Overcoming treatment-resistant depression with machine-learning based tools: a study protocol combining EEG and clinical data to personalize glutamatergic and brain stimulation interventions (SelecTool Project).

Treatment-Resistant Depression (TRD) poses a substantial health and economic challenge, persisting as a major concern despite decades of extensive research into novel treatment modalities. The considerable heterogeneity in TRD's clinical manifestations and neurobiological bases has complicated efforts toward effective interventions. Recognizing the need for precise biomarkers to guide treatment choices in TRD, herein we introduce the SelecTool Project. This initiative focuses on developing (WorkPlane 1/WP1) and conducting preliminary validation (WorkPlane 2/WP2) of a computational tool (SelecTool) that integrates clinical data, neurophysiological (EEG) and peripheral (blood sample) biomarkers through a machine-learning framework designed to optimize TRD treatment protocols. The SelecTool project aims to enhance clinical decision-making by enabling the selection of personalized interventions. It leverages multi-modal data analysis to navigate treatment choices towards two validated therapeutic options for TRD: esketamine nasal spray (ESK-NS) and accelerated repetitive Transcranial Magnetic Stimulation (arTMS). In WP1, 100 subjects with TRD will be randomized to receive either ESK-NS or arTMS, with comprehensive evaluations encompassing neurophysiological (EEG), clinical (psychometric scales), and peripheral (blood samples) assessments both at baseline (T0) and one month post-treatment initiation (T1). WP2 will utilize the data collected in WP1 to train the SelecTool algorithm, followed by its application in a second, out-of-sample cohort of 20 TRD subjects, assigning treatments based on the tool's recommendations. Ultimately, this research seeks to revolutionize the treatment of TRD by employing advanced machine learning strategies and thorough data analysis, aimed at unraveling the complex neurobiological landscape of depression. This effort is expected to provide pivotal insights that will promote the development of more effective and individually tailored treatment strategies, thus addressing a significant void in current TRD management and potentially reducing its profound societal and economic burdens.

Exploring Motor Network Connectivity in State-Dependent Transcranial Magnetic Stimulation: A Proof-of-Concept Study.

State-dependent non-invasive brain stimulation (NIBS) informed by electroencephalography (EEG) has contributed to the understanding of NIBS inter-subject and inter-session variability. While these approaches focus on local EEG characteristics, it is acknowledged that the brain exhibits an intrinsic long-range dynamic organization in networks. This proof-of-concept study explores whether EEG connectivity of the primary motor cortex (M1) in the pre-stimulation period aligns with the Motor Network (MN) and how the MN state affects responses to the transcranial magnetic stimulation (TMS) of M1. One thousand suprathreshold TMS pulses were delivered to the left M1 in eight subjects at rest, with simultaneous EEG. Motor-evoked potentials (MEPs) were measured from the right hand. The source space functional connectivity of the left M1 to the whole brain was assessed using the imaginary part of the phase locking value at the frequency of the sensorimotor µ-rhythm in a 1 s window before the pulse. Group-level connectivity revealed functional links between the left M1, left supplementary motor area, and right M1. Also, pulses delivered at high MN connectivity states result in a greater MEP amplitude compared to low connectivity states. At the single-subject level, this relation is more highly expressed in subjects that feature an overall high cortico-spinal excitability. In conclusion, this study paves the way for MN connectivity-based NIBS.

A bicoherence approach to analyze multi-dimensional cross-frequency coupling in EEG/MEG data.

We introduce a blockwise generalisation of the Antisymmetric Cross-Bicoherence (ACB), a statistical method based on bispectral analysis. The Multi-dimensional ACB (MACB) is an approach that aims at detecting quadratic lagged phase-interactions between vector time series in the frequency domain. Such a coupling can be empirically observed in functional neuroimaging data, e.g., in electro/magnetoencephalographic signals. MACB is invariant under orthogonal trasformations of the data, which makes it independent, e.g., on the choice of the physical coordinate system in the neuro-electromagnetic inverse procedure. In extensive synthetic experiments, we prove that MACB performance is significantly better than that obtained by ACB. Specifically, the shorter the data length, or the higher the dimension of the single data space, the larger the difference between the two methods.

Mindfulness meditation styles differently modulate source-level MEG microstate dynamics and complexity.

Background: The investigation of mindfulness meditation practice, classically divided into focused attention meditation (FAM), and open monitoring meditation (OMM) styles, has seen a long tradition of theoretical, affective, neurophysiological and clinical studies. In particular, the high temporal resolution of magnetoencephalography (MEG) or electroencephalography (EEG) has been exploited to fill the gap between the personal experience of meditation practice and its neural correlates. Mounting evidence, in fact, shows that human brain activity is highly dynamic, transiting between different brain states (microstates). In this study, we aimed at exploring MEG microstates at source-level during FAM, OMM and in the resting state, as well as the complexity and criticality of dynamic transitions between microstates. Methods: Ten right-handed Theravada Buddhist monks with a meditative expertise of minimum 2,265 h participated in the experiment. MEG data were acquired during a randomized block design task (6 min FAM, 6 min OMM, with each meditative block preceded and followed by 3 min resting state). Source reconstruction was performed using eLORETA on individual cortical space, and then parcellated according to the Human Connect Project atlas. Microstate analysis was then applied to parcel level signals in order to derive microstate topographies and indices. In addition, from microstate sequences, the Hurst exponent and the Lempel-Ziv complexity (LZC) were computed. Results: Our results show that the coverage and occurrence of specific microstates are modulated either by being in a meditative state or by performing a specific meditation style. Hurst exponent values in both meditation conditions are reduced with respect to the value observed during rest, LZC shows significant differences between OMM, FAM, and REST, with a progressive increase from REST to FAM to OMM. Discussion: Importantly, we report changes in brain criticality indices during meditation and between meditation styles, in line with a state-like effect of meditation on cognitive performance. In line with previous reports, we suggest that the change in cognitive state experienced in meditation is paralleled by a shift with respect to critical points in brain dynamics.

Targeting motor cortex high-excitability states defined by functional connectivity with real-time EEG-TMS.

We tested previous post-hoc findings indicating a relationship between functional connectivity (FC) in the motor network and corticospinal excitability (CsE), in a real-time EEG-TMS experiment in healthy participants. We hypothesized that high FC between left and right motor cortex predicts high CsE. FC was quantified in real-time by single-trial phase-locking value (stPLV), and TMS single pulses were delivered based on the current FC. CsE was indexed by motor-evoked potential (MEP) amplitude in a hand muscle. Possible confounding factors (pre-stimulus µ-power and phase, interstimulus interval) were evaluated post hoc. MEPs were significantly larger during high FC compared to low FC. Post hoc analysis revealed that the FC condition showed a significant interaction with µ-power in the stimulated hemisphere. Further, inter-stimulus interval (ISI) interacted with high vs. low FC conditions. In summary, FC was confirmed to be predictive of CsE, but should not be considered in isolation from µ-power and ISI. Moreover, FC was complementary to µ-phase in predicting CsE. Motor network FC is another marker of real-time accessible CsE beyond previously established markers, in particular phase and power of the µ rhythm, and may help define a more robust composite biomarker of high/low excitability states of human motor cortex.

Predicting outcome with Intranasal Esketamine treatment: A machine-learning, three-month study in Treatment-Resistant Depression (ESK-LEARNING).

Treatment-resistant depression (TRD) represents a severe clinical condition with high social and economic costs. Esketamine Nasal Spray (ESK-NS) has recently been approved for TRD by EMA and FDA, but data about predictors of response are still lacking. Thus, a tool that can predict the individual patients' probability of response to ESK-NS is needed. This study investigates sociodemographic and clinical features predicting responses to ESK-NS in TRD patients using machine learning techniques. In a retrospective, multicentric, real-world study involving 149 TRD subjects, psychometric data (Montgomery-Asberg-Depression-Rating-Scale/MADRS, Brief-Psychiatric-Rating-Scale/BPRS, Hamilton-Anxiety-Rating-Scale/HAM-A, Hamilton-Depression-Rating-Scale/HAMD-17) were collected at baseline and at one month/T1 and three months/T2 post-treatment initiation. We trained three different random forest classifiers, able to predict responses to ESK-NS with accuracies of 68.53% at T1 and 66.26% at T2 and remission at T2 with 68.60% of accuracy. Features like severe anhedonia, anxious distress, mixed symptoms as well as bipolarity were found to positively predict response and remission. At the same time, benzodiazepine usage and depression severity were linked to delayed responses. Despite some limitations (i.e., retrospective study, lack of biomarkers, lack of a correct interrater-reliability across the different centers), these findings suggest the potential of machine learning in personalized intervention for TRD.

Long-Term and Meditation-Specific Modulations of Brain Connectivity Revealed Through Multivariate Pattern Analysis.

Neuroimaging studies have provided evidence that extensive meditation practice modifies the functional and structural properties of the human brain, such as large-scale brain region interplay. However, it remains unclear how different meditation styles are involved in the modulation of these large-scale brain networks. Here, using machine learning and fMRI functional connectivity, we investigated how focused attention and open monitoring meditation styles impact large-scale brain networks. Specifically, we trained a classifier to predict the meditation style in two groups of subjects: expert Theravada Buddhist monks and novice meditators. We showed that the classifier was able to discriminate the meditation style only in the expert group. Additionally, by inspecting the trained classifier, we observed that the Anterior Salience and the Default Mode networks were relevant for the classification, in line with their theorized involvement in emotion and self-related regulation in meditation. Interestingly, results also highlighted the role of specific couplings between areas crucial for regulating attention and self-awareness as well as areas related to processing and integrating somatosensory information. Finally, we observed a larger involvement of left inter-hemispheric connections in the classification. In conclusion, our work supports the evidence that extensive meditation practice modulates large-scale brain networks, and that the different meditation styles differentially affect connections that subserve style-specific functions.

TMS-Induced Modulation of EEG Functional Connectivity Is Affected by the E-Field Orientation.

Coregistration of transcranial magnetic stimulation (TMS) and electroencephalography (EEG) allows non-invasive probing of brain circuits: TMS induces brain activation due to the generation of a properly oriented focused electric field (E-field) using a coil placed on a selected position over the scalp, while EEG captures the effects of the stimulation on brain electrical activity. Moreover, the combination of these techniques allows the investigation of several brain properties, including brain functional connectivity. The choice of E-field parameters, such as intensity, orientation, and position, is crucial for eliciting cortex-specific effects. Here, we evaluated whether and how the spatial pattern, i.e., topography and strength of functional connectivity, is modulated by the stimulus orientation. We systematically altered the E-field orientation when stimulating the left pre-supplementary motor area and showed an increase of functional connectivity in areas associated with the primary motor cortex and an E-field orientation-specific modulation of functional connectivity intensity.

Effective Intracerebral Connectivity in Acute Stroke: A TMS-EEG Study.

Stroke is a major cause of disability because of its motor and cognitive sequelae even when the acute phase of stabilization of vital parameters is overcome. The most important improvements occur in the first 8-12 weeks after stroke, indicating that it is crucial to improve our understanding of the dynamics of phenomena occurring in this time window to prospectively target rehabilitation procedures from the earliest stages after the event. Here, we studied the intracortical excitability properties of delivering transcranial magnetic stimulation (TMS) to the primary motor cortex (M1) of left and right hemispheres in 17 stroke patients who suffered a mono-lateral left hemispheric stroke, excluding pure cortical damage. All patients were studied within 10 days of symptom onset. TMS-evoked potentials (TEPs) were collected via a TMS-compatible electroencephalogram system (TMS-EEG) concurrently with motor-evoked responses (MEPs) induced in the contralateral first dorsal interosseous muscle. Comparison with age-matched healthy volunteers was made by collecting the same bilateral-stimulation data in nine healthy volunteers as controls. Excitability in the acute phase revealed relevant changes in the relationship between left lesioned and contralesionally right hemispheric homologous areas both for TEPs and MEPs. While the paretic hand displayed reduced MEPs compared to the non-paretic hand and to healthy volunteers, TEPs revealed an overexcitable lesioned hemisphere with respect to both healthy volunteers and the contra-lesion side. Our quantitative results advance the understanding of the impairment of intracortical inhibitory networks. The neuronal dysfunction most probably changes the excitatory/inhibitory on-center off-surround organization that supports already acquired learning and reorganization phenomena that support recovery from stroke sequelae.

Magnetoencephalographic spectral fingerprints differentiate evidence accumulation from saccadic motor preparation in perceptual decision-making.

The understanding of the neurobiological basis of perceptual decision-making has been profoundly shaped by studies in the monkey brain in tandem with mathematical models, providing the basis for the formulation of an intentional account of decision-making. Although much progress has been made in human studies, a characterization of the neural underpinnings of an integrative mechanism, where evidence accumulation and the selection and execution of responses are carried out by the same system, remains challenging. Here, by employing magnetoencephalographic recording in combination with an experimental protocol that measures saccadic response and leverages a systematic modulation of evidence levels, we obtained a spectral dissociation between evidence accumulation mechanisms and motor preparation within the same brain region. Specifically, we show that within the dorsomedial parietal cortex alpha power modulation reflects the amount of sensory evidence available while beta power modulations reflect motor preparation, putatively representing the human homolog of the saccadic-related LIP region.

Source-based artifact-rejection techniques for TMS-EEG.

Neuronal electroencephalography (EEG) signals arise from the cortical postsynaptic currents. Due to the conductive properties of the head, these neuronal sources produce relatively smeared spatial patterns in EEG. We can model these topographies to deduce which signals reflect genuine TMS-evoked cortical activity and which data components are merely noise and artifacts. This review will concentrate on two source-based artifact-rejection techniques developed for TMS-EEG data analysis, signal-space-projection-source-informed reconstruction (SSP-SIR), and the source-estimate-utilizing noise-discarding algorithm (SOUND). The former method was designed for rejecting TMS-evoked muscle artifacts, while the latter was developed to suppress noise signals from EEG and magnetoencephalography (MEG) in general. We shall cover the theoretical background for both methods, but most importantly, we will describe some essential practical perspectives for using these techniques effectively. We demonstrate and explain what approaches produce the most reliable inverse estimates after cleaning the data or how to perform non-biased comparisons between cleaned datasets. All noise-cleaning algorithms compromise the signals of interest to a degree. We elaborate on how the source-based methods allow objective quantification of the overcorrection. Finally, we consider possible future directions. While this article concentrates on TMS-EEG data analysis, many theoretical and practical aspects, presented here, can be readily applied in other EEG/MEG applications. Overall, the source-based cleaning methods provide a valuable set of TMS-EEG preprocessing tools. We can objectively evaluate their performance regarding possible overcorrection. Furthermore, the overcorrection can always be taken into account to compare cleaned datasets reliably. The described methods are based on current electrophysiological and anatomical understanding of the head and the EEG generators; strong assumptions of the statistical properties of the noise and artifact signals, such as independence, are not needed.

Looking through the windows: a study about the dependency of phase-coupling estimates on the data length.

Objective. Being able to characterize functional connectivity (FC) state dynamics in a real-time setting, such as in brain-computer interface, neurofeedback or closed-loop neurostimulation frameworks, requires the rapid detection of the statistical dependencies that quantify FC in short windows of data. The aim of this study is to characterize, through extensive realistic simulations, the reliability of FC estimation as a function of the data length. In particular, we focused on FC as measured by phase-coupling (PC) of neuronal oscillations, one of the most functionally relevant neural coupling modes.Approach. We generated synthetic data corresponding to different scenarios by varying the data length, the signal-to-noise ratio (SNR), the phase difference value, the spectral analysis approach (Hilbert or Fourier) and the fractional bandwidth. We compared seven PC metrics, i.e. imaginary part of phase locking value (iPLV), PLV of orthogonalized signals, phase lag index (PLI), debiased weighted PLI, imaginary part of coherency, coherence of orthogonalized signals and lagged coherence.Main results. Our findings show that, for a SNR of at least 10 dB, a data window that contains 5-8 cycles of the oscillation of interest (e.g. a 500-800 ms window at 10 Hz) is generally required to achieve reliable PC estimates. In general, Hilbert-based approaches were associated with higher performance than Fourier-based approaches. Furthermore, the results suggest that, when the analysis is performed in a narrow frequency range, a larger window is required.Significance. The achieved results pave the way to the introduction of best-practice guidelines to be followed when a real-time frequency-specific PC assessment is at target.

Looking through the windows: a study about the dependency of phase-coupling estimates on the data length.

OBJECTIVE: Being able to characterize functional connectivity (FC) state dynamics in a real-time setting, such as in brain-computer interface, neurofeedback or closed-loop neurostimulation frameworks, requires the rapid detection of the statistical dependencies that quantify FC in short windows of data. The aim of this study is to characterize, through extensive realistic simulations, the reliability of FC estimation as a function of the data length. In particular, we focused on FC as measured by phase-coupling (PC) of neuronal oscillations, one of the most functionally relevant neural coupling modes. APPROACH: We generated synthetic data corresponding to different scenarios by varying the data length, the signal-to-noise ratio, the phase difference value, the spectral analysis approach (Hilbert or Fourier) and the fractional bandwidth. We compared seven PC metrics, i.e. imaginary part of phase locking value (PLV), PLV of orthogonalized signals, phase lag index (PLI), debiased weighted PLI, imaginary part of coherency, coherence of orthogonalized signals and lagged coherence. MAIN RESULTS: Our findings show that, for a signal-to-noise-ratio of at least 10 dB, a data window that contains 5 to 8 cycles of the oscillation of interest (e.g. a 500-800ms window at 10Hz) is generally required to achieve reliable PC estimates. In general, Hilbert-based approaches were associated with higher performance than Fourier-based approaches. Furthermore, the results suggest that, when the analysis is performed in a narrow frequency range, a larger window is required. SIGNIFICANCE: The achieved results pave the way to the introduction of best-practice guidelines to be followed when a real-time frequency-specific PC assessment is at target.

Phase-coupling of neural oscillations contributes to individual differences in peripersonal space.

The peripersonal space (PPS) is a multisensory and sensorimotor interface between our body and the environment. The location of PPS boundary is not fixed. Rather, it adapts to the environmental context and differs greatly across individuals. Recent studies have started to unveil the neural correlates of individual differences in PPS extension; however, this picture is not clear yet. Here, we used approaching auditory stimuli and magnetoencephalography to capture the individual boundary of PPS and examine its neural underpinnings. In particular, building upon previous studies from our own group, we investigated the possible contribution of an intrinsic feature of the brain, that is the "resting state" functional connectivity, to the individual differences in PPS extension and the frequency specificity of this contribution. Specifically, we focused on the activity synchronized to the premotor cortex, where multisensory neurons encoding PPS have been described. Results showed that the stronger the connectivity between left premotor cortex (lPM) and a set of fronto-parietal, sensorimotor regions in the right and left hemisphere, the wider the extension of the PPS. Strikingly, such a correlation was observed only in the beta-frequency band. Overall, our results suggest that the individual extension of the PPS is coded in spatially- and spectrally-specific resting state functional links.

Decoding working memory task condition using magnetoencephalography source level long-range phase coupling patterns.

OBJECTIVE: The objective of the study is to identify phase coupling patterns that are shared across subjects via a machine learning approach that utilises source space magnetoencephalography (MEG) phase coupling data from a working memory (WM) task. Indeed, phase coupling of neural oscillations is putatively a key factor for communication between distant brain areas and is therefore crucial in performing cognitive tasks, including WM. Previous studies investigating phase coupling during cognitive tasks have often focused on a few a priori selected brain areas or a specific frequency band, and the need for data-driven approaches has been recognised. Machine learning techniques have emerged as valuable tools for the analysis of neuroimaging data since they catch fine-grained differences in the multivariate signal distribution. Here, we expect that these techniques applied to MEG phase couplings can reveal WM-related processes that are shared across individuals. APPROACH: We analysed WM data collected as part of the Human Connectome Project. The MEG data were collected while subjects (n = 83) performed N-back WM tasks in two different conditions, namely 2-back (WM condition) and 0-back (control condition). We estimated phase coupling patterns (multivariate phase slope index) for both conditions and for theta, alpha, beta, and gamma bands. The obtained phase coupling data were then used to train a linear support vector machine in order to classify which task condition the subject was performing with an across-subject cross-validation approach. The classification was performed separately based on the data from individual frequency bands and with all bands combined (multiband). Finally, we evaluated the relative importance of the different features (phase couplings) for classification by the means of feature selection probability. MAIN RESULTS: The WM condition and control condition were successfully classified based on the phase coupling patterns in the theta (62% accuracy) and alpha bands (60% accuracy) separately. Importantly, the multiband classification showed that phase coupling patterns not only in the theta and alpha but also in the gamma bands are related to WM processing, as testified by improvement in classification performance (71%). SIGNIFICANCE: Our study successfully decoded WM tasks using MEG source space functional connectivity. Our approach, combining across-subject classification and a multidimensional metric recently developed by our group, is able to detect patterns of connectivity that are shared across individuals. In other words, the results are generalisable to new individuals and allow meaningful interpretation of task-relevant phase coupling patterns.

Multi-dimensional connectivity: a conceptual and mathematical review.

The estimation of functional connectivity between regions of the brain, for example based on statistical dependencies between the time series of activity in each region, has become increasingly important in neuroimaging. Typically, multiple time series (e.g. from each voxel in fMRI data) are first reduced to a single time series that summarises the activity in a region of interest, e.g. by averaging across voxels or by taking the first principal component; an approach we call one-dimensional connectivity. However, this summary approach ignores potential multi-dimensional connectivity between two regions, and a number of recent methods have been proposed to capture such complex dependencies. Here we review the most common multi-dimensional connectivity methods, from an intuitive perspective, from a formal (mathematical) point of view, and through a number of simulated and real (fMRI and MEG) data examples that illustrate the strengths and weaknesses of each method. The paper is accompanied with both functions and scripts, which implement each method and reproduce all the examples.