Tilman Stephani

The spinal cord is an important component of the central nervous system for the processing of sensorimotor information transmitted between the body and the brain. Electrospinography (ESG) is the most accessible non-invasive technique for recording spinal signals in humans, but the detrimental impact of physiological noise (mostly of cardiac nature) has prevented widespread adoption. Here, we aim to address this issue by examining various denoising algorithms for cardiac artefact reduction—including approaches based on principal component analysis (PCA), independent component analysis (ICA), signal space projection (SSP), canonical correlation analysis (CCA), and denoising separation of sources (DSS). We observed that in situations where large number of spinal electrodes are used, ICA and SSP offer the best results in terms of balancing the removal of noise and preserving neural information of interest. In cases where only a small number of electrodes are available, an approach based on PCA is deemed helpful. Finally, we also approached this issue from a signal-enhancement perspective by applying CCA and DSS directly to signals of interest, namely spinal somatosensory evoked potentials (SEPs, especially the N13 and N22 components in the cervical and lumbar spinal cord, respectively). We observed that in cases where extensive electrode arrays are used in the context of task-based designs, CCA reveals clear evoked spinal potentials even with single-trial resolution. Taken together, there are several appropriate algorithms for physiological noise removal and/or signal enhancement in ESG, rendering this an accessible and easy-to-use technique for non-invasive assessments of human spinal cord function.

Perception is biased by expectations and previous actions. Pre-stimulus brain oscillations are a potential candidate for implementing biases in the brain. In two EEG studies (43 and 39 participants) on somatosensory near-threshold detection, we investigated the pre-stimulus neural correlates of an (implicit) previous choice bias and an explicit bias. The explicit bias was introduced by informing participants about stimulus probability on a single-trial level (volatile context) or block-wise (stable context). Behavioural analysis confirmed adjustments in the decision criterion and confidence ratings according to the cued probabilities and previous choice-induced biases. Pre-stimulus beta power with distinct sources in sensory and higher-order cortical areas predicted explicit and implicit biases, respectively, on a single subject level and partially mediated the impact of previous choice and stimulus probability on the detection response. We suggest pre-stimulus beta oscillations in distinct brain areas as a neural correlate of explicit and implicit biases in somatosensory perception.

Neural states shape perception at earliest cortical processing levels. Previous work in humans showed a relationship between initial cortical excitation, as indicated by the N20 component of the somatosensory evoked potential (SEP), prestimulus alpha oscillations, and the perceived intensity in a somatosensory discrimination paradigm. Here we address the follow-up question whether these excitability dynamics reflect changes in feedforward or feedback signals. To distinguish feedforward neural signals from feedback signals, we leveraged high-frequency oscillations (HFO) which have previously been shown to correspond to neuronal population spiking activity of the first excitatory feedforward volley in the somatosensory cortex. We examined these HFO in electroencephalography (EEG) data of 32 male human participants, performing a somatosensory intensity discrimination task. Spatial filtering and time–frequency analyses allowed to clearly distinguish HFO from the lower-frequency, conventional SEP. Using Bayesian statistics, we found evidence against the involvement of HFO in moment-to-moment variability of perceived stimulus intensity, in contrast to previously observed prestimulus alpha and N20 effects of the conventional SEP. Given that the N20 component presumably reflects backpropagating membrane potentials toward the apical dendrites (distal dendritic sites), we argue that top-down feedback processes (e.g., related to alpha oscillations) may thus rely on activity modulations at those distal dendrites of involved pyramidal cells rather than on synchronous output firing changes at their basal compartments.

Information transmission in the central nervous system (CNS) relies on the propagation of action potentials (neuronal spikes), traditionally thought to be inaccessible with electroencephalography (EEG). Converging evidence from human and animal recordings however demonstrates that high frequency oscillations (HFOs) in EEG recordings represent neuronal population spiking, complementing standard low-frequency EEG recordings in cortex. Here, we developed an approach to simultaneously record HFOs to somatosensory stimulation across the entire human CNS noninvasively: from spinal over subcortical to cortical stages. Our approach provides a detailed temporal and spectral characterization of HFOs at each CNS level, delineating how they evolve across the somatosensory processing hierarchy. Critically, we show that low and high frequency electrophysiological responses represent at least partly independent processing channels across the CNS. We thus demonstrate the feasibility of studying the full spectrum of neuronal information processing across the CNS, offering a multi-level window into human neurophysiology in health and disease.

Objective. Serving as a channel for communication with locked-in patients or control of prostheses, sensorimotor brain–computer interfaces (BCIs) decode imaginary movements from the recorded activity of the user’s brain. However, many individuals remain unable to control the BCI, and the underlying mechanisms are unclear. The user’s BCI performance was previously shown to correlate with the resting-state signal-to-noise ratio (SNR) of the mu rhythm and the phase synchronization (PS) of the mu rhythm between sensorimotor areas. Yet, these predictors of performance were primarily evaluated in a single BCI session, while the longitudinal aspect remains rather uninvestigated. In addition, different analysis pipelines were used to estimate PS in source space, potentially hindering the reproducibility of the results. Approach. To systematically address these issues, we performed an extensive validation of the relationship between pre-stimulus SNR, PS, and session-wise BCI performance using a publicly available dataset of 62 human participants performing up to 11 sessions of BCI training. We performed the analysis in sensor space using the surface Laplacian and in source space by combining 24 processing pipelines in a multiverse analysis. This way, we could investigate how robust the observed effects were to the selection of the pipeline. Main results. Our results show that SNR had both between- and within-subject effects on BCI performance for the majority of the pipelines. In contrast, the effect of PS on BCI performance was less robust to the selection of the pipeline and became non-significant after controlling for SNR. Significance. Taken together, our results demonstrate that changes in neuronal connectivity within the sensorimotor system are not critical for learning to control a BCI, and interventions that increase the SNR of the mu rhythm might lead to improvements in the user’s BCI performance.

The spinal cord is a vital component of the central nervous system for the processing of sensorimotor information transmitted between the body and the brain. Electrospinography (ESG) is the most accessible non-invasive technique for recording spinal signals in humans, but the vast and detrimental impact of physiological noise (mostly of cardiac nature) has prevented widespread adoption. Here, we aim to address this issue by examining various denoising algorithms for cardiac artefact reduction – including principal component analysis-based techniques (PCA), independent component analysis-based approaches (ICA) and signal space projection (SSP). We observed that in situations where large numbers of spinal electrodes are used, SSP offers the best results in terms of balancing the removal of harmful noise and preserving neural information of interest. In cases where only a small number of electrodes are available, an approach based on PCA is deemed helpful. Approaches based on ICA were found to be unsuitable for cardiac artefact removal in ESG, due to a suboptimal balance of artefact removal and signal preservation. Finally, we also approached this issue from a signal-enhancement perspective and observed that in cases where extensive electrode arrays are used in the context of task-based designs, a spatial filtering technique based on canonical correlation analysis (CCA) reveals clear evoked spinal potentials even with single-trial resolution. Taken together, there are several appropriate algorithms for physiological noise removal in ESG, rendering this an accessible and easy-to-use technique for non-invasive assessments of human spinal cord function.

Neural states shape perception at earliest cortical processing levels. Previous work in humans showed a relationship between initial cortical excitation, as indicated by the N20 component of the somatosensory evoked potential (SEP), pre-stimulus alpha oscillations, and the perceived intensity in a somatosensory discrimination paradigm. Here we address the follow-up question whether these excitability dynamics reflect changes in feedforward or feedback signals. We leveraged high-frequency oscillations (HFO) as a metric for neuronal population spiking activity of the first excitatory feedforward volley in the somatosensory cortex, non-invasively extracted from electroencephalography (EEG) data of 32 male human participants. Using Bayesian statistics, we found evidence against the involvement of HFO in moment-to-moment variability of perceived stimulus intensity, in contrast to previously observed pre-stimulus alpha and N20 effects. Given that the N20 component presumably reflects backpropagating membrane potentials towards the apical dendrites, we argue that top-down feedback processes (e.g., related to alpha oscillations) may thus rely on modulations at distal dendritic sites of involved pyramidal cells rather than on output firing changes at their basal compartments.

Perception is biased by expectations and previous actions. Pre-stimulus brain oscillations are a potential candidate for implementing biases in the brain. In two EEG studies (43 and 39 participants) on somatosensory near-threshold detection, we investigated the pre-stimulus neural correlates of an (implicit) previous choice bias and an explicit bias. The explicit bias was introduced by informing participants about stimulus probability on a single-trial level (volatile context) or block-wise (stable context). Behavioural analysis confirmed adjustments in the decision criterion and confidence ratings according to the cued probabilities and previous choice-induced biases. Pre-stimulus beta power with distinct sources in sensory and higher-order cortical areas predicted explicit and implicit biases, respectively, on a single subject level and partially mediated the impact of previous choice and stimulus probability on the detection response. We suggest pre-stimulus beta oscillations in distinct brain areas as a neural correlate of explicit and implicit biases in somatosensory perception.

Since the advent of Deepfakes in digital media, the development of robust and reliable detection mechanism is urgently called for. In this study, we explore a novel approach to Deepfake detection by utilizing electroencephalography (EEG) measured from the neural processing of a human participant who viewed and categorized Deepfake stimuli from the FaceForensics++ datset. These measurements serve as input features to a binary support vector classifier, trained to discriminate between real and manipulated facial images. We examine whether EEG data can inform Deepfake detection and also if it can provide a generalized representation capable of identifying Deepfakes beyond the training domain. Our preliminary results indicate that human neural processing signals can be successfully integrated into Deepfake detection frameworks and hint at the potential for a generalized neural representation of artifacts in computer generated faces. Moreover, our study provides next steps towards the understanding of how digital realism is embedded in the human cognitive system, possibly enabling the development of more realistic digital avatars in the future.

Objective: Serving as a channel for communication with locked-in patients or control of prostheses, sensorimotor brain-computer interfaces (BCIs) decode imaginary movements from the recorded activity of the user’s brain. However, many individuals remain unable to control the BCI, and the underlying mechanisms are unclear. The user’s BCI performance was previously shown to correlate with the resting-state signal-to-noise ratio (SNR) of the mu rhythm and the phase synchronization (PS) of the mu rhythm between sensorimotor areas. Yet, these predictors of performance were primarily evaluated in a single BCI session, while the longitudinal aspect remains rather uninvestigated. In addition, different analysis pipelines were used to estimate PS in source space, potentially hindering the reproducibility of the results. Approach: To systematically address these issues, we performed an extensive validation of the relationship between pre-stimulus SNR, PS, and session-wise BCI performance using a publicly available dataset of 62 human participants performing up to 11 sessions of BCI training. We performed the analysis in sensor space using the surface Laplacian and in source space by combining 24 processing pipelines in a multiverse analysis. This way, we could investigate how robust the observed effects were to the selection of the pipeline. Main results: Our results show that SNR had both between- and within-subject effects on BCI performance for the majority of the pipelines. In contrast, the effect of PS on BCI performance was less robust to the selection of the pipeline and became non-significant after controlling for SNR. Significance: Taken together, our results demonstrate that changes in neuronal connectivity within the sensorimotor system are not critical for learning to control a BCI, and interventions that increase the SNR of the mu rhythm might lead to improvements in the user’s BCI performance.