Evaluating the effect of denoising submillimeter auditory fMRI data with NORDIC

Functional magnetic resonance imaging (fMRI) has emerged as an essential tool for exploring human brain function. Submillimeter fMRI, in particular, has emerged as a tool to study mesoscopic computations. The inherently low signal-to-noise ratio (SNR) at submillimeter resolutions warrants the use of denoising approaches tailored at reducing thermal noise - the dominant contributing noise component in high resolution fMRI. NORDIC PCA is one of such approaches, and has been benchmarked against other approaches in several applications. Here, we investigate the effects that two versions of NORDIC denoising have on auditory submillimeter data. As investigating auditory functional responses poses unique challenges, we anticipated that the benefit of this technique would be especially pronounced. Our results show that NORDIC denoising improves the detection sensitivity and the reliability of estimates in submillimeter auditory fMRI data. These effects can be explained by the reduction of the noise-induced signal variability. However, we also observed a reduction in the average response amplitude (percent signal), which may suggest that a small amount of signal was also removed. We conclude that, while evaluating the effects of the signal reduction induced by NORDIC may be necessary for each application, using NORDIC in high resolution auditory fMRI studies may be advantageous because of the large reduction in variability of the estimated responses.

Voxelwise encoding models of body stimuli reveal a representational gradient from low-level visual features to postural features in occipitotemporal cortex

Research on body representation in the brain has focused on category-specific representation, using fMRI to investigate the response pattern to body stimuli in occipitotemporal cortex without so far addressing the issue of the specific computations involved in body selective regions, only defined by higher order category selectivity. This study used ultra-high field fMRI and banded ridge regression to investigate the coding of body images, by comparing the performance of three encoding models in predicting brain activity in occipitotemporal cortex and specifically the extrastriate body area (EBA). Our results suggest that bodies are encoded in occipitotemporal cortex and in the EBA according to a combination of low-level visual features and postural features.

Layer-specific correlates of detected and undetected auditory targets during attention

In everyday life, the processing of acoustic information allows us to react to subtle changes in the auditory scene. Yet even when closely attending to sounds in the context of a task, we occasionally miss task-relevant features. The neural computations that underlie our ability to detect behavioral relevant sound changes are thought to be grounded in both feedforward and feedback processes within the auditory hierarchy. Here, we assessed the role of feedforward and feedback contributions in primary and non-primary auditory areas during behavioral detection of target sounds using submillimeter spatial resolution functional magnetic resonance imaging (fMRI) at high-fields (7 T) in humans. We demonstrate that the successful detection of subtle temporal shifts in target sounds leads to a selective increase of activation in superficial layers of primary auditory cortex (PAC). These results indicate that feedback signals reaching as far back as PAC may be relevant to the detection of targets in the auditory scene.

Selective attention sharpens population receptive fields in human auditory cortex

Selective attention enables the preferential processing of relevant stimulus aspects. Invasive animal studies have shown that attending a sound feature rapidly modifies neuronal tuning throughout the auditory cortex. Human neuroimaging studies have reported enhanced auditory cortical responses with selective attention. To date, it remains unclear how the results obtained with functional magnetic resonance imaging (fMRI) in humans relate to the electrophysiological findings in animal models. Here we aim to narrow the gap between animal and human research by combining a selective attention task similar in design to those used in animal electrophysiology with high spatial resolution ultra-high field fMRI at 7 Tesla. Specifically, human participants perform a detection task, whereas the probability of target occurrence varies with sound frequency. Contrary to previous fMRI studies, we show that selective attention resulted in population receptive field sharpening, and consequently reduced responses, at the attended sound frequencies. The difference between our results to those of previous fMRI studies supports the notion that the influence of selective attention on auditory cortex is diverse and may depend on context, stimulus, and task.

Using high spatial resolution fMRI to understand representation in the auditory network

Following rapid methodological advances, ultra-high field (UHF) functional and anatomical magnetic resonance imaging (MRI) has been repeatedly and successfully used for the investigation of the human auditory system in recent years. Here, we review this work and argue that UHF MRI is uniquely suited to shed light on how sounds are represented throughout the network of auditory brain regions. That is, the provided gain in spatial resolution at UHF can be used to study the functional role of the small subcortical auditory processing stages and details of cortical processing. Further, by combining high spatial resolution with the versatility of MRI contrasts, UHF MRI has the potential to localize the primary auditory cortex in individual hemispheres. This is a prerequisite to study how sound representation in higher-level auditory cortex evolves from that in early (primary) auditory cortex. Finally, the access to independent signals across auditory cortical depths, as afforded by UHF, may reveal the computations that underlie the emergence of an abstract, categorical sound representation based on low-level acoustic feature processing. Efforts on these research topics are underway. Here we discuss promises as well as challenges that come with studying these research questions using UHF MRI, and provide a future outlook.

Audiovisual Interactions Among Near-Threshold Oscillating Stimuli in the Far Periphery Are Phase-Dependent

Recent studies have highlighted the possible contributions of direct connectivity between early sensory cortices to audiovisual integration. Anatomical connections between the early auditory and visual cortices are concentrated in visual sites representing the peripheral field of view. Here, we aimed to engage early sensory interactive pathways with simple, far-peripheral audiovisual stimuli (auditory noise and visual gratings). Using a modulation detection task in one modality performed at an 84% correct threshold level, we investigated multisensory interactions by simultaneously presenting weak stimuli from the other modality in which the temporal modulation was barely-detectable (at 55 and 65% correct detection performance). Furthermore, we manipulated the temporal congruence between the cross-sensory streams. We found evidence for an influence of barely-detectable visual stimuli on the response times for auditory stimuli, but not for the reverse effect. These visual-to-auditory influences only occurred for specific phase-differences (at onset) between the modulated audiovisual stimuli. We discuss our findings in the light of a possible role of direct interactions between early visual and auditory areas, along with contributions from the higher-order association cortex. In sum, our results extend the behavioral evidence of audio-visual processing to the far periphery, and suggest - within this specific experimental setting - an asymmetry between the auditory influence on visual processing and the visual influence on auditory processing.

Investigating the Reliability of Population Receptive Field Size Estimates Using fMRI

In functional MRI (fMRI), population receptive field (pRF) models allow a quantitative description of the response as a function of the features of the stimuli that are relevant for each voxel. The most popular pRF model used in fMRI assumes a Gaussian shape in the features space (e.g., the visual field) reducing the description of the voxel’s pRF to the Gaussian mean (the pRF preferred feature) and standard deviation (the pRF size). The estimation of the pRF mean has been proven to be highly reliable. However, the estimate of the pRF size has been shown not to be consistent within and between subjects. While this issue has been noted experimentally, here we use an optimization theory perspective to describe how the inconsistency in estimating the pRF size is linked to an inherent property of the Gaussian pRF model. When fitting such models, the goodness of fit is less sensitive to variations in the pRF size than to variations in the pRF mean. We also show how the same issue can be considered from a bias-variance perspective. We compare different estimation procedures in terms of the reliability of their estimates using simulated and real fMRI data in the visual (using the Human Connectome Project database) and auditory domain. We show that, the reliability of the estimate of the pRF size can be improved considering a linear combination of those pRF models with similar goodness of fit or a permutation based approach. This increase in reliability of the pRF size estimate does not affect the reliability of the estimate of the pRF mean and the prediction accuracy.

Estimating Exposome Score for Schizophrenia Using Predictive Modeling Approach in Two Independent Samples: The Results From the EUGEI Study

Exposures constitute a dense network of the environment: exposome. Here, we argue for embracing the exposome paradigm to investigate the sum of nongenetic "risk" and show how predictive modeling approaches can be used to construct an exposome score (ES; an aggregated score of exposures) for schizophrenia. The training dataset consisted of patients with schizophrenia and controls, whereas the independent validation dataset consisted of patients, their unaffected siblings, and controls. Binary exposures were cannabis use, hearing impairment, winter birth, bullying, and emotional, physical, and sexual abuse along with physical and emotional neglect. We applied logistic regression (LR), Gaussian Naive Bayes (GNB), the least absolute shrinkage and selection operator (LASSO), and Ridge penalized classification models to the training dataset. ESs, the sum of weighted exposures based on coefficients from each model, were calculated in the validation dataset. In addition, we estimated ES based on meta-analyses and a simple sum score of exposures. Accuracy, sensitivity, specificity, area under the receiver operating characteristic, and Nagelkerke's R2 were compared. The ESMeta-analyses performed the worst, whereas the sum score and the ESGNB were worse than the ESLR that performed similar to the ESLASSO and ESRIDGE. The ESLR distinguished patients from controls (odds ratio [OR] = 1.94, P < .001), patients from siblings (OR = 1.58, P < .001), and siblings from controls (OR = 1.21, P = .001). An increase in ESLR was associated with a gradient increase of schizophrenia risk. In reference to the remaining fractions, the ESLR at top 30%, 20%, and 10% of the control distribution yielded ORs of 3.72, 3.74, and 4.77, respectively. Our findings demonstrate that predictive modeling approaches can be harnessed to evaluate the exposome.

Mapping the human subcortical auditory system using histology, postmortem MRI and in vivo MRI at 7T

Studying the human subcortical auditory system non-invasively is challenging due to its small, densely packed structures deep within the brain. Additionally, the elaborate three-dimensional (3-D) structure of the system can be difficult to understand based on currently available 2-D schematics and animal models. Wfe addressed these issues using a combination of histological data, post mortem magnetic resonance imaging (MRI), and in vivo MRI at 7 Tesla. We created anatomical atlases based on state-of-the-art human histology (BigBrain) and postmortem MRI (50 µm). We measured functional MRI (fMRI) responses to natural sounds and demonstrate that the functional localization of subcortical structures is reliable within individual participants who were scanned in two different experiments. Further, a group functional atlas derived from the functional data locates these structures with a median distance below 2 mm. Using diffusion MRI tractography, we revealed structural connectivity maps of the human subcortical auditory pathway both in vivo (1050 µm isotropic resolution) and post mortem (200 µm isotropic resolution). This work captures current MRI capabilities for investigating the human subcortical auditory system, describes challenges that remain, and contributes novel, openly available data, atlases, and tools for researching the human auditory system.

Methods for computing the maximum performance of computational models of fMRI responses

Computational neuroimaging methods aim to predict brain responses (measured e.g. with functional magnetic resonance imaging [fMRI]) on the basis of stimulus features obtained through computational models. The accuracy of such prediction is used as an indicator of how well the model describes the computations underlying the brain function that is being considered. However, the prediction accuracy is bounded by the proportion of the variance of the brain response which is related to the measurement noise and not to the stimuli (or cognitive functions). This bound to the performance of a computational model has been referred to as the noise ceiling. In previous fMRI applications two methods have been proposed to estimate the noise ceiling based on either a split-half procedure or Monte Carlo simulations. These methods make different assumptions over the nature of the effects underlying the data, and, importantly, their relation has not been clarified yet. Here, we derive an analytical form for the noise ceiling that does not require computationally expensive simulations or a splitting procedure that reduce the amount of data. The validity of this analytical definition is proved in simulations, we show that the analytical solution results in the same estimate of the noise ceiling as the Monte Carlo method. Considering different simulated noise structure, we evaluate different estimators of the variance of the responses and their impact on the estimation of the noise ceiling. We furthermore evaluate the interplay between regularization (often used to estimate model fits to the data when the number of computational features in the model is large) and model complexity on the performance with respect to the noise ceiling. Our results indicate that when considering the variance of the responses across runs, computing the noise ceiling analytically results in similar estimates as the split half estimator and approaches the true noise ceiling under a variety of simulated noise scenarios. Finally, the methods are tested on real fMRI data acquired at 7 Tesla.

Encoding computational models in brain responses measured with fMRI allows testing the algorithmic representations carried out by the neural population within voxels. The accuracy of a model in predicting new responses is used as a measure of the brain validity of the computational model being tested, but the result of this analysis is determined not only by how precisely the model describes the responses but also by the quality of the data. In this article, we evaluate existing approaches to estimate the best possible accuracy that any computational model can achieve conditioned to the amount of measurement noise that is present in the experimental data (i.e. the noise ceiling). Additionally we introduce a close form estimation of the noise ceiling that does not require computationally or data expensive procedures. All the methods are compared using simulated and real fMRI data. We draw conclusions over the impact of regularization procedures and make practical recommendations on how to report the results of computational models in neuroimaging.

Fast Gaussian Naïve Bayes for searchlight classification analysis

The searchlight technique is a variant of multivariate pattern analysis (MVPA) that examines neural activity across large sets of small regions, exhaustively covering the whole brain. This usually involves application of classifier algorithms across all searchlights, which entails large computational costs especially when testing the statistical significance of the accuracies with permutation methods. In this article, a new implementation of the Gaussian Naive Bayes classifier is presented (henceforth massive-GNB). This approach allows classification in all searchlights simultaneously, and is faster than previously published searchlight GNB implementations, as well as other more complex classifiers including support vector machines (SVM). To ensure that the gain in speed for GNB would be useful in searchlight analysis, we compared the accuracies of massive-GNB and SVM in detecting the lateral occipital complex (LOC) in an fMRI localizer experiment (26 subjects). Moreover, this region as defined in a meta-analysis of many activation studies was used as a gold standard to compare error rates for both classifiers. In individual searchlights, SVM was somewhat more accurate than massive-GNB and more selective in detecting the meta-analytic LOC. However, with multiple comparison correction at the cluster-level the two classifiers performed equivalently. Thus for cluster-level analysis, massive-GNB produces an accuracy similar to more sophisticated classifiers but with a substantial gain in speed. Massive-GNB (available as a public Matlab toolbox) could facilitate the more widespread use of searchlight analysis.

3D statistical parametric mapping of quiet sleep EEG in the first year of life

This paper extends previously developed 3D SPM for Electrophysiological Source Imaging (Bosch et al., 2001) for neonate EEG. It builds on a prior paper by our group that established age dependent means and standard deviations for the scalp EEG Broad Band Spectral Parameters of children in the first year of life. We now present developmental equations for the narrow band log spectral power of EEG sources, obtained from a sample of 93 normal neonates from age 1 to 10 months in quiet sleep. The main finding from these regressions is that EEG power from 0.78 to 7.5 Hz decreases with age and also for 45-50 Hz. By contrast, there is an increase with age in the frequency band of 19-32 Hz localized to parietal, temporal and occipital areas. Deviations from the norm were analyzed for normal neonates and 17 with brain damage. The diagnostic accuracy (measured by the area under the ROC curve) of EEG source SPM is 0.80, 0.69 for average reference scalp EEG SPM, and 0.48 for Laplacian EEG SPM. This superior performance of 3D SPM over scalp qEEG suggests that it might be a promising approach for the evaluation of brain damage in the first year of life.

A zero-training algorithm for EEG single-trial classification applied to a face recognition ERP experiment

This paper proposes a machine learning based approach to discriminate between EEG single trials of two experimental conditions in a face recognition experiment. The algorithm works using a single-trial EEG database of multiple subjects and thus does not require subject-specific training data. This approach supports the idea that zero-training classification and on-line detection Brain Computer Interface (BCI) systems are areas with a significant amount of potential.