Prediction of Behavioral Improvement Through Resting-State Electroencephalography and Clinical Severity in a Randomized Controlled Trial Testing Bumetanide in Autism Spectrum Disorder.

BACKGROUND: Mechanism-based treatments such as bumetanide are being repurposed for autism spectrum disorder. We recently reported beneficial effects on repetitive behavioral symptoms that might be related to regulating excitation-inhibition (E/I) balance in the brain. Here, we tested the neurophysiological effects of bumetanide and the relationship to clinical outcome variability and investigated the potential for machine learning-based predictions of meaningful clinical improvement. METHODS: Using modified linear mixed models applied to intention-to-treat population, we analyzed E/I-sensitive electroencephalography (EEG) measures before and after 91 days of treatment in the double-blind, randomized, placebo-controlled Bumetanide in Autism Medication and Biomarker study. Resting-state EEG of 82 subjects out of 92 participants (7-15 years) were available. Alpha frequency band absolute and relative power, central frequency, long-range temporal correlations, and functional E/I ratio treatment effects were related to the Repetitive Behavior Scale-Revised (RBS-R) and the Social Responsiveness Scale 2 as clinical outcomes. RESULTS: We observed superior bumetanide effects on EEG, reflected in increased absolute and relative alpha power and functional E/I ratio and in decreased central frequency. Associations between EEG and clinical outcome change were restricted to subgroups with medium to high RBS-R improvement. Using machine learning, medium and high RBS-R improvement could be predicted by baseline RBS-R score and EEG measures with 80% and 92% accuracy, respectively. CONCLUSIONS: Bumetanide exerts neurophysiological effects related to clinical changes in more responsive subsets, in whom prediction of improvement was feasible through EEG and clinical measures.

Improved Manual Annotation of EEG Signals through Convolutional Neural Network Guidance.

The development of validated algorithms for automated handling of artifacts is essential for reliable and fast processing of EEG signals. Recently, there have been methodological advances in designing machine-learning algorithms to improve artifact detection of trained professionals who usually meticulously inspect and manually annotate EEG signals. However, validation of these methods is hindered by the lack of a gold standard as data are mostly private and data annotation is time consuming and error prone. In the effort to circumvent these issues, we propose an iterative learning model to speed up and reduce errors of manual annotation of EEG. We use a convolutional neural network (CNN) to train on expert-annotated eyes-open and eyes-closed resting-state EEG data from typically developing children (n = 30) and children with neurodevelopmental disorders (n = 141). To overcome the circular reasoning of aiming to develop a new algorithm and benchmarking to a manually-annotated gold standard, we instead aim to improve the gold standard by revising the portion of the data that was incorrectly learned by the network. When blindly presented with the selected signals for re-assessment (23% of the data), the two independent expert-annotators changed the annotation in 25% of the cases. Subsequently, the network was trained on the expert-revised gold standard, which resulted in improved separation between artifacts and nonartifacts as well as an increase in balanced accuracy from 74% to 80% and precision from 59% to 76%. These results show that CNNs are promising to enhance manual annotation of EEG artifacts and can be improved further with better gold-standard data.

Bumetanide Effects on Resting-State EEG in Tuberous Sclerosis Complex in Relation to Clinical Outcome: An Open-Label Study.

Neuronal excitation-inhibition (E/I) imbalances are considered an important pathophysiological mechanism in neurodevelopmental disorders. Preclinical studies on tuberous sclerosis complex (TSC), suggest that altered chloride homeostasis may impair GABAergic inhibition and thereby E/I-balance regulation. Correction of chloride homeostasis may thus constitute a treatment target to alleviate behavioral symptoms. Recently, we showed that bumetanide-a chloride-regulating agent-improved behavioral symptoms in the open-label study Bumetanide to Ameliorate Tuberous Sclerosis Complex Hyperexcitable Behaviors trial (BATSCH trial; Eudra-CT: 2016-002408-13). Here, we present resting-state EEG as secondary analysis of BATSCH to investigate associations between EEG measures sensitive to network-level changes in E/I balance and clinical response to bumetanide. EEGs of 10 participants with TSC (aged 8-21 years) were available. Spectral power, long-range temporal correlations (LRTC), and functional E/I ratio (fE/I) in the alpha-frequency band were compared before and after 91 days of treatment. Pre-treatment measures were compared against 29 typically developing children (TDC). EEG measures were correlated with the Aberrant Behavioral Checklist-Irritability subscale (ABC-I), the Social Responsiveness Scale-2 (SRS-2), and the Repetitive Behavior Scale-Revised (RBS-R). At baseline, TSC showed lower alpha-band absolute power and fE/I than TDC. Absolute power increased through bumetanide treatment, which showed a moderate, albeit non-significant, correlation with improvement in RBS-R. Interestingly, correlations between baseline EEG measures and clinical outcomes suggest that most responsiveness might be expected in children with network characteristics around the E/I balance point. In sum, E/I imbalances pointing toward an inhibition-dominated network are present in TSC. We established neurophysiological effects of bumetanide although with an inconclusive relationship with clinical improvement. Nonetheless, our results further indicate that baseline network characteristics might influence treatment response. These findings highlight the possible utility of E/I-sensitive EEG measures to accompany new treatment interventions for TSC. Clinical Trial Registration: EU Clinical Trial Register, EudraCT 2016-002408-13 (www.clinicaltrialsregister.eu/ctr-search/trial/2016-002408-13/NL). Registered 25 July 2016.

Robin's Viewer: Using deep-learning predictions to assist EEG annotation.

Machine learning techniques such as deep learning have been increasingly used to assist EEG annotation, by automating artifact recognition, sleep staging, and seizure detection. In lack of automation, the annotation process is prone to bias, even for trained annotators. On the other hand, completely automated processes do not offer the users the opportunity to inspect the models' output and re-evaluate potential false predictions. As a first step toward addressing these challenges, we developed Robin's Viewer (RV), a Python-based EEG viewer for annotating time-series EEG data. The key feature distinguishing RV from existing EEG viewers is the visualization of output predictions of deep-learning models trained to recognize patterns in EEG data. RV was developed on top of the plotting library Plotly, the app-building framework Dash, and the popular M/EEG analysis toolbox MNE. It is an open-source, platform-independent, interactive web application, which supports common EEG-file formats to facilitate easy integration with other EEG toolboxes. RV includes common features of other EEG viewers, e.g., a view-slider, tools for marking bad channels and transient artifacts, and customizable preprocessing. Altogether, RV is an EEG viewer that combines the predictive power of deep-learning models and the knowledge of scientists and clinicians to optimize EEG annotation. With the training of new deep-learning models, RV could be developed to detect clinical patterns other than artifacts, for example sleep stages and EEG abnormalities.

Measurement of excitation-inhibition ratio in autism spectrum disorder using critical brain dynamics.

Balance between excitation (E) and inhibition (I) is a key principle for neuronal network organization and information processing. Consistent with this notion, excitation-inhibition imbalances are considered a pathophysiological mechanism in many brain disorders including autism spectrum disorder (ASD). However, methods to measure E/I ratios in human brain networks are lacking. Here, we present a method to quantify a functional E/I ratio (fE/I) from neuronal oscillations, and validate it in healthy subjects and children with ASD. We define structural E/I ratio in an in silico neuronal network, investigate how it relates to power and long-range temporal correlations (LRTC) of the network's activity, and use these relationships to design the fE/I algorithm. Application of this algorithm to the EEGs of healthy adults showed that fE/I is balanced at the population level and is decreased through GABAergic enforcement. In children with ASD, we observed larger fE/I variability and stronger LRTC compared to typically developing children (TDC). Interestingly, visual grading for EEG abnormalities that are thought to reflect E/I imbalances revealed elevated fE/I and LRTC in ASD children with normal EEG compared to TDC or ASD with abnormal EEG. We speculate that our approach will help understand physiological heterogeneity also in other brain disorders.

Virtual localization of the seizure onset zone: Using non-invasive MEG virtual electrodes at stereo-EEG electrode locations in refractory epilepsy patients.

In some patients with medically refractory epilepsy, EEG with intracerebrally placed electrodes (stereo-electroencephalography, SEEG) is needed to locate the seizure onset zone (SOZ) for successful epilepsy surgery. SEEG has limitations and entails risk of complications because of its invasive character. Non-invasive magnetoencephalography virtual electrodes (MEG-VEs) may overcome SEEG limitations and optimize electrode placement making SOZ localization safer. Our purpose was to assess whether interictal activity measured by MEG-VEs and SEEG at identical anatomical locations were comparable, and whether MEG-VEs activity properties could determine the location of a later resected brain area (RA) as an approximation of the SOZ. We analyzed data from nine patients who underwent MEG and SEEG evaluation, and surgery for medically refractory epilepsy. MEG activity was retrospectively reconstructed using beamforming to obtain VEs at the anatomical locations corresponding to those of SEEG electrodes. Spectral, functional connectivity and functional network properties were obtained for both, MEG-VEs and SEEG time series, and their correlation and reliability were established. Based on these properties, the approximation of the SOZ was characterized by the differences between RA and non-RA (NRA). We found significant positive correlation and reliability between MEG-VEs and SEEG spectral measures (particularly in delta [0.5-4 Hz], alpha2 [10-13 Hz], and beta [13-30 Hz] bands) and broadband functional connectivity. Both modalities showed significantly slower activity and a tendency towards increased broadband functional connectivity in the RA compared to the NRA. Our findings show that spectral and functional connectivity properties of non-invasively obtained MEG-VEs match those of invasive SEEG recordings, and can characterize the SOZ. This suggests that MEG-VEs might be used for optimal SEEG planning and fewer depth electrode implantations, making the localization of the SOZ safer and more successful.