The lab runs a number of projects, centered around three themes, outlined below. Many of our techniques—brain imaging, stimulation, behavioural measurements, modelling, and electrophysiology—are shared across the themes.
Tools and Fundamentals of Neuroimaging
The projects in this theme investigate physiological and informative components of functional MRI (fMRI) signals, develop and validate advanced MRI hardware for high-resolution imaging or simultaneous imaging and stimulation, and develop tools for advanced imaging, analysis of images, and modelling.
General background [UNDER CONSTRUCTION]
Functional Magnetic Resonance Imaging has been the work horse for mapping human brain functions in the past decades. In an example of the classical approach, subjects are shown are shown a cloud of randomly moving dots on a screen for 20 seconds (condition A), then a similar cloud of dots that are now coherently moving toward the same direction (condition B). Condition B evokes two psychological phenomena, the perception of moving dots or elements and the perception of coherent or global movement, whereas Condition A evokes only on of the two concepts evoked in condition B, the perception of moving elements. Under the assumption of linear summation, subtracting condition A from condition B should cancel the effects of moving elements and allow the experimenter to specifically study global motion. For mapping the brain areas responsible for the perception of global motion, the experimenter will, for each fMRI voxels (a pixel in 3D), subtract the signal acquired in condition B from the signal acquired in condition A and test whether this difference is significant. Results are then overlaid on top of the image of the brain, yielding the colorful pictures of fMRI papers have accustomed us to, with blobs indicating the “mapped” brain areas.
Although this approach was extremely useful for mapping the brain, i.e. the identification which brain area does what, advances in the achievable resolution and sensitivity of MRI and the development of analysis methods open the way for the more interesting question of how does the brain do what it does? Examples of questions addressed around this center theme in the lab... [direct to other sections here] However, the greater MRI capabilities come with their lot of challenges. The indirect relation between neural activity and measured fMRI signal did not matter much when fMRI resolution allowed only to study large swats (>3mm) of the cortex, but understanding this relation becomes crucial for studying the small-scale processes now accessible with modern MRI scanner. Next sections highlight our approaches and contributions to technological advances in MRI (subsection title) and to define the link between neural information processing and fMRI signals (Spatio-Temporal Properties / Neural and Vascular Origins of fMRI Signals section).
Physiology & Information Content of fMRI Signals [UNDER CONSTRUCTION]
Key Concepts: hemodynamic response; information content of multivoxel activity patterns, spatiotemporal dynamics of hemodynamic signals
Seminal work on the neural determinants of the fMRI signals both related synaptic and spiking activity to changes in fMRI signals (Logothetis 2002). Upon increased levels of neural activity, multiple signaling pathways induces blood vessels dilation, triggering a localized vascular response involving increased blood flow and volume in the piece of brain involved, that in order to meet the increased metabolic demands. In the paradigmatic case of shifting from a resting (no stimulation) to a high (active) level of neural activity, the increase in blood supply greatly exceeds the increase in oxygen supply. The consequent increase in oxygenation level changes the magnetic properties of the tissue and give rise to the most commonly used fMRI signal, the blood-oxygenation-level-dependent (BOLD) signal.
Temporally, the slow nature of the BOLD response owes to its vascular origin – the vascular response is delayed and sluggish relative to triggering neural. There is however animal (Uhlirova, Kilic et al. 2016) and human (Farivar, Thompson et al. 2011) evidences that the shape of the response, the hemodynamic response function, might not be completely explained by vascular dynamics. Both optogenetic stimulation of specific sub-population of neurons (Uhlirova, Kilic et al. 2016) and modulating the excitation/inhibition ratio of a neural response with targeted visual stimuli (Farivar, Thompson et al. 2011) affected the shape of the hemodynamic response. A main interest in our lab is to establish methodologies for using these differences in temporal shape in order to move from measures of the global level of neural activity within a voxel (from the amplitude of the response) to measures of the qualitative nature of a neural activation like its excitation/inhibition balance (from the shape of the response).
Spatially, patterns of BOLD activation, even at very high resolution, do not perfectly overlap with the pattern of neural activity, being heavily biased toward nearby penetrating and pial veins. That is not to say that spatial patterns of BOLD do not carry information about the underlying spatial pattern of neural activity, which itself can carry important information on neural information processing, the real interesting thing we want to learn about. A privileged approach in our lab to circumvent this issue is to move away from the precise millimeter-range localization of neural processes and rather focus on spatial patterns of BOLD relevant to information processing. This is the decoding approach to fMRI, where the capacity of the multi-voxel pattern of BOLD signal within a brain region to predict features of a visual stimuli is taken as evidence for the processing of that feature in that brain area.
The spatial and temporal characteristics of a BOLD response are intrinsically interrelated through the complex dynamics of adaptive blood flow in the vascular network. A series of experiments in our lab has shown spatiotemporal properties of BOLD signal quite difficult to study with standard approaches. A cardinal findings in our lab is that as the BOLD response unfold through time and space in the visual cortex, the neurally-relevant component (capacity to predict stimulus feature from the spatial pattern) of the response is delayed by an extra 1-2 seconds relative to the largely non-specific (to stimulus feature) BOLD response.
Farivar, R., et al. (2011). "Interocular suppression in strabismic amblyopia results in an attenuated and delayed hemodynamic response function in early visual cortex." Journal of Vision 11(14).
Logothetis, N. K. (2002). "The neural basis of the blood-oxygen-level-dependent functional magnetic resonance imaging signal." Philosophical Transactions of the Royal Society of London Series B-Biological Sciences 357(1424): 1003-1037.
Uhlirova, H., et al. (2016). "Cell type specificity of neurovascular coupling in cerebral cortex." Elife 5.
MRI RF Coils and Optimization
Through novel hardware designs, we aim to improve on four key aspects of magnetic resonance imaging (MRI): accuracy, quality, comfort, and speed. Higher accuracy and quality of images leads to more accurate and precise findings in both research and clinical settings. For instance, higher resolution images with minimized noise and heightened quality may mean earlier detection of illness, leading to higher life expectancy and quality of life; or more precise conclusions to research questions, fostering a better understanding of our brains, leading to more accurate therapeutic techniques. The comfort of a patient and the length of time spent in the scanner can greatly improve the happiness of the patient while also minimizing error introduced by movement or premature termination of scans. From a research perspective, subject happiness, again, lends to improved accuracy and quality of results, as well as higher temporal resolution of brain activity scans (functional scans).
We accomplish these goals by designing and testing radio-frequency coils, a type of specialized antenna. They offer improved accuracy and quality through better signal-to-noise ratio, higher patient comfort through more form fitting and flexible assemblies, and higher speed through large coil arrays.
Currently, our projects include combining transcranial magnetic stimulation (TMS), a method to stimulate precise areas of the brain, with functional imaging of subjects performing a visual task to investigate the instantaneous effects of TMS and how these effects permeate across the brain. A large, flexible coil array, featuring 128-channels, is currently in development and will provide leading-edge image quality and accuracy, while improving patient comfort through a flexible, size-adjustable design.
We hope to provide insight, tools, and designs that will motivate others to contribute and grow the field of diagnostic imaging hardware and techniques.
Analysis and Modelling of Structural and Functional MRI signals
Anatomical structures among experimental subjects are very different. However, neuroscientists make inference about the population based on the confounded fact that our experimental subjects represent human population. Previously, researches addressed the problem by aligning anatomical landmarks of the cortical surface and bring subject to a common template. This approach still leaves gaps to be bridged that anatomical landmarks are not perfect reflections of functional brain area variety between different people. Recently, researches have developed ways to circumvent the inconsistency of anatomical structure to functional areas. One of them is functional hyper-alignment, which brings subjects to a same template based on functional patterns. Representational similarity analysis (RSA) also helps to eliminate this inconsistency by making it possible of comparisons in different multidimensional spaces. In our lab, we are interested in this framework of moving from representing anatomical template to functional template of the group. We are moving towards models that could provide insights to the neuroimaging society.
Brain Representation and Processing of Vision
People navigate complex environments, making sense of objects, scenes, and their constancy despite changing illumination, point of view, etc. How does the brain do it? We’re trying to understand aspects of these as they pertain to the Natural 3-D world by studying how depth cues serve to represent objects, how naturalistic scenes are represented, and how some of these representations arise from early stages in visual processing.
Brain Signals During Vivid, Natural Movie Viewing
The visual system evolved to make sense of a complex natural world combining colours, shapes, motion, textures, luminance, etc., yet the most common approach in visual neuroscience is to isolate the processing mechanism of a single visual feature by measuring the subjects’ response to an artificially generated stimulus. There is rising debate in the field of visual processing that the whole is not quite the sum of its parts – essentially that the interplay of visual processes jointly results in the representation of the stimuli and if these processes are individually activated, these interactions are overlooked. The unique advantage of using naturalistic stimuli is that many processes of the visual system are being activated at any given time, providing the closest image of how the visual system works. Our lab uses vivid natural movies to 1) explore the representation of objects and scenes 2) show that subjects exhibit prototypical fine-scale spatial patterns and 3) compare the injured brain to normal templates of activity.
The challenge to studying the visual system under naturalistic conditions is that the stimulus itself is too complex to parameterize so the brain response is extremely difficult to predict and model. Instead of trying to predict how the brain will react, we may approach the problem by exploring whether different brains react the same way to a naturalistic stimulus. Brains are indeed synchronized over a large area of the cortex when viewing a movie (cite), showing that a naturalistic stimulus can illicit the same temporal patterns in different brains. This leads us to the question: does a naturalistic stimulus produce the same representations of the movie in each subject?
Representation of Objects from 3-D depth cues
Human can recognize object regardless of variant visual appearance (color, shape, texture, etc.) and regardless of retinal image variation (size, position, orientation, etc.). For example, if we see a mug far away or really close by, rotated or upside-down, made with porcelain or plastic, been white or green, we can still say it’s a mug, even though its image representation was completely different in our eyes. This property is called invariance of object recognition. Two previous papers in our lab addressing the behavior aspect of invariant face recognition demonstrate that face recognition is depth cue invariant (Akhavein & Farivar, 2017; Dehmoobadsharifabadi & Farivar, 2016).
One way to make inference about the brain is to discover differences in brain activity when subjects are watching different objects. The basic assumption is that if the activity in one brain area can tell which object the subject is watching invariantly, this brain area must bear information about the objects, and thus participate in the object recognition. The heated debate arises around what aspects of brain activity differs when people are watching different objects. Activity amplitude (univariate activity) is shown to related to object recognition. More recent ideas tend to attract attentions that the activity patterns (multivariate) of certain patch of the brain is more related to invariant object recognition. In our lab, we take the multivariate pattern approach and investigate depth cue invariance using MEG and fMRI (Akhavein & Farivar, 2018, in press). We extent the idea of current multivariate activity pattern to a new level by postulating that representations of invariant object recognition should not be confined by local activity pattern. The invariance of object recognition should include the property of brain networks. Working progress in the lab include investigating the object representations in the brain at network level.
Akhavein, H., & Farivar, R. (2017). Gaze behavior during 3-D face identification is depth cue invariant. Journal of Vision, 17(2):9, 1–12.
Dehmoobadsharifabadi, A., & Farivar, R. (2016). Are face representations depth cue invariant? Journal of Vision, 16(8):6. doi: 10.1167/16.8.6.
Early Visual Representations
The sclera, the cornea, the iris, and the pupil are most visible parts that are seen in external view of the eyes.
The retina absorbs light rays and change the electromagnetic information into information that can be transmitted by the neurons. The retina contains the fovea, where vision is sharpest, and the optic disc, where the absence of light receptors creates a blind spot. Cells in retina include the photoreceptors, bipolar cells, ganglion cells, horizontal cells, and amacrine cells.
Light energy breaks down the photopigments found in rods and cones. This process causes the photoreceptor to become hyperpolarized, which is the neural signal that initiates the sensory experience of vision.
Cones and Rods differ in their reaction to a change from brightly lit to dimly lit conditions(dark adaptation). Cones adapt rapidly but are relatively insensitive. Rods adapt more slowly but allow us to see in very dimly lit conditions. The eyes require about 30 minutes to completely adapt to the dark.
Cones and Rods differ in their reaction to a change from dimly lit to brightly lit conditions(light adaptation). Cones adapt rapidly, allowing us to see clearly. Rods rapidly become bleached and are not functional under brightly lit conditions.
Ganglion-cell electrical activity is studied by inserting a microelectrode near a ganglion cell, using single-cell recording techniques. The receptive fields of ganglion cells are either on-center, off-surround or off-center, on-surround.
Amacrine cells and horizontal cells provide lateral connections within the retina to influence the activity of the cells providing vertical connections (i.e., the photoreceptors, bipolar cells, and ganglion cells).
Pathway from Retina to the Visual Cortex:
The visual system has two kinds of crossovers: (a) visual material is reversed by the lens onto the retina and (b) at the optic chiasm, half of the fibres in each optic nerve cross over. As a result of these crossovers, everything from the left side of the visual filed ends up on the right hand side of the head, and everything from the right side of the visual filed ends up on the left hand side of the head.
The optic nerve is called the optic tract beyond the optic chiasm. The optic track travels to the superior colliculus, which is important in the detection of movement, and to the lateral geniculate nucleus, which is an important way station for processing visual input.
The lateral geniculate nucleus is organized into layers that keep separate the information from the two eyes; cells in the LGN function like ganglion cells.
The visual cortex, which is responsible for higher levels of visual processing, is divided into the primary visual cortex (also called Area 27, striate cortex, and V1) and the secondary visual cortex (also called the extrastriate cortex).
Neuronal messages from the LGN arrives in layer 4C of V1, which has a retinotopic arrangement.
Outside of the layer 4C, the primary visual cortex has two basic kinds of neurons: simple cortical cells(responding to lines and edges) and complex cortical cells (responding to movement). Some of both types of cells are end-stopped, which means that they respond most to lines that end with their receptive fields.
Neurons in V1 are arranged in columns. Neurons within each column respond best to a line of one particular orientation. Cells in an adjacent column have the highest respond rate to a line whose orientation has shifted by only about 10o.
Three visual pathways (P, M and K) originate in the retina with different ganglion cells. Each pathway has different characteristics and functions.
The secondary visual cortex receives information from the primary visual cortex. Within the secondary visual cortex are two important pathways: the Where pathway and the What pathway.
The Where pathway is important for spatial location; it runs from the primary visual cortex through the secondary visual cortex to the parietal lobe. The What pathway is important for object recognition; it turns from the primary visual cortex through the secondary visual cortex to the temporal lobe.
Disease and Recovery Models
We actively seek to contribute to the study of disease and recovery and to learn from these interactions to better understand fundamental aspects of cortical processing. We currently study three different diseases, and seek to understand recovery mechanisms in these as well.
Amblyopia is a neurodevelopmental disorder of the visual cortex. It results from a disruption of binocular vision in the early years of life. Amblyopic patients have low visual acuity, low sensitivity to contrast, as well as visual distortions. Different models were developed in the goal of understanding those visual deficits further, but none could come to a common agreement. We believed that amblyopia could be modeled based on higher levels of intrinsic blur in their visual system compared to normal observers. We used psychophysics to test image and edge blur discrimination in amblyopic subjects and found a level of intrinsic blur that was superior to normal observers. Another project that we are working on is developing a model to realize the magnitude of visual perceptual distortion in amblyopia.