A massive 7T fMRI dataset to bridge cognitive neuroscience and artificial intelligence.

Extensive sampling of neural activity during rich cognitive phenomena is critical for robust understanding of brain function. Here we present the Natural Scenes Dataset (NSD), in which high-resolution functional magnetic resonance imaging responses to tens of thousands of richly annotated natural scenes were measured while participants performed a continuous recognition task. To optimize data quality, we developed and applied novel estimation and denoising techniques. Simple visual inspections of the NSD data reveal clear representational transformations along the ventral visual pathway. Further exemplifying the inferential power of the dataset, we used NSD to build and train deep neural network models that predict brain activity more accurately than state-of-the-art models from computer vision. NSD also includes substantial resting-state and diffusion data, enabling network neuroscience perspectives to constrain and enhance models of perception and memory. Given its unprecedented scale, quality and breadth, NSD opens new avenues of inquiry in cognitive neuroscience and artificial intelligence.

Generative Feedback Explains Distinct Brain Activity Codes for Seen and Mental Images.

The relationship between mental imagery and vision is a long-standing problem in neuroscience. Currently, it is not known whether differences between the activity evoked during vision and reinstated during imagery reflect different codes for seen and mental images. To address this problem, we modeled mental imagery in the human brain as feedback in a hierarchical generative network. Such networks synthesize images by feeding abstract representations from higher to lower levels of the network hierarchy. When higher processing levels are less sensitive to stimulus variation than lower processing levels, as in the human brain, activity in low-level visual areas should encode variation in mental images with less precision than seen images. To test this prediction, we conducted an fMRI experiment in which subjects imagined and then viewed hundreds of spatially varying naturalistic stimuli. To analyze these data, we developed imagery-encoding models. These models accurately predicted brain responses to imagined stimuli and enabled accurate decoding of their position and content. They also allowed us to compare, for every voxel, tuning to seen and imagined spatial frequencies, as well as the location and size of receptive fields in visual and imagined space. We confirmed our prediction, showing that, in low-level visual areas, imagined spatial frequencies in individual voxels are reduced relative to seen spatial frequencies and that receptive fields in imagined space are larger than in visual space. These findings reveal distinct codes for seen and mental images and link mental imagery to the computational abilities of generative networks.

Reproducibility of the Structural Brain Connectome Derived from Diffusion Tensor Imaging.

RATIONALE: Disruptions of brain anatomical connectivity are believed to play a central role in several neurological and psychiatric illnesses. The structural brain connectome is typically derived from diffusion tensor imaging (DTI), which may be influenced by methodological factors related to signal processing, MRI scanners and biophysical properties of neuroanatomical regions. In this study, we evaluated how these variables affect the reproducibility of the structural connectome. METHODS: Twenty healthy adults underwent 3 MRI scanning sessions (twice in the same MRI scanner and a third time in a different scanner unit) within a short period of time. The scanning sessions included similar T1 weighted and DTI sequences. Deterministic or probabilistic tractography was performed to assess link weight based on the number of fibers connecting gray matter regions of interest (ROI). Link weight and graph theory network measures were calculated and reproducibility was assessed through intra-class correlation coefficients, assuming each scanning session as a rater. RESULTS: Connectome reproducibility was higher with data from the same scanner. The probabilistic approach yielded larger reproducibility, while the individual variation in the number of tracked fibers from deterministic tractography was negatively associated with reproducibility. Links connecting larger and anatomically closer ROIs demonstrated higher reproducibility. In general, graph theory measures demonstrated high reproducibility across scanning sessions. DISCUSSION: Anatomical factors and tractography approaches can influence the reproducibility of the structural connectome and should be factored in the interpretation of future studies. Our results demonstrate that connectome mapping is a largely reproducible technique, particularly as it relates to the geometry of network architecture measured by graph theory methods.

Oscillating Square Wave Transcranial Direct Current Stimulation (tDCS) Delivered During Slow Wave Sleep Does Not Improve Declarative Memory More Than Sham: A Randomized Sham Controlled Crossover Study.

BACKGROUND: A 2006 trial in healthy medical students found that anodal slow oscillating tDCS delivered bi-frontally during slow wave sleep had an enhancing effect in declarative, but not procedural memory. Although there have been supporting animal studies, and similar findings in pathological groups, this study has not been replicated, or refuted, in the intervening years. We therefore tested these earlier results for replication using similar methods with the exception of current waveform (square in our study, nearly sinusoidal in the original). OBJECTIVE/HYPOTHESIS: Our objective was to test the findings of a 2006 trial suggesting bi-frontal anodal tDCS during slow wave sleep enhances declarative memory. METHODS: Twelve students (mean age 25, 9 women) free of medical problems underwent two testing conditions (active, sham) in a randomized counterbalanced fashion. Active stimulation consisted of oscillating square wave tDCS delivered during early Non-Rapid Eye Movement (NREM) sleep. The sham condition consisted of setting-up the tDCS device and electrodes, but not turning it on during sleep. tDCS was delivered bi-frontally with anodes placed at F3/F4, and cathodes placed at mastoids. Current density was 0.517 mA/cm(2), and oscillated between zero and maximal current at a frequency of 0.75 Hz. Stimulation occurred during five-five minute blocks with 1-min inter-block intervals (25 min total stimulation). The primary outcomes were both declarative memory consolidation measured by a paired word association test (PWA), and non-declarative memory, measured by a non-dominant finger-tapping test (FTT). We also recorded and analyzed sleep EEG. RESULTS: There was no difference in the number of paired word associations remembered before compared to after sleep [(active = 3.1 ± 3.0 SD more associations) (sham = 3.8 ± 3.1 SD more associations)]. Finger tapping improved, (non-significantly) following active stimulation [(3.6 ± 2.7 SD correctly typed sequences) compared to sham stimulation (2.3 ± 2.2 SD correctly typed sequences)]. CONCLUSION: In this study, we failed to find improvements in declarative or performance memory and could not replicate an earlier study using nearly identical settings. Specifically we failed to find a beneficial effect on either overnight declarative or non-declarative memory consolidation via square-wave oscillating tDCS intervention applied bi-frontally during early NREM sleep. It is unclear if the morphology of the tDCS pulse is critical in any memory related improvements.