A voxel-wise encoding model for early visual areas decodes mental images of remembered scenes

Reconstructing faces from fMRI patterns using deep generative neural networks

Although distinct categories are reliably decoded from fMRI brain responses, it has proved more difficult to distinguish visually similar inputs, such as different faces. Here, we apply a recently developed deep learning system to reconstruct face images from human fMRI. We trained a variational auto-encoder (VAE) neural network using a GAN (Generative Adversarial Network) unsupervised procedure over a large data set of celebrity faces. The auto-encoder latent space provides a meaningful, topologically organized 1024-dimensional description of each image. We then presented several thousand faces to human subjects, and learned a simple linear mapping between the multi-voxel fMRI activation patterns and the 1024 latent dimensions. Finally, we applied this mapping to novel test images, translating fMRI patterns into VAE latent codes, and codes into face reconstructions. The system not only performed robust pairwise decoding (>95% correct), but also accurate gender classification, and even decoded which face was imagined, rather than seen.

Fragmented ambiguous objects: Stimuli with stable low-level features for object recognition tasks

Visual object recognition is a complex skill that relies on the interaction of many spatially distinct and specialized visual areas in the human brain. One tool that can help us better understand these specializations and interactions is a set of visual stimuli that do not differ along low-level dimensions (e.g., orientation, contrast) but do differ along high-level dimensions, such as whether a real-world object can be detected. The present work creates a set of line segment-based images that are matched for luminance, contrast, and orientation distribution (both for single elements and for pair-wise combinations) but result in a range of object and non-object percepts. Image generation started with images of isolated objects taken from publicly available databases and then progressed through 3-stages: a computer algorithm generating 718 candidate images, expert observers selecting 217 for further consideration, and naïve observers performing final ratings. This process identified a set of 100 images that all have the same low-level properties but cover a range of recognizability (proportion of naïve observers (N = 120) who indicated that the stimulus "contained a known object") and semantic stability (consistency across the categories of living, non-living/manipulable, and non-living/non-manipulable when the same observers named "known" objects). Stimuli are available at https://github.com/caolman/FAOT.git.

Decoding the contents and strength of imagery before volitional engagement

Is it possible to predict the freely chosen content of voluntary imagery from prior neural signals? Here we show that the content and strength of future voluntary imagery can be decoded from activity patterns in visual and frontal areas well before participants engage in voluntary imagery. Participants freely chose which of two images to imagine. Using functional magnetic resonance (fMRI) and multi-voxel pattern analysis, we decoded imagery content as far as 11 seconds before the voluntary decision, in visual, frontal and subcortical areas. Decoding in visual areas in addition to perception-imagery generalization suggested that predictive patterns correspond to visual representations. Importantly, activity patterns in the primary visual cortex (V1) from before the decision, predicted future imagery vividness. Our results suggest that the contents and strength of mental imagery are influenced by sensory-like neural representations that emerge spontaneously before volition.

Disparity level identification using the voxel-wise Gabor model of fMRI data

Perceiving disparities is the intuitive basis for our understanding of the physical world. Although many electrophysiology studies have revealed the disparity-tuning characteristics of the neurons in the visual areas of the macaque brain, neuron population responses to disparity processing have seldom been investigated. Many disparity studies using functional magnetic resonance imaging (fMRI) have revealed the disparity-selective visual areas in the human brain. However, it is unclear how to characterize neuron population disparity-tuning responses using fMRI technique. In the present study, we constructed three voxel-wise encoding Gabor models to predict the voxel responses to novel disparity levels and used a decoding method to identify the new disparity levels from population responses in the cortex. Among the three encoding models, the fine-coarse model (FCM) that used fine/coarse disparities to fit the voxel responses to disparities outperformed the single model and uncrossed-crossed model. Moreover, the FCM demonstrated high accuracy in predicting voxel responses in V3A complex and high accuracy in identifying novel disparities from responses in V3A complex. Our results suggest that the FCM can better characterize the voxel responses to disparities than the other two models and V3A complex is a critical visual area for representing disparity information.

The Somatotopy of Mental Tactile Imagery

To what degree mental imagery (MI) bears on the same neuronal processes as perception has been a central question in the neurophysiological study of imagery. Sensory-recruitment models suggest that imagery of sensory material heavily relies on the involvement of sensory cortices. Empirical evidence mainly stems from the study of visual imagery and suggests that it depends on the mentally imagined material whether hierarchically lower regions are recruited. However, evidence from other modalities is necessary to infer generalized principles. In this fMRI study we used the somatotopic organization of the primary somatosensory cortex (SI) to test in how far MI of tactile sensations activates topographically sensory brain areas. Participants (N = 19) either perceived or imagined vibrotactile stimuli on their left or right thumbs or big toes. The direct comparison to a corresponding perception condition revealed that SI was somatotopically recruited during imagery. While stimulus driven bottom-up processing induced activity throughout all SI subareas, i.e., BA1, BA3a, BA3b, and BA2 defined by probabilistic cytoarchitectonic maps, top-down recruitment during imagery was limited to the hierarchically highest subarea BA2.

Reconstructing imagined letters from early visual cortex reveals tight topographic correspondence between visual mental imagery and perception

Visual mental imagery is the quasi-perceptual experience of “seeing in the mind’s eye”. While a tight correspondence between imagery and perception in terms of subjective experience is well established, their correspondence in terms of neural representations remains insufficiently understood. In the present study, we exploit the high spatial resolution of functional magnetic resonance imaging (fMRI) at 7T, the retinotopic organization of early visual cortex, and machine-learning techniques to investigate whether visual imagery of letter shapes preserves the topographic organization of perceived shapes. Sub-millimeter resolution fMRI images were obtained from early visual cortex in six subjects performing visual imagery of four different letter shapes. Predictions of imagery voxel activation patterns based on a population receptive field-encoding model and physical letter stimuli provided first evidence in favor of detailed topographic organization. Subsequent visual field reconstructions of imagery data based on the inversion of the encoding model further showed that visual imagery preserves the geometric profile of letter shapes. These results open new avenues for decoding, as we show that a denoising autoencoder can be used to pretrain a classifier purely based on perceptual data before fine-tuning it on imagery data. Finally, we show that the autoencoder can project imagery-related voxel activations onto their perceptual counterpart allowing for visually recognizable reconstructions even at the single-trial level. The latter may eventually be utilized for the development of content-based BCI letter-speller systems.

Deep image reconstruction from human brain activity

The mental contents of perception and imagery are thought to be encoded in hierarchical representations in the brain, but previous attempts to visualize perceptual contents have failed to capitalize on multiple levels of the hierarchy, leaving it challenging to reconstruct internal imagery. Recent work showed that visual cortical activity measured by functional magnetic resonance imaging (fMRI) can be decoded (translated) into the hierarchical features of a pre-trained deep neural network (DNN) for the same input image, providing a way to make use of the information from hierarchical visual features. Here, we present a novel image reconstruction method, in which the pixel values of an image are optimized to make its DNN features similar to those decoded from human brain activity at multiple layers. We found that our method was able to reliably produce reconstructions that resembled the viewed natural images. A natural image prior introduced by a deep generator neural network effectively rendered semantically meaningful details to the reconstructions. Human judgment of the reconstructions supported the effectiveness of combining multiple DNN layers to enhance the visual quality of generated images. While our model was solely trained with natural images, it successfully generalized to artificial shapes, indicating that our model was not simply matching to exemplars. The same analysis applied to mental imagery demonstrated rudimentary reconstructions of the subjective content. Our results suggest that our method can effectively combine hierarchical neural representations to reconstruct perceptual and subjective images, providing a new window into the internal contents of the brain.

A Varying Role for Abstraction in Models of Category Learning Constructed from Neural Representations in Early Visual Cortex

The human capacity for visual categorization is core to how we make sense of the visible world. Although a substantive body of research in cognitive neuroscience has localized this capacity to regions of human visual cortex, relatively few studies have investigated the role of abstraction in how representations for novel object categories are constructed from the neural representation of stimulus dimensions. Using human fMRI coupled with formal modeling of observer behavior, we assess a wide range of categorization models that vary in their level of abstraction from collections of subprototypes to representations of individual exemplars. The category learning tasks range from simple linear and unidimensional category rules to complex crisscross rules that require a nonlinear combination of multiple dimensions. We show that models based on neural responses in primary visual cortex favor a variable, but often limited, extent of abstraction in the construction of representations for novel categories, which differ in degree across tasks and individuals.

Predicting eye movement patterns from fMRI responses to natural scenes

Eye tracking has long been used to measure overt spatial attention, and computational models of spatial attention reliably predict eye movements to natural images. However, researchers lack techniques to noninvasively access spatial representations in the human brain that guide eye movements. Here, we use functional magnetic resonance imaging (fMRI) to predict eye movement patterns from reconstructed spatial representations evoked by natural scenes. First, we reconstruct fixation maps to directly predict eye movement patterns from fMRI activity. Next, we use a model-based decoding pipeline that aligns fMRI activity to deep convolutional neural network activity to reconstruct spatial priority maps and predict eye movements in a zero-shot fashion. We predict human eye movement patterns from fMRI responses to natural scenes, provide evidence that visual representations of scenes and objects map onto neural representations that predict eye movements, and find a novel three-way link between brain activity, deep neural network models, and behavior.

Visual imagery of faces and cars in face-selective visual areas

Neuroimaging provides a unique tool to investigate otherwise difficult-to-access mental processes like visual imagery. Prior studies support the idea that visual imagery is a top-down reinstatement of visual perception, and it is likely that this extends to object processing. Here we use functional MRI and multi-voxel pattern analysis to ask if mental imagery of cars engages the fusiform face area, similar to what is found during perception. We test only individuals who we assumed could imagine individual car models based on their above-average perceptual abilities with cars. Our results provide evidence that cars are represented differently from common objects in face-selective visual areas, at least in those with above-average car recognition ability. Moreover, pattern classifiers trained on data acquired during imagery can decode the neural response pattern acquired during perception, suggesting that the tested object categories are represented similarly during perception and visual imagery. The results suggest that, even at high-levels of visual processing, visual imagery mirrors perception to some extent, and that face-selective areas may in part support non-face object imagery.

Exploring the functional nature of synaesthetic colour: Dissociations from colour perception and imagery

Individuals with grapheme-colour synaesthesia experience anomalous colours when reading achromatic text. These unusual experiences have been said to resemble 'normal' colour perception or colour imagery, but studying the nature of synaesthesia remains difficult. In the present study, we report novel evidence that synaesthetic colour impacts conscious vision in a way that is different from both colour perception and imagery. Presenting 'normal' colour prior to binocular rivalry induces a location-dependent suppressive bias reflecting local habituation. By contrast, a grapheme that evokes synaesthetic colour induces a facilitatory bias reflecting priming that is not constrained to the inducing grapheme's location. This priming does not occur in non-synaesthetes and does not result from response bias. It is sensitive to diversion of visual attention away from the grapheme, but resistant to sensory perturbation, reflecting a reliance on cognitive rather than sensory mechanisms. Whereas colour imagery in non-synaesthetes causes local priming that relies on the locus of imagined colour, imagery in synaesthetes caused global priming not dependent on the locus of imagery. These data suggest a unique psychophysical profile of high-level colour processing in synaesthetes. Our novel findings and method will be critical to testing theories of synaesthesia and visual awareness.

Decoding natural images from evoked brain activities using encoding models with invertible mapping

Recent studies have built encoding models in the early visual cortex, and reliable mappings have been made between the low-level visual features of stimuli and brain activities. However, these mappings are irreversible, so that the features cannot be directly decoded. To solve this problem, we designed a sparse framework-based encoding model that predicted brain activities from a complete feature representation. Moreover, according to the distribution and activation rules of neurons in the primary visual cortex (V1), three key transformations were introduced into the basic feature to improve the model performance. In this setting, the mapping was simple enough that it could be inverted using a closed-form formula. Using this mapping, we designed a hybrid identification method based on the support vector machine (SVM), and tested it on a published functional magnetic resonance imaging (fMRI) dataset. The experiments confirmed the rationality of our encoding model, and the identification accuracies for 2 subjects increased from 92% and 72% to 98% and 92% with the chance level only 0.8%.

fMRI-based Multivariate Pattern Analyses Reveal Imagery Modality and Imagery Content Specific Representations in Primary Somatosensory, Motor and Auditory Cortices

Previous studies have shown that the early visual cortex contains content-specific representations of stimuli during visual imagery, and that these representational patterns of imagery content have a perceptual basis. To date, there is little evidence for the presence of a similar organization in the auditory and tactile domains. Using fMRI-based multivariate pattern analyses we showed that primary somatosensory, auditory, motor, and visual cortices are discriminative for imagery of touch versus sound. In the somatosensory, motor and visual cortices the imagery modality discriminative patterns were similar to perception modality discriminative patterns, suggesting that top-down modulations in these regions rely on similar neural representations as bottom-up perceptual processes. Moreover, we found evidence for content-specific representations of the stimuli during auditory imagery in the primary somatosensory and primary motor cortices. Both the imagined emotions and the imagined identities of the auditory stimuli could be successfully classified in these regions.

Human V4 Activity Patterns Predict Behavioral Performance in Imagery of Object Color

Color is special among basic visual features in that it can form a defining part of objects that are engrained in our memory. Whereas most neuroimaging research on human color vision has focused on responses related to external stimulation, the present study investigated how sensory-driven color vision is linked to subjective color perception induced by object imagery. We recorded fMRI activity in male and female volunteers during viewing of abstract color stimuli that were red, green, or yellow in half of the runs. In the other half we asked them to produce mental images of colored, meaningful objects (such as tomato, grapes, banana) corresponding to the same three color categories. Although physically presented color could be decoded from all retinotopically mapped visual areas, only hV4 allowed predicting colors of imagined objects when classifiers were trained on responses to physical colors. Importantly, only neural signal in hV4 was predictive of behavioral performance in the color judgment task on a trial-by-trial basis. The commonality between neural representations of sensory-driven and imagined object color and the behavioral link to neural representations in hV4 identifies area hV4 as a perceptual hub linking externally triggered color vision with color in self-generated object imagery.SIGNIFICANCE STATEMENT Humans experience color not only when visually exploring the outside world, but also in the absence of visual input, for example when remembering, dreaming, and during imagery. It is not known where neural codes for sensory-driven and internally generated hue converge. In the current study we evoked matching subjective color percepts, one driven by physically presented color stimuli, the other by internally generated color imagery. This allowed us to identify area hV4 as the only site where neural codes of corresponding subjective color perception converged regardless of its origin. Color codes in hV4 also predicted behavioral performance in an imagery task, suggesting it forms a perceptual hub for color perception.

Imaging at ultrahigh magnetic fields: History, challenges, and solutions

Following early efforts in applying nuclear magnetic resonance (NMR) spectroscopy to study biological processes in intact systems, and particularly since the introduction of 4 T human scanners circa 1990, rapid progress was made in imaging and spectroscopy studies of humans at 4 T and animal models at 9.4 T, leading to the introduction of 7 T and higher magnetic fields for human investigation at about the turn of the century. Work conducted on these platforms has provided numerous technological solutions to challenges posed at these ultrahigh fields, and demonstrated the existence of significant advantages in signal-to-noise ratio and biological information content. Primary difference from lower fields is the deviation from the near field regime at the radiofrequencies (RF) corresponding to hydrogen resonance conditions. At such ultrahigh fields, the RF is characterized by attenuated traveling waves in the human body, which leads to image non-uniformities for a given sample-coil configuration because of destructive and constructive interferences. These non-uniformities were initially considered detrimental to progress of imaging at high field strengths. However, they are advantageous for parallel imaging in signal reception and transmission, two critical technologies that account, to a large extend, for the success of ultrahigh fields. With these technologies and improvements in instrumentation and imaging methods, today ultrahigh fields have provided unprecedented gains in imaging of brain function and anatomy, and started to make inroads into investigation of the human torso and extremities. As extensive as they are, these gains still constitute a prelude to what is to come given the increasingly larger effort committed to ultrahigh field research and development of ever better instrumentation and techniques.

Domain-specific reports of visual imagery vividness are not related to perceptual expertise

Just as people vary in their perceptual expertise with a given domain, they also vary in their abilities to imagine objects. Visual imagery and perception share common mechanisms. However, it is unclear whether domain-specific expertise is relevant to visual imagery. Although the vividness of visual imagery is typically measured as a domain-general construct, a component of this vividness may be domain-specific. For example, individuals who have gained perceptual expertise with a specific domain might experience clearer mental images within this domain. Here we investigated whether perceptual expertise for cars relates to visual imagery vividness in the same domain, by assessing the correlations between a widely used domain-general measure of visual imagery vividness (the Vividness of Visual Imagery Questionnaire; Marks in British Journal of Psychology, 64, 17-24, 1973), a new measure of visual imagery vividness specific to cars, and behavioral tests of car expertise. We found that domain-specific imagery relates most strongly to general imagery vividness and less strongly to self-reported expertise, while it does not relate to perceptual or semantic expertise.

Linking numbers to perceptions and experiences: Why we need transdisciplinary mixed-methods combining neurophysiological and qualitative data

Today, more and more problems that scientists need to tackle are complex problems. Many examples of these can be found in the health sciences, medicine and ecology. Typical features of complex problems are that they cannot be studied by one discipline and that they need to take into account subjective data as well as objective data. Two promising responses to deal with complex problems are Transdisciplinary and Mixed Method approaches. However, there is still a lacuna to fill, with transdisciplinary studies bridging the social sciences and biomedical sciences. More specifically, we need more and better studies that combine qualitative data about subjective experiences, perception and so on with objective, quantitative, neurophysiological data. We believe that the combination of qualitative and neurophysiological data is a good example of what we would like to call transdisciplinary mixed methods. In this article, we aim to explore the opportunities of transdisciplinary mixed-methods studies in which qualitative and neurophysiological data are used. We give a brief overview of what is characteristic for this kind of studies and illustrate this with examples; we point out strengths and limitations and propose an agenda for the future. We conclude that transdisciplinary mixed-methods studies in which qualitative and neurophysiological data are used have the potential to improve our knowledge about complex problems. A main obstacle seems to be that most scientists from the biomedical sciences are not familiar with the (qualitative) methods from the social sciences and vice versa. To end this ‘clash of paradigms’™, we urgently need to cultivate transdisciplinary thinking.

What is feasible with imaging human brain function and connectivity using functional magnetic resonance imaging

When we consider all of the methods we employ to detect brain function, from electrophysiology to optical techniques to functional magnetic resonance imaging (fMRI), we do not really have a 'golden technique' that meets all of the needs for studying the brain. We have methods, each of which has significant limitations but provide often complimentary information. Clearly, there are many questions that need to be answered about fMRI, which unlike other methods, allows us to study the human brain. However, there are also extraordinary accomplishments or demonstration of the feasibility of reaching new and previously unexpected scales of function in the human brain. This article reviews some of the work we have pursued, often with extensive collaborations with other co-workers, towards understanding the underlying mechanisms of the methodology, defining its limitations, and developing solutions to advance it. No doubt, our knowledge of human brain function has vastly expanded since the introduction of fMRI. However, methods and instrumentation in this dynamic field have evolved to a state that discoveries about the human brain based on fMRI principles, together with information garnered at a much finer spatial and temporal scale through other methods, are poised to significantly accelerate in the next decade.This article is part of the themed issue 'Interpreting BOLD: a dialogue between cognitive and cellular neuroscience'.

Effects of spatial fMRI resolution on the classification of naturalistic movies

Studies involving multivariate pattern analysis (MVPA) of BOLD fMRI data generally attribute the success of the information-theoretic approach to BOLD signal contrast on the fine spatial scale of millimeters facilitating the classification or decoding of perceptual stimuli. However, to date MVPA studies that have actually explored fMRI resolutions at less than 2 mm voxel size are rare and limited to small sets of unnatural stimuli (like visual gratings) as well as specific sub-regions of the brain, notably the primary somatosensory cortices. To investigate what spatial scale best supports high information extraction under more general conditions this study combined naturalistic movie stimuli with high-resolution fMRI at 7 T and linear discriminant analysis (LDA) of global and local BOLD signal patterns. Contrary to predictions, LDA and similar classifiers reached a maximum in classification accuracy (CA) at a smoothed resolution close to 3 mm, well above the 1.2 mm voxel size of the fMRI acquisition. Maximal CAs around 90% were contingent upon global fMRI signal patterns comprising 4 k-16 k of the most reactive voxels distributed sparsely throughout the occipital and ventro-temporal cortices. A Searchlight analysis of local fMRI patterns largely confirmed the global results, but also revealed a small subset of brain regions in early visual cortex showing limited increases in CA with higher resolution. Principal component analysis of the global and local fMRI signal patterns suggested that reproducible neuronal contributions were spatially auto-correlated and smooth, while other components of higher spatial frequency were likely related to physiological noise and responsible for the reduced CA at higher resolution. Systematic differences between experiments and subjects suggested that higher CA was significantly correlated with more consistent behavior revealed by eye tracking. Thus, the optimal resolution of fMRI data for MVPA was mainly limited by physiological noise of high spatial frequency as well as behavioral (in-)consistency.

Laminar fMRI and computational theories of brain function

Recently developed methods for functional MRI at the resolution of cortical layers (laminar fMRI) offer a novel window into neurophysiological mechanisms of cortical activity. Beyond physiology, laminar fMRI also offers an unprecedented opportunity to test influential theories of brain function. Specifically, hierarchical Bayesian theories of brain function, such as predictive coding, assign specific computational roles to different cortical layers. Combined with computational models, laminar fMRI offers a unique opportunity to test these proposals noninvasively in humans. This review provides a brief overview of predictive coding and related hierarchical Bayesian theories, summarises their predictions with regard to layered cortical computations, examines how these predictions could be tested by laminar fMRI, and considers methodological challenges. We conclude by discussing the potential of laminar fMRI for clinically useful computational assays of layer-specific information processing.

The blind mind: No sensory visual imagery in aphantasia

For most people the use of visual imagery is pervasive in daily life, but for a small group of people the experience of visual imagery is entirely unknown. Research based on subjective phenomenology indicates that otherwise healthy people can completely lack the experience of visual imagery, a condition now referred to as aphantasia. As congenital aphantasia has thus far been based on subjective reports, it remains unclear whether individuals are really unable to imagine visually, or if they have very poor metacognition - they have images in their mind, but are blind to them. Here we measured sensory imagery in subjectively self-diagnosed aphantasics, using the binocular rivalry paradigm, as well as measuring their self-rated object and spatial imagery with multiple questionnaires (VVIQ, SUIS and OSIQ). Unlike, the general population, experimentally naive aphantasics showed almost no imagery-based rivalry priming. Aphantasic participants' self-rated visual object imagery was significantly below average, however their spatial imagery scores were above average. These data suggest that aphantasia is a condition involving a lack of sensory and phenomenal imagery, and not a lack of metacognition. The possible underlying neurological cause of aphantasia is discussed as well as future research directions.

Reading Minds: Brain-Decoding Scientists Move Closer to Discovering the Keys to Unlock the Brain

When you see or think about an object, your brain engages in a unique pattern of activity tied specifically to that object. That's how you know a cat is a cat, and not a dog or a house or a cloud. Using functional magnetic resonance imaging (fMRI) and other techniques, scientists are not only able to measure those activity patterns but are also deciphering what each pattern means. Essentially, they are beginning to read minds.

Decoupling of BOLD amplitude and pattern classification of orientation-selective activity in human visual cortex

Multivariate pattern analysis (MVPA) of fMRI data has allowed the investigation of neural representations of stimuli on the basis of distributed patterns of activity within a brain region, independently from overall brain activity. For instance, several studies on early visual cortex have reported reliable MVPA decoding of the identity of a stimulus representation that was kept in working memory or internally generated, despite the fact that the overall BOLD response was low or even at baseline levels. Here we ask how it is possible that reliable stimulus information can be decoded from early visual cortex even when the overall BOLD signal remains low. We reanalyzed a data set in which human participants (N = 24) imagined or kept in working memory an oriented visual grating. We divided voxels from V1, V2, and V3 into groups based on orientation preference, and compared the time course of mean BOLD responses to preferred and non-preferred orientations with the time course of the multivariate decoding performance. Decoding accuracy related to a numerically small, but reliable univariate difference in the mean BOLD response to preferred and non-preferred stimuli. The time course of the difference in BOLD responses to preferred and non-preferred orientations was highly similar to the time course of the multivariate pattern classification accuracy. The reliability of the classification strongly correlated with the magnitude of differences in BOLD signal between preferred and non-preferred stimuli. These activity differences were small compared to the large overall BOLD modulations. This suggests that a substantial part of the task-related BOLD response to visual stimulation might not be stimulus-specific. Rather, stimulus-evoked BOLD signals in early visual cortex during a task context may be an amalgam of small stimulus-specific responses and large task-related but non-stimulus-specific responses. The latter are not evident during the maintenance or internal generation of stimulus representations, but provide an explanation of how reliable stimulus information can be decoded from early visual cortex even though its overall BOLD signal remains low.

Decoding naturalistic experiences from human brain activity via distributed representations of words

Natural visual scenes induce rich perceptual experiences that are highly diverse from scene to scene and from person to person. Here, we propose a new framework for decoding such experiences using a distributed representation of words. We used functional magnetic resonance imaging (fMRI) to measure brain activity evoked by natural movie scenes. Then, we constructed a high-dimensional feature space of perceptual experiences using skip-gram, a state-of-the-art distributed word embedding model. We built a decoder that associates brain activity with perceptual experiences via the distributed word representation. The decoder successfully estimated perceptual contents consistent with the scene descriptions by multiple annotators. Our results illustrate three advantages of our decoding framework: (1) three types of perceptual contents could be decoded in the form of nouns (objects), verbs (actions), and adjectives (impressions) contained in 10,000 vocabulary words; (2) despite using such a large vocabulary, we could decode novel words that were absent in the datasets to train the decoder; and (3) the inter-individual variability of the decoded contents co-varied with that of the contents of scene descriptions. These findings suggest that our decoding framework can recover diverse aspects of perceptual experiences in naturalistic situations and could be useful in various scientific and practical applications.

Convolutional neural network-based encoding and decoding of visual object recognition in space and time

Representations learned by deep convolutional neural networks (CNNs) for object recognition are a widely investigated model of the processing hierarchy in the human visual system. Using functional magnetic resonance imaging, CNN representations of visual stimuli have previously been shown to correspond to processing stages in the ventral and dorsal streams of the visual system. Whether this correspondence between models and brain signals also holds for activity acquired at high temporal resolution has been explored less exhaustively. Here, we addressed this question by combining CNN-based encoding models with magnetoencephalography (MEG). Human participants passively viewed 1,000 images of objects while MEG signals were acquired. We modelled their high temporal resolution source-reconstructed cortical activity with CNNs, and observed a feed-forward sweep across the visual hierarchy between 75 and 200 ms after stimulus onset. This spatiotemporal cascade was captured by the network layer representations, where the increasingly abstract stimulus representation in the hierarchical network model was reflected in different parts of the visual cortex, following the visual ventral stream. We further validated the accuracy of our encoding model by decoding stimulus identity in a left-out validation set of viewed objects, achieving state-of-the-art decoding accuracy.

The influence of low-level stimulus features on the representation of contexts, items, and their mnemonic associations

Since the earliest attempts to characterize the "receptive fields" of neurons, a central aim of many neuroscience experiments is to elucidate the information that is represented in various regions of the brain. Recent studies suggest that, in the service of memory, information is represented in the medial temporal lobe in a conjunctive or associative form with the contextual aspects of the experience being the primary factor or highest level of the conjunctive hierarchy. A critical question is whether the information that has been observed in these studies reflects notions such as a cognitive representation of context or whether the information reflects the low-level sensory differences between stimuli. We performed two functional magnetic resonance imaging experiments to address this question and we found that associative representations observed between context and item (and order) in the human brain can be highly influenced by low-level sensory differences between stimuli. Our results place clear constraints on the experimental design of studies that aim to investigate the representation of contexts and items during performance of associative memory tasks. Moreover, our results raise interesting theoretical questions regarding the disambiguation of memory-related representations from processing-related representations.

Object Recognition in Mental Representations: Directions for Exploring Diagnostic Features through Visual Mental Imagery

One of the fundamental goals of object recognition research is to understand how a cognitive representation produced from the output of filtered and transformed sensory information facilitates efficient viewer behavior. Given that mental imagery strongly resembles perceptual processes in both cortical regions and subjective visual qualities, it is reasonable to question whether mental imagery facilitates cognition in a manner similar to that of perceptual viewing: via the detection and recognition of distinguishing features. Categorizing the feature content of mental imagery holds potential as a reverse pathway by which to identify the components of a visual stimulus which are most critical for the creation and retrieval of a visual representation. This review will examine the likelihood that the information represented in visual mental imagery reflects distinctive object features thought to facilitate efficient object categorization and recognition during perceptual viewing. If it is the case that these representational features resemble their sensory counterparts in both spatial and semantic qualities, they may well be accessible through mental imagery as evaluated through current investigative techniques. In this review, methods applied to mental imagery research and their findings are reviewed and evaluated for their efficiency in accessing internal representations, and implications for identifying diagnostic features are discussed. An argument is made for the benefits of combining mental imagery assessment methods with diagnostic feature research to advance the understanding of visual perceptive processes, with suggestions for avenues of future investigation.

Generic decoding of seen and imagined objects using hierarchical visual features

Object recognition is a key function in both human and machine vision. While brain decoding of seen and imagined objects has been achieved, the prediction is limited to training examples. We present a decoding approach for arbitrary objects using the machine vision principle that an object category is represented by a set of features rendered invariant through hierarchical processing. We show that visual features, including those derived from a deep convolutional neural network, can be predicted from fMRI patterns, and that greater accuracy is achieved for low-/high-level features with lower-/higher-level visual areas, respectively. Predicted features are used to identify seen/imagined object categories (extending beyond decoder training) from a set of computed features for numerous object images. Furthermore, decoding of imagined objects reveals progressive recruitment of higher-to-lower visual representations. Our results demonstrate a homology between human and machine vision and its utility for brain-based information retrieval.

Sensory Substitution and Multimodal Mental Imagery

Many philosophers use findings about sensory substitution devices in the grand debate about how we should individuate the senses. The big question is this: Is "vision" assisted by (tactile) sensory substitution really vision? Or is it tactile perception? Or some sui generis novel form of perception? My claim is that sensory substitution assisted "vision" is neither vision nor tactile perception, because it is not perception at all. It is mental imagery: visual mental imagery triggered by tactile sensory stimulation. But it is a special form of mental imagery that is triggered by corresponding sensory stimulation in a different sense modality, which I call "multimodal mental imagery."

Preparatory attention in visual cortex

Top-down attention is the mechanism that allows us to selectively process goal-relevant aspects of a scene while ignoring irrelevant aspects. A large body of research has characterized the effects of attention on neural activity evoked by a visual stimulus. However, attention also includes a preparatory phase before stimulus onset in which the attended dimension is internally represented. Here, we review neurophysiological, functional magnetic resonance imaging, magnetoencephalography, electroencephalography, and transcranial magnetic stimulation (TMS) studies investigating the neural basis of preparatory attention, both when attention is directed to a location in space and when it is directed to nonspatial stimulus attributes (content-based attention) ranging from low-level features to object categories. Results show that both spatial and content-based attention lead to increased baseline activity in neural populations that selectively code for the attended attribute. TMS studies provide evidence that this preparatory activity is causally related to subsequent attentional selection and behavioral performance. Attention thus acts by preactivating selective neurons in the visual cortex before stimulus onset. This appears to be a general mechanism that can operate on multiple levels of representation. We discuss the functional relevance of this mechanism, its limitations, and its relation to working memory, imagery, and expectation. We conclude by outlining open questions and future directions.

Temporal Information Entropy of the Blood-Oxygenation Level-Dependent Signals Increases in the Activated Human Primary Visual Cortex

Time-domain analysis of blood-oxygenation level-dependent (BOLD) signals allows the identification of clusters of voxels responding to photic stimulation in primary visual cortex (V1). However, the characterization of information encoding into temporal properties of the BOLD signals of an activated cluster is poorly investigated. Here, we used Shannon entropy to determine spatial and temporal information encoding in the BOLD signal within the most strongly activated area of the human visual cortex during a hemifield photic stimulation. We determined the distribution profile of BOLD signals during epochs at rest and under stimulation within small (19-121 voxels) clusters designed to include only voxels driven by the stimulus as highly and uniformly as possible. We found consistent and significant increases (2-4% on average) in temporal information entropy during activation in contralateral but not ipsilateral V1, which was mirrored by an expected loss of spatial information entropy. These opposite changes coexisted with increases in both spatial and temporal mutual information (i.e., dependence) in contralateral V1. Thus, we showed that the first cortical stage of visual processing is characterized by a specific spatiotemporal rearrangement of intracluster BOLD responses. Our results indicate that while in the space domain BOLD maps may be incapable of capturing the functional specialization of small neuronal populations due to relatively low spatial resolution, some information encoding may still be revealed in the temporal domain by an increase of temporal information entropy.

Hierarchical Neural Representation of Dreamed Objects Revealed by Brain Decoding with Deep Neural Network Features

Dreaming is generally thought to be generated by spontaneous brain activity during sleep with patterns common to waking experience. This view is supported by a recent study demonstrating that dreamed objects can be predicted from brain activity during sleep using statistical decoders trained with stimulus-induced brain activity. However, it remains unclear whether and how visual image features associated with dreamed objects are represented in the brain. In this study, we used a deep neural network (DNN) model for object recognition as a proxy for hierarchical visual feature representation, and DNN features for dreamed objects were analyzed with brain decoding of fMRI data collected during dreaming. The decoders were first trained with stimulus-induced brain activity labeled with the feature values of the stimulus image from multiple DNN layers. The decoders were then used to decode DNN features from the dream fMRI data, and the decoded features were compared with the averaged features of each object category calculated from a large-scale image database. We found that the feature values decoded from the dream fMRI data positively correlated with those associated with dreamed object categories at mid- to high-level DNN layers. Using the decoded features, the dreamed object category could be identified at above-chance levels by matching them to the averaged features for candidate categories. The results suggest that dreaming recruits hierarchical visual feature representations associated with objects, which may support phenomenal aspects of dream experience.

Mental rotation task specifically modulates functional connectivity strength of intrinsic brain activity in low frequency domains: A maximum uncertainty linear discriminant analysis

Neuroimaging studies have highlighted that intrinsic brain activity is modified to implement task demands. However, the relation between mental rotation and intrinsic brain activity remains unclear. To answer this question, we collected functional MRI (fMRI) data from 30 healthy participants in two mental rotation task periods (1st-task state, 2nd-task state) and two rest periods before (pre-task resting state) and after the task (post-task resting state) respectively. By combining the spatial independent component analysis (ICA) and voxel-wise functional connectivity strength (FCS), we identified FCS maps of 10 brain resting state networks (RSNs) within six different bands (i.e., 0-0.05, 0.05-0.1, 0.1-0.15, 0.15-0.2, 0.2-0.25, and 0.01-0.08Hz) corresponding to the four states for each subject. The maximum uncertainty linear discriminant analysis (MLDA) method showed that the FCS within the low frequency bandwidth of 0.05-0.1Hz could effectively classify the mental rotation task state from pre-/post-task resting states but failed to discriminate the pre- and post-task resting states. Discriminative FCSs were observed in the cognitive executive-control network (central executive and attention) and the imagery-based internal mental manipulation network (default mode, primary sensorimotor, and primary visual). Imagery manipulation is a stable mental element of mental rotation, and the involvement of executive control is dependent on the degree of task familiarity. Together, the present study provides evidence that mental rotation task specifically modifies intrinsic brain activity to complement cognitive demands, which provides further insight into the neural basis of mental rotation manipulation.