presentation intervals, which was extended to the right parietal and the right occipito-temporal regions during the interval immediately after the operands presentation. In contrast, during multiplication higher phase-locking in lower alpha band was evident in the centro-parietal regions during the operator presentation, which was extended to the left fronto-central and anterior regions during the operands presentation. Besides, we found stronger theta phase synchrony between the parietal areas and the right occipital areas for single-digit addition than for multiplication during operands encoding. These findings of oscillatory brain activity extend the previous observations on functional dissociation between the two arithmetic operations.

Radiotherapy utilization in developing countries: An IAEA study.

BACKGROUND: The planning of national radiotherapy (RT) services requires a thorough knowledge of the country's cancer epidemiology profile, the radiotherapy utilization (RTU) rates and a future projection of these data. Previous studies have established RTU rates in high-income countries. METHODS: Optimal RTU (oRTU) rates were determined for nine middle-income countries, following the epidemiological evidence-based method. The actual RTU (aRTU) rates were calculated dividing the total number of new notifiable cancer patients treated with radiotherapy in 2012 by the total number of cancer patients diagnosed in the same year in each country. An analysis of the characteristics of patients and treatments in a series of 300 consecutive radiotherapy patients shed light on the particular patient and treatments profile in the participating countries. RESULTS: The median oRTU rate for the group of nine countries was 52% (47-56%). The median aRTU rate for the nine countries was 28% (9-46%). These results show that the real proportion of cancer patients receiving RT is lower than the optimal RTU with a rate difference between 10-42.7%. The median percent-unmet need was 47% (18-82.3%). CONCLUSIONS: The optimal RTU rate in middle-income countries did not differ significantly from that previously found in high-income countries. The actual RTU rates were consistently lower than the optimal, in particular in countries with limited resources and a large population.

Neurocognitive mechanisms of mathematical giftedness: A literature review.

Mathematically gifted children/adolescents have demonstrated exceptional abilities and traits in logical reasoning, mental imagery, and creative thinking. In the field of cognitive neuroscience, the past studies on mathematically gifted brains have concentrated on investigating event-related brain activation regions, cerebral laterality of cognitive functions, functional specialization that is uniquely dedicated for specific cognitive purposes, and functional interactions among discrete brain regions. From structural and functional perspectives, these studies have witnessed both "general" and "unique" neural characteristics of mathematically gifted brains. In this article, the theoretical background, empirical studies, and neurocognitive mechanisms of mathematically gifted children/adolescents are reviewed. Based on the integration of the findings, some potential directions for the future research are identified and discussed.