Interpretable deep learning for deconvolutional analysis of neural signals.

The widespread adoption of deep learning to build models that capture the dynamics of neural populations is typically based on "black-box" approaches that lack an interpretable link between neural activity and function. Here, we propose to apply algorithm unrolling, a method for interpretable deep learning, to design the architecture of sparse deconvolutional neural networks and obtain a direct interpretation of network weights in relation to stimulus-driven single-neuron activity through a generative model. We characterize our method, referred to as deconvolutional unrolled neural learning (DUNL), and show its versatility by applying it to deconvolve single-trial local signals across multiple brain areas and recording modalities. To exemplify use cases of our decomposition method, we uncover multiplexed salience and reward prediction error signals from midbrain dopamine neurons in an unbiased manner, perform simultaneous event detection and characterization in somatosensory thalamus recordings, and characterize the responses of neurons in the piriform cortex. Our work leverages the advances in interpretable deep learning to gain a mechanistic understanding of neural dynamics.

A Randomized Controlled Pilot Trial Comparing Effects of Qigong and Exercise/Nutrition Training on Fatigue and Other Outcomes in Female Cancer Survivors.

Cancer-related fatigue (CRF) is a common and burdensome, often long-term side effect of cancer and its treatment. Many non-pharmacological treatments have been investigated as possible CRF therapies, including exercise, nutrition, health/psycho-education, and mind-body therapies. However, studies directly comparing the efficacy of these treatments in randomized controlled trials are lacking. To fill this gap, we conducted a parallel single blind randomized controlled pilot efficacy trial with women with CRF to directly compare the effects of Qigong (a form of mind-body intervention) (n = 11) to an intervention that combined strength and aerobic exercise, plant-based nutrition and health/psycho-education (n = 13) in a per protocol analysis. This design was chosen to determine the comparative efficacy of 2 non-pharmacologic interventions, with different physical demand intensities, in reducing the primary outcome measure of self-reported fatigue (FACIT "Additional Concerns" subscale). Both interventions showed a mean fatigue improvement of more than double the pre-established minimal clinically important difference of 3 (qigong: 7.068 ± 10.30, exercise/nutrition: 8.846 ± 12.001). Mixed effects ANOVA analysis of group × time interactions revealed a significant main effect of time, such that both groups significantly improved fatigue from pre- to post-treatment (F(1,22) = 11.898, P = .002, generalized eta squared effect size = 0.116) There was no significant difference between fatigue improvement between groups (independent samples t-test: P = .70 ), suggesting a potential equivalence or non-inferiority of interventions, which we could not definitively establish due to our small sample size. This study provides evidence from a small sample of n = 24 women with CRF that qigong improves fatigue similarly to exercise-nutrition courses. Qigong additionally significantly improved secondary measures of mood, emotion regulation, and stress, while exercise/nutrition significantly improved secondary measures of sleep/fatigue. These findings provide preliminary evidence for divergent mechanisms of fatigue improvement across interventions, with qigong providing a gentler and lower-intensity alternative to exercise/nutrition.

Fronto-central resting-state 15-29 Hz transient beta events change with therapeutic transcranial magnetic stimulation for posttraumatic stress disorder and major depressive disorder.

Repetitive transcranial magnetic stimulation (rTMS) is an established treatment for major depressive disorder (MDD) and shows promise for posttraumatic stress disorder (PTSD), yet effectiveness varies. Electroencephalography (EEG) can identify rTMS-associated brain changes. EEG oscillations are often examined using averaging approaches that mask finer time-scale dynamics. Recent advances show some brain oscillations emerge as transient increases in power, a phenomenon termed "Spectral Events," and that event characteristics correspond with cognitive functions. We applied Spectral Event analyses to identify potential EEG biomarkers of effective rTMS treatment. Resting 8-electrode EEG was collected from 23 patients with MDD and PTSD before and after 5 Hz rTMS targeting the left dorsolateral prefrontal cortex. Using an open-source toolbox ( https://github.com/jonescompneurolab/SpectralEvents ), we quantified event features and tested for treatment associated changes. Spectral Events in delta/theta (1-6 Hz), alpha (7-14 Hz), and beta (15-29 Hz) bands occurred in all patients. rTMS-induced improvement in comorbid MDD PTSD were associated with pre- to post-treatment changes in fronto-central electrode beta event features, including frontal beta event frequency spans and durations, and central beta event maxima power. Furthermore, frontal pre-treatment beta event duration correlated negatively with MDD symptom improvement. Beta events may provide new biomarkers of clinical response and advance the understanding of rTMS.

Fronto-central resting-state 15-29Hz transient beta events change with therapeutic transcranial magnetic stimulation for posttraumatic stress disorder and major depressive disorder.

Repetitive transcranial magnetic stimulation (rTMS) is an established treatment for major depressive disorder (MDD) and shows promise for posttraumatic stress disorder (PTSD), yet effectiveness varies. Electroencephalography (EEG) can identify rTMS-associated brain changes. EEG oscillations are often examined using averaging approaches that mask finer time-scale dynamics. Recent advances show some brain oscillations emerge as transient increases in power, a phenomenon termed "Spectral Events," and that event characteristics correspond with cognitive functions. We applied Spectral Event analyses to identify potential EEG biomarkers of effective rTMS treatment. Resting 8-electrode EEG was collected from 23 patients with MDD and PTSD before and after 5Hz rTMS targeting the left dorsolateral prefrontal cortex. Using an open-source toolbox ( https://github.com/jonescompneurolab/SpectralEvents ), we quantified event features and tested for treatment associated changes. Spectral Events in delta/theta (1-6 Hz), alpha (7-14 Hz), and beta (15-29 Hz) bands occurred in all patients. rTMS-induced improvement in comorbid MDD PTSD were associated with pre-to post-treatment changes in fronto-central electrode beta event features, including frontal beta event frequency spans and durations, and central beta event maxima power. Furthermore, frontal pre-treatment beta event duration correlated negatively with MDD symptom improvement. Beta events may provide new biomarkers of clinical response and advance the understanding of rTMS.

Corrigendum: Mechanical affective touch therapy for anxiety disorders: Feasibility, clinical outcomes, and electroencephalography biomarkers from an open-label trial.

[This corrects the article DOI: 10.3389/fpsyt.2022.877574.].

Mechanical Affective Touch Therapy for Anxiety Disorders: Feasibility, Clinical Outcomes, and Electroencephalography Biomarkers From an Open-Label Trial.

Background: Most external peripheral nerve stimulation devices designed to alter mood states use electrical energy, but mechanical stimulation for activation of somatosensory pathways may be harnessed for potential therapeutic neuromodulation. A novel investigational device for Mechanical Affective Touch Therapy (MATT) was created to stimulate C-tactile fibers through gentle vibrations delivered by piezoelectric actuators on the bilateral mastoid processes. Methods: 22 adults with anxiety disorders and at least moderate anxiety symptom severity enrolled in an open-label pilot trial that involved MATT self-administration using a simple headset at home at least twice per day for 4 weeks. Resting EEG data were acquired before and after a baseline MATT session and again before the final MATT session. Self-report measures of mood and anxiety were collected at baseline, week 2, and week 4, while interoception was assessed pre- and post-treatment. Results: Anxiety and depressive symptoms improved significantly from baseline to endpoint, and mindfulness was enhanced. EEG metrics confirmed an association between acute MATT stimulation and oscillatory power in alpha and theta bands; symptom changes correlated with changes in some metrics. Conclusion: Open-label data suggest MATT is a promising non-invasive therapeutic approach to anxiety disorders that warrants further development.

A preliminary investigation of childhood anxiety/depressive symptomatology and working memory across multiple units of analysis.

This exploratory study examined multiple units of working memory (WM) analysis in a transdiagnostic, treatment-seeking, pediatric sample. This included a) an electroencephalography marker of WM (coupling of theta and gamma oscillations [i.e., theta-gamma coupling] in frontal brain regions), b) WM test performance, and c) parent-reported WM symptoms. A composite score combining each of these units of analysis correlated with self-reported depressive and anxiety symptoms, with only theta-gamma coupling independently predicted anxiety/depressive symptoms. Results confirm prior findings on the association between WM and anxiety/depression, although the majority of this variance was explained by frontal theta-gamma coupling during WM demands.

Left-Lateralized Contributions of Saccades to Cortical Activity During a One-Back Word Recognition Task.

Saccadic eye movements are an inherent component of natural reading, yet their contribution to information processing at subsequent fixation remains elusive. Here we use anatomically-constrained magnetoencephalography (MEG) to examine cortical activity following saccades as healthy human subjects engaged in a one-back word recognition task. This activity was compared with activity following external visual stimulation that mimicked saccades. A combination of procedures was employed to eliminate saccadic ocular artifacts from the MEG signal. Both saccades and saccade-like external visual stimulation produced early-latency responses beginning ~70 ms after onset in occipital cortex and spreading through the ventral and dorsal visual streams to temporal, parietal and frontal cortices. Robust differential activity following the onset of saccades vs. similar external visual stimulation emerged during 150-350 ms in a left-lateralized cortical network. This network included: (i) left lateral occipitotemporal (LOT) and nearby inferotemporal (IT) cortex; (ii) left posterior Sylvian fissure (PSF) and nearby multimodal cortex; and (iii) medial parietooccipital (PO), posterior cingulate and retrosplenial cortices. Moreover, this left-lateralized network colocalized with word repetition priming effects. Together, results suggest that central saccadic mechanisms influence a left-lateralized language network in occipitotemporal and temporal cortex above and beyond saccadic influences at preceding stages of information processing during visual word recognition.

Measuring the signal-to-noise ratio of a neuron.

The signal-to-noise ratio (SNR), a commonly used measure of fidelity in physical systems, is defined as the ratio of the squared amplitude or variance of a signal relative to the variance of the noise. This definition is not appropriate for neural systems in which spiking activity is more accurately represented as point processes. We show that the SNR estimates a ratio of expected prediction errors and extend the standard definition to one appropriate for single neurons by representing neural spiking activity using point process generalized linear models (PP-GLM). We estimate the prediction errors using the residual deviances from the PP-GLM fits. Because the deviance is an approximate χ(2) random variable, we compute a bias-corrected SNR estimate appropriate for single-neuron analysis and use the bootstrap to assess its uncertainty. In the analyses of four systems neuroscience experiments, we show that the SNRs are -10 dB to -3 dB for guinea pig auditory cortex neurons, -18 dB to -7 dB for rat thalamic neurons, -28 dB to -14 dB for monkey hippocampal neurons, and -29 dB to -20 dB for human subthalamic neurons. The new SNR definition makes explicit in the measure commonly used for physical systems the often-quoted observation that single neurons have low SNRs. The neuron's spiking history is frequently a more informative covariate for predicting spiking propensity than the applied stimulus. Our new SNR definition extends to any GLM system in which the factors modulating the response can be expressed as separate components of a likelihood function.

Algorithms for the analysis of ensemble neural spiking activity using simultaneous-event multivariate point-process models.

Understanding how ensembles of neurons represent and transmit information in the patterns of their joint spiking activity is a fundamental question in computational neuroscience. At present, analyses of spiking activity from neuronal ensembles are limited because multivariate point process (MPP) models cannot represent simultaneous occurrences of spike events at an arbitrarily small time resolution. Solo recently reported a simultaneous-event multivariate point process (SEMPP) model to correct this key limitation. In this paper, we show how Solo's discrete-time formulation of the SEMPP model can be efficiently fit to ensemble neural spiking activity using a multinomial generalized linear model (mGLM). Unlike existing approximate procedures for fitting the discrete-time SEMPP model, the mGLM is an exact algorithm. The MPP time-rescaling theorem can be used to assess model goodness-of-fit. We also derive a new marked point-process (MkPP) representation of the SEMPP model that leads to new thinning and time-rescaling algorithms for simulating an SEMPP stochastic process. These algorithms are much simpler than multivariate extensions of algorithms for simulating a univariate point process, and could not be arrived at without the MkPP representation. We illustrate the versatility of the SEMPP model by analyzing neural spiking activity from pairs of simultaneously-recorded rat thalamic neurons stimulated by periodic whisker deflections, and by simulating SEMPP data. In the data analysis example, the SEMPP model demonstrates that whisker motion significantly modulates simultaneous spiking activity at the 1 ms time scale and that the stimulus effect is more than one order of magnitude greater for simultaneous activity compared with non-simultaneous activity. Together, the mGLM, the MPP time-rescaling theorem and the MkPP representation of the SEMPP model offer a theoretically sound, practical tool for measuring joint spiking propensity in a neuronal ensemble.

Eye movements modulate the spatiotemporal dynamics of word processing.

Active reading requires coordination between frequent eye movements (saccades) and short fixations in text. Yet, the impact of saccades on word processing remains unknown, as neuroimaging studies typically employ constant eye fixation. Here we investigate eye-movement effects on word recognition processes in healthy human subjects using anatomically constrained magnetoencephalography, psychophysical measurements, and saccade detection in real time. Word recognition was slower and brain responses were reduced to words presented early versus late after saccades, suggesting an overall transient impairment of word processing after eye movements. Response reductions occurred early in visual cortices and later in language regions, where they colocalized with repetition priming effects. Qualitatively similar effects occurred when words appeared early versus late after background movement that mimicked saccades, suggesting that retinal motion contributes to postsaccadic inhibition. Further, differences in postsaccadic and background-movement effects suggest that central mechanisms also contribute to postsaccadic modulation. Together, these results suggest a complex interplay between visual and central saccadic mechanisms during reading.

A spatiotemporal dynamic distributed solution to the MEG inverse problem.

MEG/EEG are non-invasive imaging techniques that record brain activity with high temporal resolution. However, estimation of brain source currents from surface recordings requires solving an ill-conditioned inverse problem. Converging lines of evidence in neuroscience, from neuronal network models to resting-state imaging and neurophysiology, suggest that cortical activation is a distributed spatiotemporal dynamic process, supported by both local and long-distance neuroanatomic connections. Because spatiotemporal dynamics of this kind are central to brain physiology, inverse solutions could be improved by incorporating models of these dynamics. In this article, we present a model for cortical activity based on nearest-neighbor autoregression that incorporates local spatiotemporal interactions between distributed sources in a manner consistent with neurophysiology and neuroanatomy. We develop a dynamic maximum a posteriori expectation-maximization (dMAP-EM) source localization algorithm for estimation of cortical sources and model parameters based on the Kalman Filter, the Fixed Interval Smoother, and the EM algorithms. We apply the dMAP-EM algorithm to simulated experiments as well as to human experimental data. Furthermore, we derive expressions to relate our dynamic estimation formulas to those of standard static models, and show how dynamic methods optimally assimilate past and future data. Our results establish the feasibility of spatiotemporal dynamic estimation in large-scale distributed source spaces with several thousand source locations and hundreds of sensors, with resulting inverse solutions that provide substantial performance improvements over static methods.

STATE-SPACE SOLUTIONS TO THE DYNAMIC MAGNETOENCEPHALOGRAPHY INVERSE PROBLEM USING HIGH PERFORMANCE COMPUTING.

Determining the magnitude and location of neural sources within the brain that are responsible for generating magnetoencephalography (MEG) signals measured on the surface of the head is a challenging problem in functional neuroimaging. The number of potential sources within the brain exceeds by an order of magnitude the number of recording sites. As a consequence, the estimates for the magnitude and location of the neural sources will be ill-conditioned because of the underdetermined nature of the problem. One well-known technique designed to address this imbalance is the minimum norm estimator (MNE). This approach imposes an L(2) regularization constraint that serves to stabilize and condition the source parameter estimates. However, these classes of regularizer are static in time and do not consider the temporal constraints inherent to the biophysics of the MEG experiment. In this paper we propose a dynamic state-space model that accounts for both spatial and temporal correlations within and across candidate intra-cortical sources. In our model, the observation model is derived from the steady-state solution to Maxwell's equations while the latent model representing neural dynamics is given by a random walk process. We show that the Kalman filter (KF) and the Kalman smoother [also known as the fixed-interval smoother (FIS)] may be used to solve the ensuing high-dimensional state-estimation problem. Using a well-known relationship between Bayesian estimation and Kalman filtering, we show that the MNE estimates carry a significant zero bias. Calculating these high-dimensional state estimates is a computationally challenging task that requires High Performance Computing (HPC) resources. To this end, we employ the NSF Teragrid Supercomputing Network to compute the source estimates. We demonstrate improvement in performance of the state-space algorithm relative to MNE in analyses of simulated and actual somatosensory MEG experiments. Our findings establish the benefits of high-dimensional state-space modeling as an effective means to solve the MEG source localization problem.

Rapid changes in thalamic firing synchrony during repetitive whisker stimulation.

Thalamic firing synchrony is thought to ensure selective transmission of relevant sensory information to the recipient cortical neurons by rendering them more responsive to temporally correlated input spikes. However, direct evidence for a synchrony code in the thalamus is limited. Here, we directly measure thalamic firing synchrony and its stimulus-induced modulation over time, using simultaneous single unit recordings from individual thalamic barreloids in the rat somatosensory whisker/barrel system. Employing whisker deflections varying in velocity or frequency and a cross-correlation approach, we find systematic changes in both time course and strength of thalamic firing synchrony as a function of stimulus parameters and sensory adaptation. Synchrony develops faster and is greater with higher velocity deflections. Greater firing synchrony reflects stimulus-dependent increases in instantaneous firing rates, greater spike time precision relative to stimulus onset as well as common input that likely arises from divergent trigeminothalamic and corticothalamic neurons. With adaptation, synchrony decreases and takes longer to develop but is more dependent on the cells' common inputs. Rapid, sharp increases in thalamic synchrony mirroring quick increases in whisker velocity occur also during ongoing random, high-frequency whisker vibrations. Together, results demonstrate millisecond by millisecond changes in thalamic near-synchronous firing during complex patterns of ongoing vibrissa movements that may ensure transmission of preferred sensory information in local thalamocortical circuits during whisking and active touch.

A non-invasive method to relate the timing of neural activity to white matter microstructural integrity.

The neurophysiological basis of variability in the latency of evoked neural responses has been of interest for decades. We describe a method to identify white matter pathways that may contribute to inter-individual variability in the timing of neural activity. We investigated the relation of the latency of peak visual responses in occipital cortex as measured by magnetoencephalography (MEG) to fractional anisotropy (FA) in the entire brain as measured with diffusion tensor imaging (DTI) in eight healthy young adults. This method makes no assumptions about the anatomy of white matter connections. Visual responses were evoked during a saccadic paradigm and were time-locked to arrival at a saccadic goal. The latency of the peak visual response was inversely related to FA in bilateral parietal and right lateral frontal white matter adjacent to cortical regions that modulate early visual responses. These relations suggest that biophysical properties of white matter affect the timing of early visual responses. This preliminary report demonstrates a non-invasive, unbiased method to relate the timing information from evoked-response experiments to the biophysical properties of white matter measured with DTI.

Functional topography of corticothalamic feedback enhances thalamic spatial response tuning in the somatosensory whisker/barrel system.

Corticothalamic (CT) projections are approximately 10 times more numerous than thalamocortical projections, yet their function in sensory processing is poorly understood. In particular, the functional significance of the topographic precision of CT feedback is unknown. We addressed these issues in the rodent somatosensory whisker/barrel system by deflecting individual whiskers and pharmacologically enhancing activity in layer VI of single whisker-related cortical columns. Enhancement of corticothalamic activity in a cortical column facilitated whisker-evoked responses in topographically aligned thalamic barreloid neurons, while activation of an adjacent column weakly suppressed activity at the same thalamic site. Both effects were more pronounced when stimulating the preferred, or principal, whisker than for adjacent whiskers. Thus, facilitation by homologous CT feedback sharpens thalamic receptive field focus, while suppression by nonhomologous feedback diminishes it. Our findings demonstrate that somatosensory cortex can selectively regulate thalamic spatial response tuning by engaging topographically specific excitatory and inhibitory mechanisms in the thalamus.

Processing of periodic whisker deflections by neurons in the ventroposterior medial and thalamic reticular nuclei.

Rats employ rhythmic whisker movements to sample information in their sensory environment. To study frequency tuning and filtering characteristics of thalamic circuitry, we recorded single-unit responses of ventroposterior medial (VPm) and thalamic reticular (Rt) neurons to 1- to 40-Hz sinusoidal and pulsatile whisker deflection in lightly narcotized rats. Neuronal entrainment was assessed by a measure of the relative modulation (RM) of firing at the stimulus frequency given by the first harmonic (F1) of the cycle time histogram divided by the mean firing rate (F0). VPm signaling of both sinusoidal and periodic pulsatile whisker movements improved gradually over 1-16 and was maximal at 20-40 Hz. By contrast, the RM of Rt responses increased over 1-8 Hz, but deteriorated progressively over the 12- to 40-Hz range. In Rt, response adaptation occurred at lower stimulus frequencies and to a greater extent than in VPm. Within a train of high-frequency stimuli, Rt responses progressively decremented, possibly due to the accumulation of inhibition, whereas those of VPm neurons augmented. Mean firing rates in Rt increased 42 spikes/s over 1-40 Hz, providing tonic (low RM) inhibition during high-frequency stimulation that may enhance VPm signal-to-noise ratios. Consistent with this view, VPm mean firing rates increased only 13 spikes/s over 1-40 Hz, and inter-deflection activity was suppressed to a greater extent than stimulus-evoked responses. Rt inhibition is likely to act in concert with actions of neuromodulators in optimizing thalamic temporal signaling of high-frequency whisker movements.

State-dependent processing of sensory stimuli by thalamic reticular neurons.

Inhibitory neurons of the thalamic reticular (RT) nucleus fire in two activity modes, burst and tonic, depending on an animal's behavioral state. In tonic mode, depolarized RT cells fire single action potentials continuously, whereas burst firing consists of grouped discharges separated by periods of quiescence. To determine how these firing modes affect sensory-evoked RT responses, single-unit responses to controlled whisker deflections were analyzed according to the burst versus tonic mode of spontaneous activity (SA) preceding the response. After burst mode activity (i.e., either quiescence or spontaneous bursts), responses exhibited a slow approximately 15 msec rise to peak firing rates followed by a approximately 35 msec decay. Interspike intervals within the response exhibited accelerando-decelerando patterns similar to those of spontaneous bursts. After tonic mode activity (i.e., single spikes), responses had a nearly instantaneous approximately 1.5 msec rise-to-peak followed by a approximately 40 msec decay, with large spike counts (5.2 spikes per stimulus) similar to those evoked in burst mode (6.2 spikes per stimulus). Interspike intervals were longer in tonic mode and exhibited a decelerando pattern. Initial evoked spikes, however, had shorter latencies and greater synchrony, contributing to the rapid onset of tonic population response. Shifts from quiescent (presumed burst mode) to tonic SA could be induced by either previous whisker deflections or iontophoretic application of NMDA; both manipulations effected appropriate shifts from burst to tonic response spike patterns. In awake animals, burst and tonic firing in RT, as in thalamocortical relay nuclei, may reflect sensory processing strategies appropriate for different behavioral and attentional states.

Local field potentials and the encoding of whisker deflections by population firing synchrony in thalamic barreloids.

In layer IV of rat somatosensory cortex, barrel circuitry is highly sensitive to thalamic population firing rates during the first few milliseconds of the whisker-evoked response. This sensitivity of barrel neurons to thalamic firing synchrony was inferred previously from analysis of simulated barrel circuitry and from single-unit recordings performed one at a time. In this study, we investigate stimulus-dependent synchronous activity in the thalamic ventral posteromedial nucleus (VPm) using the more direct approach of local field potential (LFP) recording. We report that thalamic barreloid neurons generate larger magnitude LFP responses to principal versus adjacent whiskers, to preferred versus nonpreferred movement directions, and to high- versus low-velocity/acceleration deflections. Responses were better predicted by acceleration than velocity, and they were insensitive to the final amplitude of whisker deflection. Importantly, reliable and robust stimulus/response relationships were found only for the initial 1.2-7.5 ms of the thalamic LFP response, reflecting arrival of afferent information from the brain stem. Later components of the thalamic response, which are likely to coincide with arrival of inhibitory inputs from the thalamic reticular nucleus and excitatory inputs from the barrel cortex itself, are variable and poorly predicted by stimulus parameters. Together with previous results, these findings underscore a critical role for thalamic firing synchrony in the encoding of small but rapidly changing perturbations of specific whiskers in particular directions.