Challenges and opportunities of mesoscopic brain mapping with fMRI

Towards whole brain mapping of the haemodynamic response function

Functional magnetic resonance imaging time-series are conventionally processed by linear modelling the evoked response as the convolution of the experimental conditions with a stereotyped haemodynamic response function (HRF). However, the neural signal in response to a stimulus can vary according to task, brain region, and subject-specific conditions. Moreover, HRF shape has been suggested to carry physiological information. The BOLD signal across a range of sensorial and cognitive tasks was fitted using a sine series expansion, and modelled signals were deconvolved, thus giving rise to a task-specific deconvolved HRF (dHRF), which was characterized in terms of amplitude, latency, time-to-peak and full-width at half maximum for each task. We found that the BOLD response shape changes not only across activated regions and tasks, but also across subjects despite the age homogeneity of the cohort. Largest variabilities were observed in mean amplitude and latency across tasks and regions, while time-to-peak and full width at half maximum were relatively more consistent. Additionally, the dHRF was found to deviate from canonicity in several brain regions. Our results suggest that the choice of a standard, uniform HRF may be not optimal for all fMRI analyses and may lead to model misspecifications and statistical bias.

In the brain of the beholder: bi-stable motion reveals mesoscopic-scale feedback modulation in V1

Understanding the neural processes underlying conscious perception remains a central goal in neuroscience. Visual illusions, whether static or dynamic, provide an effective ecological paradigm for studying conscious perception, as they induce subjective experiences from constant visual inputs. While previous neuroimaging studies have dissociated perceptual interpretation of visual motion from sensory input within the motion-sensitive area (hMT+) in humans, less is known about the role of the primary visual area (V1) and its relationship to hMT+ during a bistable perception. To address this, we conducted a layer-fMRI study at 7 T with human participants exposed to a bistable motion quartet stimulus. Despite a constant sensory input, the bistable motion quartet elicits switching horizontal and vertical apparent motion percepts likely due to lateral and feedback connections across low and high-level brain regions (feedback processing). As control, we used an "unambiguous" version of the motion quartet, hereafter referred to as "physical" motion stimulus, where horizontal and vertical motion is physically presented as visual stimulus in an alternated fashion (feedforward processing). With the advantage of a sub-millimeter resolution gained at ultra-high magnetic field (7 Tesla), we aimed to unveil the differential laminar modulation of V1 (early visual area) and hMT+ (high-order visual area) during the physical and bistable condition. Our results indicate that: (1) hMT+ functional activity correlates with conscious perception during both physical and ambiguous stimuli with similar strength. There is no evidence of differential laminar profiles in hMT+ between the two experimental conditions. (2) Between inducer squares, V1 shows a significantly reduced functional response to the ambiguous stimulus compared to the physical stimulus, as it primarily reflects feedback signals with diminished feedforward input. Distinct V1 laminar profiles differentiate the two experimental conditions. (3) The temporal dynamics of V1 and hMT+ become more similar during the ambiguous condition. (4) V1 exhibits reduced specificity to horizontal and vertical motion perception during the ambiguous condition at the retinotopic locations corresponding to the perceived motion. Our findings demonstrate that during the ambiguous condition, there is a stronger temporal coupling between hMT+ and V1 due to feedback signals from hMT+ to V1. Such feedback to V1 might be contributing to the stabilization of the vivid perception of directed motion at the face of constant ambiguous stimulation.

The laminar pattern of proprioceptive activation in human primary motor cortex

The primary motor cortex (M1) is increasingly being recognized for its vital role in proprioceptive somatosensation. However, our current understanding of proprioceptive processing at the laminar scale is limited. Empirical findings in primates and rodents suggest a pronounced role of superficial cortical layers, but the involvement of deep layers has yet to be examined in humans. Submillimeter resolution functional magnetic resonance imaging (fMRI) has emerged in recent years, paving the way for studying layer-dependent activity in humans (laminar fMRI). In the present study, laminar fMRI was employed to investigate the influence of proprioceptive somatosensation on M1 deep layer activation using passive finger movements. Significant M1 deep layer activation was observed in response to proprioceptive stimulation across 10 healthy subjects using a vascular space occupancy (VASO)-sequence at 7 T. For further validation, two additional datasets were included which were obtained using a balanced steady-state free precession sequence with ultrahigh (0.3 mm) in-plane resolution, yielding converging results. These results were interpreted in the light of previous laminar fMRI studies and the active inference account of motor control. We propose that a considerable proportion of M1 deep layer activation is due to proprioceptive influence and that deep layers of M1 constitute a key component in proprioceptive circuits.

Layer-specific BOLD effects in gradient and spin-echo acquisitions in somatosensory cortex

PURPOSE

Previous studies have shown varied BOLD signals with gradient echo (GE) across cortical depth. To interpret these variations, and understand the effects of vascular geometry and size, the magnitudes and layer distributions of GE and spin-echo (SE) BOLD functional MRI signals were compared in the somatosensory cortex of squirrel monkeys during tactile stimulation and in a resting state at high spatial resolution and high field.

METHODS

A block-design stimulation was used to identify tactile-evoked activation signals in somatosensory Areas 3b and 1. Layer-specific connectivities were calculated using resting-state data. Signal power spectra were compared by depth and pulse sequence. The measured ratios of transverse relaxation rate changes were compared with Anderson and Weiss's model.

RESULTS

SE signals showed a 26% lower percentage signal change during tactile stimulation compared with GE, along with a slower time course. SE signals remained consistent but weaker in lower layers, whereas GE signals decreased with cortical depth. This pattern extended to resting-state power spectra. Resting-state functional connectivity indicated larger connectivity between the top layers of Area 3b and Area 1 for GE, with minimal changes for SE. Comparisons with theory suggest vessel diameters ranging from 19.4 to 9 microns are responsible for BOLD effects across cortical layers at 9.4 T.

CONCLUSION

These results provide further evidence that at high field, SE BOLD signals are relatively free of contributions from sources other than microvascular changes in response to neural activity, whereas GE signals, even in the superficial layers, are not dominated by very large vessels.

Multilayer Network Analysis across Cortical Depths in Resting-State 7T fMRI

In graph theory, "multilayer networks" represent systems involving several interconnected topological levels. One example in neuroscience is the stratification of connections between different cortical depths or "laminae", which is becoming non-invasively accessible in humans using ultra-high-resolution functional MRI (fMRI). Here, we applied multilayer graph theory to examine functional connectivity across different cortical depths in humans, using 7T fMRI (1-mm3 voxels; 30 participants). Blood oxygenation level dependent (BOLD) signals were derived from five depths between the white matter and pial surface. We compared networks where the inter-regional connections were limited to a single cortical depth only ("layer-by-layer matrices") to those considering all possible connections between areas and cortical depths ("multilayer matrix"). We utilized global and local graph theory features that quantitatively characterize network attributes including network composition, nodal centrality, path-based measures, and hub segregation. Detecting functional differences between cortical depths was improved using multilayer connectomics compared to the layer-by-layer versions. Superficial depths of the cortex dominated information transfer and deeper depths drove clustering. These differences were largest in frontotemporal and limbic regions. fMRI functional connectivity across different cortical depths may contain neurophysiologically relevant information; thus, multilayer connectomics could provide a methodological framework for studies on how information flows across this stratification.

Non-Cartesian 3D-SPARKLING vs Cartesian 3D-EPI encoding schemes for functional Magnetic Resonance Imaging at 7 Tesla

The quest for higher spatial and/or temporal resolution in functional MRI (fMRI) while preserving a sufficient temporal signal-to-noise ratio (tSNR) has generated a tremendous amount of methodological contributions in the last decade ranging from Cartesian vs. non-Cartesian readouts, 2D vs. 3D acquisition strategies, parallel imaging and/or compressed sensing (CS) accelerations and simultaneous multi-slice acquisitions to cite a few. In this paper, we investigate the use of a finely tuned version of 3D-SPARKLING. This is a non-Cartesian CS-based acquisition technique for high spatial resolution whole-brain fMRI. We compare it to state-of-the-art Cartesian 3D-EPI during both a retinotopic mapping paradigm and resting-state acquisitions at 1mm3 (isotropic spatial resolution). This study involves six healthy volunteers and both acquisition sequences were run on each individual in a randomly-balanced order across subjects. The performances of both acquisition techniques are compared to each other in regards to tSNR, sensitivity to the BOLD effect and spatial specificity. Our findings reveal that 3D-SPARKLING has a higher tSNR than 3D-EPI, an improved sensitivity to detect the BOLD contrast in the gray matter, and an improved spatial specificity. Compared to 3D-EPI, 3D-SPARKLING yields, on average, 7% more activated voxels in the gray matter relative to the total number of activated voxels.

An Agent-Based Model to Reproduce the Boolean Logic Behaviour of Neuronal Self-Organised Communities through Pulse Delay Modulation and Generation of Logic Gates

The human brain is arguably the most complex "machine" to ever exist. Its detailed functioning is yet to be fully understood, let alone modelled. Neurological processes have logical signal-processing and biophysical aspects, and both affect the brain's structure, functioning and adaptation. Mathematical approaches based on both information and graph theory have been extensively used in an attempt to approximate its biological functioning, along with Artificial Intelligence frameworks inspired by its logical functioning. In this article, an approach to model some aspects of the brain learning and signal processing is presented, mimicking the metastability and backpropagation found in the real brain while also accounting for neuroplasticity. Several simulations are carried out with this model to demonstrate how dynamic neuroplasticity, neural inhibition and neuron migration can reshape the brain's logical connectivity to synchronise signal processing and obtain certain target latencies. This work showcases the importance of dynamic logical and biophysical remodelling in brain plasticity. Combining mathematical (agents, graph theory, topology and backpropagation) and biomedical ingredients (metastability, neuroplasticity and migration), these preliminary results prove complex brain phenomena can be reproduced-under pertinent simplifications-via affordable computations, which can be construed as a starting point for more ambitiously accurate simulations.

Statistical inference on representational geometries

Neuroscience has recently made much progress, expanding the complexity of both neural activity measurements and brain-computational models. However, we lack robust methods for connecting theory and experiment by evaluating our new big models with our new big data. Here, we introduce new inference methods enabling researchers to evaluate and compare models based on the accuracy of their predictions of representational geometries: A good model should accurately predict the distances among the neural population representations (e.g. of a set of stimuli). Our inference methods combine novel 2-factor extensions of crossvalidation (to prevent overfitting to either subjects or conditions from inflating our estimates of model accuracy) and bootstrapping (to enable inferential model comparison with simultaneous generalization to both new subjects and new conditions). We validate the inference methods on data where the ground-truth model is known, by simulating data with deep neural networks and by resampling of calcium-imaging and functional MRI data. Results demonstrate that the methods are valid and conclusions generalize correctly. These data analysis methods are available in an open-source Python toolbox (rsatoolbox.readthedocs.io).

Improved sensitivity and microvascular weighting of 3T laminar fMRI with GE-BOLD using NORDIC and phase regression

INTRODUCTION

Functional MRI with spatial resolution in the submillimeter domain enables measurements of activation across cortical layers in humans. This is valuable as different types of cortical computations, e.g., feedforward versus feedback related activity, take place in different cortical layers. Laminar fMRI studies have almost exclusively employed 7T scanners to overcome the reduced signal stability associated with small voxels. However, such systems are relatively rare and only a subset of those are clinically approved. In the present study, we examined if the feasibility of laminar fMRI at 3T could be improved by use of NORDIC denoising and phase regression.

METHODS

5 healthy subjects were scanned on a Siemens MAGNETOM Prisma 3T scanner. To assess across-session reliability, each subject was scanned in 3-8 sessions on 3-4 consecutive days. A 3D gradient echo EPI (GE-EPI) sequence was used for BOLD acquisitions (voxel size 0.82 mm isotopic, TR = 2.2 s) using a block design finger tapping paradigm. NORDIC denoising was applied to the magnitude and phase time series to overcome limitations in temporal signal-to-noise ratio (tSNR) and the denoised phase time series were subsequently used to correct for large vein contamination through phase regression.

RESULTS AND CONCLUSION

NORDIC denoising resulted in tSNR values comparable to or higher than commonly observed at 7T. Layer-dependent activation profiles could thus be extracted robustly, within and across sessions, from regions of interest located in the hand knob of the primary motor cortex (M1). Phase regression led to substantially reduced superficial bias in obtained layer profiles, although residual macrovascular contribution remained. We believe the present results support an improved feasibility of laminar fMRI at 3T.

Novel developments in non-contrast enhanced MRI of the perivascular clearance system: What are the possibilities for Alzheimer's disease research?

The cerebral waste clearance system (i.e, glymphatic or intramural periarterial drainage) works through a network of perivascular spaces (PVS). Dysfunction of this system likely contributes to aggregation of Amyloid-β and subsequent toxic plaques in Alzheimer's disease (AD). A promising, non-invasive technique to study this system is MRI, though applications in dementia are still scarce. This review focusses on recent non-contrast enhanced (non-CE) MRI techniques which determine and visualise physiological aspects of the clearance system at multiple levels, i.e., cerebrospinal fluid flow, PVS-flow and interstitial fluid movement. Furthermore, various MRI studies focussing on aspects of the clearance system which are relevant to AD are discussed, such as studies on ageing, sleep alterations, and cognitive decline. Additionally, the complementary function of non-CE to CE methods is elaborated upon. We conclude that non-CE studies have great potential to determine which parts of the waste clearance system are affected by AD and in which stages of cognitive impairment dysfunction of this system occurs, which could allow future clinical trials to target these specific mechanisms.

Surgery of Motor Eloquent Glioblastoma Guided by TMS-Informed Tractography: Driving Resection Completeness Towards Prolonged Survival

Background

Surgical treatment of patients with glioblastoma affecting motor eloquent brain regions remains critically discussed given the risk–benefit dilemma of prolonging survival at the cost of motor-functional damage. Tractography informed by navigated transcranial magnetic stimulation (nTMS-informed tractography, TIT) provides a rather robust estimate of the individual location of the corticospinal tract (CST), a highly vulnerable structure with poor functional reorganisation potential. We hypothesised that by a more comprehensive, individualised surgical decision-making using TIT, tumours in close relationship to the CST can be resected with at least equal probability of gross total resection (GTR) than less eloquently located tumours without causing significantly more gross motor function harm. Moreover, we explored whether the completeness of TIT-aided resection translates to longer survival.

Methods

A total of 61 patients (median age 63 years, m = 34) with primary glioblastoma neighbouring or involving the CST were operated on between 2010 and 2015. TIT was performed to inform surgical planning in 35 of the patients (group T; vs. 26 control patients). To achieve largely unconfounded group comparisons for each co-primary outcome (i.e., gross-motor functional worsening, GTR, survival), (i) uni- and multivariate regression analyses were performed to identify features of optimal outcome prediction; (ii), optimal propensity score matching (PSM) was applied to balance those features pairwise across groups, followed by (iii) pairwise group comparison.

Results

Patients in group T featured a significantly higher lesion-CST overlap compared to controls (8.7 ± 10.7% vs. 3.8 ± 5.7%; p = 0.022). The frequency of gross motor worsening was higher in group T, albeit non-significant (n = 5/35 vs. n = 0/26; p = 0.108). PSM-based paired-sample comparison, controlling for the confounders of preoperative tumour volume and vicinity to the delicate vasculature of the insula, showed higher GTR rates in group T (77% vs. 69%; p = 0.025), particularly in patients with a priori intended GTR (87% vs. 78%; p = 0.003). This translates into a prolonged PFS in the same PSM subgroup (8.9 vs. 5.8 months; p = 0.03), with GTR representing the strongest predictor of PFS (p = 0.001) and OS (p = 0.0003) overall.

Conclusion

The benefit of TIT-aided GTR appears to overcome the drawbacks of potentially elevated motor functional risk in motor eloquent tumour localisation, leading to prolonged survival of patients with primary glioblastoma close to the CST.

Layer-specific activation in human primary somatosensory cortex during tactile temporal prediction error processing

The human brain continuously generates predictions of incoming sensory input and calculates corresponding prediction errors from the perceived inputs to update internal predictions. In human primary somatosensory cortex (area 3b), different cortical layers are involved in receiving the sensory input and generation of error signals. It remains unknown, however, how the layers in the human area 3b contribute to the temporal prediction error processing. To investigate prediction error representation in the area 3b across layers, we acquired layer-specific functional magnetic resonance imaging (fMRI) data at 7T from human area 3b during a task of index finger poking with no-delay, short-delay and long-delay touching sequences. We demonstrate that all three tasks increased activity in both superficial and deep layers of area 3b compared to the random sensory input. The fMRI signal was differentially modulated solely in the deep layers rather than the superficial layers of area 3b by the delay time. Compared with the no-delay stimuli, activity was greater in the deep layers of area 3b during the short-delay stimuli but lower during the long-delay stimuli. This difference activity features in the superficial and deep layers suggest distinct functional contributions of area 3b layers to tactile temporal prediction error processing. The functional segregation in area 3b across layers may reflect that the excitatory and inhibitory interplay in the sensory cortex contributions to flexible communication between cortical layers or between cortical areas.