A novel method for automated classification of epileptiform activity in the human electroencephalogram-based on independent component analysis.

Diagnosis of several neurological disorders is based on the detection of typical pathological patterns in the electroencephalogram (EEG). This is a time-consuming task requiring significant training and experience. Automatic detection of these EEG patterns would greatly assist in quantitative analysis and interpretation. We present a method, which allows automatic detection of epileptiform events and discrimination of them from eye blinks, and is based on features derived using a novel application of independent component analysis. The algorithm was trained and cross validated using seven EEGs with epileptiform activity. For epileptiform events with compensation for eyeblinks, the sensitivity was 65 +/- 22% at a specificity of 86 +/- 7% (mean +/- SD). With feature extraction by PCA or classification of raw data, specificity reduced to 76 and 74%, respectively, for the same sensitivity. On exactly the same data, the commercially available software Reveal had a maximum sensitivity of 30% and concurrent specificity of 77%. Our algorithm performed well at detecting epileptiform events in this preliminary test and offers a flexible tool that is intended to be generalized to the simultaneous classification of many waveforms in the EEG.

Decoding decisions in healthy subjects and auditory discrimination in comatose patients through single-trial topographic EEG analyses

Neuroimaging studies analyzing neurophysiological signals are typically based on comparing averages of peri-stimulus epochs across experimental conditions. This approach can however be problematic in the case of high-level cognitive tasks, where response variability across trials is expected to be high and in cases where subjects cannot be considered part of a group. The main goal of this thesis has been to address this issue by developing a novel approach for analyzing electroencephalography (EEG) responses at the single-trial level. This approach takes advantage of the spatial distribution of the electric field on the scalp (topography) and exploits repetitions across trials for quantifying the degree of discrimination between experimental conditions through a classification scheme. In the first part of this thesis, I developed and validated this new method (Tzovara et al., 2012a,b). Its general applicability was demonstrated with three separate datasets, two in the visual modality and one in the auditory. This development allowed then to target two new lines of research, one in basic and one in clinical neuroscience, which represent the second and third part of this thesis respectively. For the second part of this thesis (Tzovara et al., 2012c), I employed the developed method for assessing the timing of exploratory decision-making. Using single-trial topographic EEG activity during presentation of a choice's payoff, I could predict the subjects' subsequent decisions. This prediction was due to a topographic difference which appeared on average at ~516ms after the presentation of payoff and was subject-specific. These results exploit for the first time the temporal correlates of individual subjects' decisions and additionally show that the underlying neural generators start differentiating their responses already ~880ms before the button press. Finally, in the third part of this project, I focused on a clinical study with the goal of assessing the degree of intact neural functions in comatose patients. Auditory EEG responses were assessed through a classical mismatch negativity paradigm, during the very early phase of coma, which is currently under-investigated. By taking advantage of the decoding method developed in the first part of the thesis, I could quantify the degree of auditory discrimination at the single patient level (Tzovara et al., in press). Our results showed for the first time that even patients who do not survive the coma can discriminate sounds at the neural level, during the first hours after coma onset. Importantly, an improvement in auditory discrimination during the first 48hours of coma was predictive of awakening and survival, with 100% positive predictive value. - L'analyse des signaux électrophysiologiques en neuroimagerie se base typiquement sur la comparaison des réponses neurophysiologiques à différentes conditions expérimentales qui sont moyennées après plusieurs répétitions d'une tâche. Pourtant, cette approche peut être problématique dans le cas des fonctions cognitives de haut niveau, où la variabilité des réponses entre les essais peut être très élevéeou dans le cas où des sujets individuels ne peuvent pas être considérés comme partie d'un groupe. Le but principal de cette thèse est d'investiguer cette problématique en développant une nouvelle approche pour l'analyse des réponses d'électroencephalographie (EEG) au niveau de chaque essai. Cette approche se base sur la modélisation de la distribution du champ électrique sur le crâne (topographie) et profite des répétitions parmi les essais afin de quantifier, à l'aide d'un schéma de classification, le degré de discrimination entre des conditions expérimentales. Dans la première partie de cette thèse, j'ai développé et validé cette nouvelle méthode (Tzovara et al., 2012a,b). Son applicabilité générale a été démontrée avec trois ensembles de données, deux dans le domaine visuel et un dans l'auditif. Ce développement a permis de cibler deux nouvelles lignes de recherche, la première dans le domaine des neurosciences cognitives et l'autre dans le domaine des neurosciences cliniques, représentant respectivement la deuxième et troisième partie de ce projet. En particulier, pour la partie cognitive, j'ai appliqué cette méthode pour évaluer l'information temporelle de la prise des décisions (Tzovara et al., 2012c). En se basant sur l'activité topographique de l'EEG au niveau de chaque essai pendant la présentation de la récompense liée à un choix, on a pu prédire les décisions suivantes des sujets (en termes d'exploration/exploitation). Cette prédiction s'appuie sur une différence topographique qui apparaît en moyenne ~516ms après la présentation de la récompense. Ces résultats exploitent pour la première fois, les corrélés temporels des décisions au niveau de chaque sujet séparément et montrent que les générateurs neuronaux de ces décisions commencent à différentier leurs réponses déjà depuis ~880ms avant que les sujets appuient sur le bouton. Finalement, pour la dernière partie de ce projet, je me suis focalisée sur une étude Clinique afin d'évaluer le degré des fonctions neuronales intactes chez les patients comateux. Des réponses EEG auditives ont été examinées avec un paradigme classique de mismatch negativity, pendant la phase précoce du coma qui est actuellement sous-investiguée. En utilisant la méthode de décodage développée dans la première partie de la thèse, j'ai pu quantifier le degré de discrimination auditive au niveau de chaque patient (Tzovara et al., in press). Nos résultats montrent pour la première fois que même des patients comateux qui ne vont pas survivre peuvent discriminer des sons au niveau neuronal, lors de la phase aigue du coma. De plus, une amélioration dans la discrimination auditive pendant les premières 48heures du coma a été observée seulement chez des patients qui se sont réveillés par la suite (100% de valeur prédictive pour un réveil).

Decoding sequence learning from single-trial intracranial EEG in humans.

We propose and validate a multivariate classification algorithm for characterizing changes in human intracranial electroencephalographic data (iEEG) after learning motor sequences. The algorithm is based on a Hidden Markov Model (HMM) that captures spatio-temporal properties of the iEEG at the level of single trials. Continuous intracranial iEEG was acquired during two sessions (one before and one after a night of sleep) in two patients with depth electrodes implanted in several brain areas. They performed a visuomotor sequence (serial reaction time task, SRTT) using the fingers of their non-dominant hand. Our results show that the decoding algorithm correctly classified single iEEG trials from the trained sequence as belonging to either the initial training phase (day 1, before sleep) or a later consolidated phase (day 2, after sleep), whereas it failed to do so for trials belonging to a control condition (pseudo-random sequence). Accurate single-trial classification was achieved by taking advantage of the distributed pattern of neural activity. However, across all the contacts the hippocampus contributed most significantly to the classification accuracy for both patients, and one fronto-striatal contact for one patient. Together, these human intracranial findings demonstrate that a multivariate decoding approach can detect learning-related changes at the level of single-trial iEEG. Because it allows an unbiased identification of brain sites contributing to a behavioral effect (or experimental condition) at the level of single subject, this approach could be usefully applied to assess the neural correlates of other complex cognitive functions in patients implanted with multiple electrodes.

Event-related potential analyses via single-subject topographic classification.

Introduction: Responses to external stimuli are typically investigated by averaging peri-stimulus electroencephalography (EEG) epochs in order to derive event-related potentials (ERPs) across the electrode montage, under the assumption that signals that are related to the external stimulus are fixed in time across trials. We demonstrate the applicability of a single-trial model based on patterns of scalp topographies (De Lucia et al, 2007) that can be used for ERP analysis at the single-subject level. The model is able to classify new trials (or groups of trials) with minimal a priori hypotheses, using information derived from a training dataset. The features used for the classification (the topography of responses and their latency) can be neurophysiologically interpreted, because a difference in scalp topography indicates a different configuration of brain generators. An above chance classification accuracy on test datasets implicitly demonstrates the suitability of this model for EEG data. Methods: The data analyzed in this study were acquired from two separate visual evoked potential (VEP) experiments. The first entailed passive presentation of checkerboard stimuli to each of the four visual quadrants (hereafter, "Checkerboard Experiment") (Plomp et al, submitted). The second entailed active discrimination of novel versus repeated line drawings of common objects (hereafter, "Priming Experiment") (Murray et al, 2004). Four subjects per experiment were analyzed, using approx. 200 trials per experimental condition. These trials were randomly separated in training (90%) and testing (10%) datasets in 10 independent shuffles. In order to perform the ERP analysis we estimated the statistical distribution of voltage topographies by a Mixture of Gaussians (MofGs), which reduces our original dataset to a small number of representative voltage topographies. We then evaluated statistically the degree of presence of these template maps across trials and whether and when this was different across experimental conditions. Based on these differences, single-trials or sets of a few single-trials were classified as belonging to one or the other experimental condition. Classification performance was assessed using the Receiver Operating Characteristic (ROC) curve. Results: For the Checkerboard Experiment contrasts entailed left vs. right visual field presentations for upper and lower quadrants, separately. The average posterior probabilities, indicating the presence of the computed template maps in time and across trials revealed significant differences starting at ~60-70 ms post-stimulus. The average ROC curve area across all four subjects was 0.80 and 0.85 for upper and lower quadrants, respectively and was in all cases significantly higher than chance (unpaired t-test, p<0.0001). In the Priming Experiment, we contrasted initial versus repeated presentations of visual object stimuli. Their posterior probabilities revealed significant differences, which started at 250ms post-stimulus onset. The classification accuracy rates with single-trial test data were at chance level. We therefore considered sub-averages based on five single trials. We found that for three out of four subjects' classification rates were significantly above chance level (unpaired t-test, p<0.0001). Conclusions: The main advantage of the present approach is that it is based on topographic features that are readily interpretable along neurophysiologic lines. As these maps were previously normalized by the overall strength of the field potential on the scalp, a change in their presence across trials and between conditions forcibly reflects a change in the underlying generator configurations. The temporal periods of statistical difference between conditions were estimated for each training dataset for ten shuffles of the data. Across the ten shuffles and in both experiments, we observed a high level of consistency in the temporal periods over which the two conditions differed. With this method we are able to analyze ERPs at the single-subject level providing a novel tool to compare normal electrophysiological responses versus single cases that cannot be considered part of any cohort of subjects. This aspect promises to have a strong impact on both basic and clinical research.

A definition and classification of status epilepticus - Report of the ILAE Task Force on Classification of Status Epilepticus.

The Commission on Classification and Terminology and the Commission on Epidemiology of the International League Against Epilepsy (ILAE) have charged a Task Force to revise concepts, definition, and classification of status epilepticus (SE). The proposed new definition of SE is as follows: Status epilepticus is a condition resulting either from the failure of the mechanisms responsible for seizure termination or from the initiation of mechanisms, which lead to abnormally, prolonged seizures (after time point t1 ). It is a condition, which can have long-term consequences (after time point t2 ), including neuronal death, neuronal injury, and alteration of neuronal networks, depending on the type and duration of seizures. This definition is conceptual, with two operational dimensions: the first is the length of the seizure and the time point (t1 ) beyond which the seizure should be regarded as "continuous seizure activity." The second time point (t2 ) is the time of ongoing seizure activity after which there is a risk of long-term consequences. In the case of convulsive (tonic-clonic) SE, both time points (t1 at 5 min and t2 at 30 min) are based on animal experiments and clinical research. This evidence is incomplete, and there is furthermore considerable variation, so these time points should be considered as the best estimates currently available. Data are not yet available for other forms of SE, but as knowledge and understanding increase, time points can be defined for specific forms of SE based on scientific evidence and incorporated into the definition, without changing the underlying concepts. A new diagnostic classification system of SE is proposed, which will provide a framework for clinical diagnosis, investigation, and therapeutic approaches for each patient. There are four axes: (1) semiology; (2) etiology; (3) electroencephalography (EEG) correlates; and (4) age. Axis 1 (semiology) lists different forms of SE divided into those with prominent motor systems, those without prominent motor systems, and currently indeterminate conditions (such as acute confusional states with epileptiform EEG patterns). Axis 2 (etiology) is divided into subcategories of known and unknown causes. Axis 3 (EEG correlates) adopts the latest recommendations by consensus panels to use the following descriptors for the EEG: name of pattern, morphology, location, time-related features, modulation, and effect of intervention. Finally, axis 4 divides age groups into neonatal, infancy, childhood, adolescent and adulthood, and elderly.

Interrater variability of EEG interpretation in comatose cardiac arrest patients.

OBJECTIVE: EEG is widely used to predict outcome in comatose cardiac arrest patients, but its value has been limited by lack of a uniform classification. We used the EEG terminology proposed by the American Clinical Neurophysiology Society (ACNS) to assess interrater variability in a cohort of cardiac arrest patients included in the Target Temperature Management trial. The main objective was to evaluate if malignant EEG-patterns could reliably be identified. METHODS: Full-length EEGs from 103 comatose cardiac arrest patients were interpreted by four EEG-specialists with different nationalities who were blinded for patient outcome. Percent agreement and kappa (κ) for the categories in the ACNS EEG terminology and for prespecified malignant EEG-patterns were calculated. RESULTS: There was substantial interrater agreement (κ 0.71) for highly malignant patterns and moderate agreement (κ 0.42) for malignant patterns. Substantial agreement was found for malignant periodic or rhythmic patterns (κ 0.72) while agreement for identifying an unreactive EEG was fair (κ 0.26). CONCLUSIONS: The ACNS EEG terminology can be used to identify highly malignant EEG-patterns in post cardiac arrest patients in an international context with high reliability. SIGNIFICANCE: The establishment of strict criteria with high transferability between interpreters will increase the usefulness of routine EEG to assess neurological prognosis after cardiac arrest.

Urban image classification with semisupervised multiscale cluster kernels

This paper presents a semisupervised support vector machine (SVM) that integrates the information of both labeled and unlabeled pixels efficiently. Method's performance is illustrated in the relevant problem of very high resolution image classification of urban areas. The SVM is trained with the linear combination of two kernels: a base kernel working only with labeled examples is deformed by a likelihood kernel encoding similarities between labeled and unlabeled examples. Results obtained on very high resolution (VHR) multispectral and hyperspectral images show the relevance of the method in the context of urban image classification. Also, its simplicity and the few parameters involved make the method versatile and workable by unexperienced users.

Diagnostic yield of short-term video-EEG monitoring for epilepsy and PNESs: a European assessment.

INTRODUCTION: Although long-term video-EEG monitoring (LVEM) is routinely used to investigate paroxysmal events, short-term video-EEG monitoring (SVEM) lasting <24 h is increasingly recognized as a cost-effective tool. Since, however, relatively few studies addressed the yield of SVEM among different diagnostic groups, we undertook the present study to investigate this aspect. METHODS: We retrospectively analyzed 226 consecutive SVEM recordings over 6 years. All patients were referred because routine EEGs were inconclusive. Patients were classified into 3 suspected diagnostic groups: (1) group with epileptic seizures, (2) group with psychogenic nonepileptic seizures (PNESs), and (3) group with other or undetermined diagnoses. We assessed recording lengths, interictal epileptiform discharges, epileptic seizures, PNESs, and the definitive diagnoses obtained after SVEM. RESULTS: The mean age was 34 (±18.7) years, and the median recording length was 18.6 h. Among the 226 patients, 127 referred for suspected epilepsy - 73 had a diagnosis of epilepsy, none had a diagnosis of PNESs, and 54 had other or undetermined diagnoses post-SVEM. Of the 24 patients with pre-SVEM suspected PNESs, 1 had epilepsy, 12 had PNESs, and 11 had other or undetermined diagnoses. Of the 75 patients with other diagnoses pre-SVEM, 17 had epilepsy, 11 had PNESs, and 47 had other or undetermined diagnoses. After SVEM, 15 patients had definite diagnoses other than epilepsy or PNESs, while in 96 patients, diagnosis remained unclear. Overall, a definitive diagnosis could be reached in 129/226 (57%) patients. CONCLUSIONS: This study demonstrates that in nearly 3/5 patients without a definitive diagnosis after routine EEG, SVEM allowed us to reach a diagnosis. This procedure should be encouraged in this setting, given its time-effectiveness compared with LVEM.

Automatic Mallampati Classification Using Active Appearance Models

Difficult tracheal intubation assessment is an important research topic in anesthesia as failed intubations are important causes of mortality in anesthetic practice. The modified Mallampati score is widely used, alone or in conjunction with other criteria, to predict the difficulty of intubation. This work presents an automatic method to assess the modified Mallampati score from an image of a patient with the mouth wide open. For this purpose we propose an active appearance models (AAM) based method and use linear support vector machines (SVM) to select a subset of relevant features obtained using the AAM. This feature selection step proves to be essential as it improves drastically the performance of classification, which is obtained using SVM with RBF kernel and majority voting. We test our method on images of 100 patients undergoing elective surgery and achieve 97.9% accuracy in the leave-one-out crossvalidation test and provide a key element to an automatic difficult intubation assessment system.

EEG-based functional brain networks: does the network size matter?

Functional connectivity in human brain can be represented as a network using electroencephalography (EEG) signals. These networks--whose nodes can vary from tens to hundreds--are characterized by neurobiologically meaningful graph theory metrics. This study investigates the degree to which various graph metrics depend upon the network size. To this end, EEGs from 32 normal subjects were recorded and functional networks of three different sizes were extracted. A state-space based method was used to calculate cross-correlation matrices between different brain regions. These correlation matrices were used to construct binary adjacency connectomes, which were assessed with regards to a number of graph metrics such as clustering coefficient, modularity, efficiency, economic efficiency, and assortativity. We showed that the estimates of these metrics significantly differ depending on the network size. Larger networks had higher efficiency, higher assortativity and lower modularity compared to those with smaller size and the same density. These findings indicate that the network size should be considered in any comparison of networks across studies.

Effect of training-sample size and classification difficulty on the accuracy of genomic predictors.

Introduction: As part of the MicroArray Quality Control (MAQC)-II project, this analysis examines how the choice of univariate feature-selection methods and classification algorithms may influence the performance of genomic predictors under varying degrees of prediction difficulty represented by three clinically relevant endpoints. Methods: We used gene-expression data from 230 breast cancers (grouped into training and independent validation sets), and we examined 40 predictors (five univariate feature-selection methods combined with eight different classifiers) for each of the three endpoints. Their classification performance was estimated on the training set by using two different resampling methods and compared with the accuracy observed in the independent validation set. Results: A ranking of the three classification problems was obtained, and the performance of 120 models was estimated and assessed on an independent validation set. The bootstrapping estimates were closer to the validation performance than were the cross-validation estimates. The required sample size for each endpoint was estimated, and both gene-level and pathway-level analyses were performed on the obtained models. Conclusions: We showed that genomic predictor accuracy is determined largely by an interplay between sample size and classification difficulty. Variations on univariate feature-selection methods and choice of classification algorithm have only a modest impact on predictor performance, and several statistically equally good predictors can be developed for any given classification problem.

Anti-tumor necrosis factor drug survival in axial spondyloarthritis is independent of the classification criteria.

To compare the impact of meeting specific classification criteria [modified New York (mNY), European Spondyloarthropathy Study Group (ESSG), and Assessment of SpondyloArthritis international Society (ASAS) criteria] on anti-tumor necrosis factor (anti-TNF) drug retention, and to determine predictive factors of better drug survival. All patients fulfilling the ESSG criteria for axial spondyloarthritis (SpA) with available data on the axial ASAS and mNY criteria, and who had received at least one anti-TNF treatment were retrospectively retrieved in a single academic institution in Switzerland. Drug retention was computed using survival analysis (Kaplan-Meier), adjusted for potential confounders. Of the 137 patients classified as having axial SpA using the ESSG criteria, 112 also met the ASAS axial SpA criteria, and 77 fulfilled the mNY criteria. Drug retention rates at 12 and 24 months for the first biologic therapy were not significantly different between the diagnostic groups. Only the small ASAS non-classified axial SpA group (25 patients) showed a nonsignificant trend toward shorter drug survival. Elevated CRP level, but not the presence of bone marrow edema on magnetic resonance imaging (MRI) scans, was associated with significantly better drug retention (OR 7.9, ICR 4-14). In this cohort, anti-TNF drug survival was independent of the classification criteria. Elevated CRP level, but not positive MRI, was associated with better drug retention.

A Survey of Active Learning Algorithms for Supervised Remote Sensing Image Classification

Defining an efficient training set is one of the most delicate phases for the success of remote sensing image classification routines. The complexity of the problem, the limited temporal and financial resources, as well as the high intraclass variance can make an algorithm fail if it is trained with a suboptimal dataset. Active learning aims at building efficient training sets by iteratively improving the model performance through sampling. A user-defined heuristic ranks the unlabeled pixels according to a function of the uncertainty of their class membership and then the user is asked to provide labels for the most uncertain pixels. This paper reviews and tests the main families of active learning algorithms: committee, large margin, and posterior probability-based. For each of them, the most recent advances in the remote sensing community are discussed and some heuristics are detailed and tested. Several challenging remote sensing scenarios are considered, including very high spatial resolution and hyperspectral image classification. Finally, guidelines for choosing the good architecture are provided for new and/or unexperienced user.