Fundamental mechanisms of visual motion detection: models,cells and functions

Taking a comparative approach, data from a range of visual species are discussed in the context of ideas about mechanisms of motion detection. The cellular basis of motion detection in the vertebrate retina, sub-cortical structures and visual cortex is reviewed alongside that of the insect optic lobes. Special care is taken to relate concepts from theoretical models to the neural circuitry in biological systems. Motion detection involves spatiotemporal pre-filters, temporal delay filters and non-linear interactions. A number of different types of non-linear mechanism such as facilitation, inhibition and division have been proposed to underlie direction selectivity. The resulting direction-selective mechanisms can be combined to produce speed-tuned motion detectors. Motion detection is a dynamic process with adaptation as a fundamental property. The behavior of adaptive mechanisms in motion detection is discussed, focusing on the informational basis of motion adaptation, its phenomenology in human vision, and its cellular basis. The question of whether motion adaptation serves a function or is simply the result of neural fatigue is critically addressed.     

Primary sensory cortices, top-down projections and conscious experience

Beginning with a prominent article by Crick and Koch in 1995 (Nature 375, 121-123), cognitive neuroscience has witnessed an intensive debate about whether or not neural activity in primary visual cortex correlates with conscious visual experience. While some studies--especially those employing functional magnetic resonance imaging--imply that this is the case, others--particularly those recording from single neurons--suggest that it is not. In the light of this ongoing controversy, it is surprising that the analogous question in other sensory modalities has received far less attention. The first part of the present article reviews studies relevant to the role of primary auditory and primary somatosensory cortices in conscious auditory and tactile experience. As will become evident, the results of these studies, at least at first sight, appear no less contradictory than those obtained in the visual modality--in fact, they evidence discrepancies that resemble those found in the visual system to an impressive degree. The second part of the article attempts to reconcile the seemingly contradictory data by suggesting that only activity induced in the primary sensory cortices through cortico-cortical top-down signals can become consciously accessible, whereas activity induced by bottom-up signals from the thalamus cannot. This conclusion is in line with the earlier proposals of several prominent neuroscientists that portrayed conscious perception as the result of an active interpretative process by the brain, rather than a passive reflection of the environment.     

Neural correlates of spatial working memory in humans: a functional magnetic resonance imaging study comparing visual and tactile processes

Recent studies of neural correlates of working memory components have identified both low-level perceptual processes and higher-order supramodal mechanisms through which sensory information can be integrated and manipulated. In addition to the primary sensory cortices, working memory relies on a widely distributed neural system of higher-order association areas that includes posterior parietal and occipital areas, and on prefrontal cortex for maintaining and manipulating information. The present study was designed to determine brain patterns of neural response to the same spatial working memory task presented either visually or in a tactile format, and to evaluate the relationship between spatial processing in the visual and tactile sensory modalities. Brain activity during visual and tactile spatial working memory tasks was measured in six young right-handed healthy male volunteers by using functional magnetic resonance imaging. Results indicated that similar fronto-parietal networks were recruited during spatial information processing across the two sensory modalities-specifically the posterior parietal cortex, the dorsolateral prefrontal cortex and the anterior cingulate cortex. These findings provide a neurobiological support to behavioral observations by indicating that common cerebral regions subserve generation of higher order mental representations involved in working memory independently from a specific sensory modality.     

Transcranial magnetic stimulation-induced ‘visual echoes’ are generated in early visual cortex

Transcranial magnetic stimulation (TMS) of the early visual areas can trigger perception of a flash of light, a so-called phosphene. Here we show that a very brief presentation of a stimulus can modulate features of a subsequent TMS-induced phosphene, to a level that participants mistake phosphenes for real stimuli, inducing ‘visual echoes’ of a previously seen stimulus. These ‘echoes’ are modulated by visual context at the moment of magnetic stimulation, showing that they are generated in early visual areas, and that the brain processes these ‘echoes’ as if they are factually presented stimuli. This shows that TMS can re-activate weak visual representations in early visual areas. Based on the pattern of contextual modulation of visual echoes, we theorize that perception of these echoes is not a passive reactivation of residual activity in early visual cortex, but an active interpretation of the combined activity of TMS-induced neural noise and cortical state.     

Mind does really matter: evidence from neuroimaging studies of emotional self-regulation, psychotherapy, and placebo effect

This article reviews neuroimaging studies of conscious and voluntary regulation of various emotional states (sexual arousal, sadness, negative emotion). The results of these studies show that metacognition and cognitive recontextualization selectively alters the way the brain processes and reacts to emotional stimuli. Neuroimaging studies of the effect of psychotherapy in patients suffering from diverse forms of psychopathology (obsessive-compulsive disorder, panic disorder, unipolar major depressive disorder, social phobia, spider phobia, borderline personality) are also examined. The results of these studies indicate that the mental functions and processes involved in diverse forms of psychotherapy exert a significant influence on brain activity. Neuroimaging investigations of the placebo effect in healthy individuals (placebo analgesia, psychostimulant expectation) and patients with Parkinson's disease or unipolar major depressive disorder are also reviewed. The results of these investigations demonstrate that beliefs and expectations can markedly modulate neurophysiological and neurochemical activity in brain regions involved in perception, movement, pain, and various aspects of emotion processing. Collectively, the findings of the neuroimaging studies reviewed here strongly support the view that the subjective nature and the intentional content (what they are "about" from a first-person perspective) of mental processes (e.g., thoughts, feelings, beliefs, volition) significantly influence the various levels of brain functioning (e.g., molecular, cellular, neural circuit) and brain plasticity. Furthermore, these findings indicate that mentalistic variables have to be seriously taken into account to reach a correct understanding of the neural bases of behavior in humans. An attempt is made to interpret the results of these neuroimaging studies with a new theoretical framework called the Psychoneural Translation Hypothesis.     

TMS disruption of V5/MT+ indicates a role for the dorsal stream in word recognition

Although word recognition is a skill commonly expected to rely more on ventral rather than dorsal stream processing, there is some evidence for a magnocellular/dorsal impairment in dyslexia. The early rapid feedforward/feedback loop through the dorsal stream seen in primate has been suggested to allow an initial global analysis, and in human early activation of parietal attention mechanisms for detecting salient stimuli, facilitating more local level detailed ventral stream processing. To test this model in humans, transcranial magnetic stimulation (TMS) was used to probe the role of early visual cortex (V1/V2) and V5/MT+ in single word identification. TMS over V1/V2 between word onset and 36 ms post word onset disrupted accurate word discrimination, with disruption also evident at approximately 99 ms. TMS over V5/MT+ also disrupted accuracy following stimulation at approximately the same time as word onset and again at 130 ms post word onset. Thus, a role for V5/MT+ in accurate single word identification is apparent suggesting rapid triggering of attention to salient exogenous stimuli may be required prior to processing in primary and temporal cortical regions.     

Perception of illusory contours forms intermodulation responses of steady state visual evoked potentials as a neural signature of spatial integration

Perception of illusory contours was shown to be a consequence of neural activity related to spatial integration in early visual areas. Candidates for such filling-in phenomena are long-range horizontal connections of neurons in V1/V2, and feedback from higher order visual areas. To get a direct measure of spatial integration in early visual cortex, we presented two differently flickering inducers, which evoked steady-state visual evoked potentials (SSVEPs) while manipulating the formation of an illusory rectangle. As a neural marker of integration we tested differences in amplitudes of intermodulation frequencies i.e. linear combinations of the driving frequencies. These were significantly increased when an illusory rectangle was perceived. Increases were neither due to changes of any of the two driving frequencies nor in the frequency that tagged the processing of the compound object, indicating that results are not a consequence of paying more attention to inducers when the illusory rectangle was visible.     

基于深度残差网络的皮肤镜图像黑色素瘤的识别

恶性黑色素瘤是最常见和最致命的皮肤癌之一。临床上,皮肤镜检查是恶性黑色素瘤早期诊断的常规手段。但是人工检查费力、费时,并且高度依赖于皮肤科医生的临床经验。因此,研究出自动识别皮肤镜图像中的黑色素瘤算法显得尤为重要。提出一种皮肤镜图像自动评估的新框架,利用深度学习方法,使其在有限的训练数据下产生更具区分性的特征。具体来说,首先在大规模自然图像数据集上预训练一个深度为152层的残差神经网络(Res-152),用来提取皮肤病变图像的深度卷积层特征,并对其使用均值池化得到特征向量,然后利用支持向量机(SVM)对提取的黑色素瘤特征进行分类。在公开的皮肤病变图像ISBI 2016挑战数据集中,用所提出的方法对248幅黑色素瘤图像和1 031幅非黑色素瘤图像进行评估,达到86.28%的准确率及84.18%的AUC值。同时,为论证神经网络深度对分类结果的影响,比较不同深度的模型框架。与现有使用传统手工特征的研究(如基于密集采样SIFT描述符的词袋模型)相比,或仅从深层神经网络的全连接层提取特征进行分类的方法相比,新方法能够产生区分性能更强的特征表达,可以在有限的训练数据下解决黑色素瘤的类内差异大、黑色素瘤与非黑素瘤之间的类间差异小的问题。     

Objects and their icons in the brain: the neural correlates of visual concept formation

We are constantly exposed to symbols such as traffic signs, emoticons in internet communication, or other abstract representations of objects as well as, of course, the written words. However, aside from the word reading, little is known about the way our brain responds when we read non-lexical iconic symbols. By using functional MRI, we found that the watching of icons recruited manifold brain areas including frontal and parietal cortices in addition to the temporo-occipital junction in the ventral pathway. Remarkably, the brain response for icons was contrasted with the response for corresponding concrete objects with the pattern of 'hyper-cortical and hypo-subcortical' brain activation. This neural underpinning might be called the neural correlates for visual concept formation.     

Responses of primate area MT during the execution of optokinetic nystagmus and afternystagmus

The directional selectivity of the visual response properties was determined in 148 neurons, all located in area MT of three hemispheres of two macaque monkeys. The perferred direction of every neuron was obtained by analyzing the response obtained by a circular movement of the background while the monkeys fixated a stationary target. The distribution of the preferred directions was isotropic and showed no ipsiversive bias. MT neurons were excited in a directionally selective manner during the execution of optokinetic nystagmus, in a similar way to that produced by visual stimulation during fixation. The majority of neurons showed a sensitivity to the velocity of retinal image slip. Activity during the execution of optokinetic nystagmus could be traced back to residual retinal image slip in the direction of optokinetic stimulation. No dynamic effects of the neuronal activity during the build-up of eye velocity in early optokinetic nystagmus were observed. Obviously, the activity in area MT did not reflect the charging of the velocity storage mechanism. Accordingly, following the cessation of stimulation, the activity dropped to the level of spontaneous activity and did not parallel the execution of optokinetic afternystagmus. These results suggest that area MT is not part of the velocity storage mechanism and, furthermore, that the storage mechanism must be downstream of area MT in the processing of visual motion for the generation of the optokinetic nystagmus and afternystagmus.     

Auditory processing that leads to conscious perception: a unique window to central auditory processing opened by the mismatch negativity and related responses

In this review, we will present a model of brain events leading to conscious perception in audition. This represents an updated version of Näätänen's previous model of automatic and attentive central auditory processing. This revised model is mainly based on the mismatch negativity (MMN) and N1 indices of automatic processing, the processing negativity (PN) index of selective attention, and their magnetoencephalographic (MEG) and functional magnetic resonance imaging (fMRI) equivalents. Special attention is paid to determining the neural processes that might underlie conscious perception and the borderline between automatic and attention‐dependent processes in audition.     

C-Deoxyglucose mapping of the monkey brain during reaching to visual targets

The strategies used by the macaca monkey brain in controlling the performance of a reaching movement to a visual target have been studied by the quantitative autoradiographic ¹⁴C-DG method.Experiments on visually intact monkeys reaching to a visual target indicate that V1 and V2 convey visuomotor information to the cortex of the superior temporal and parietoccipital sulci which may encode the position of the moving forelimb, and to the cortex in the ventral part and lateral bank of the intraparietal sulcus which may encode the location of the visual target. The involvement of the medial bank of the intraparietal sulcus in proprioceptive guidance of movement is also suggested on the basis of the parallel metabolic effects estimated in this region and in the forelimb representations of the primary somatosensory and motor cortices. The network including the inferior postarcuate skeletomotor and prearcuate oculomotor cortical fields and the caudal periprincipal area 46 may participate in sensory-to-motor and oculomotor-to-skeletomotor transformations, in parallel with the medial and lateral intraparietal cortices.Experiments on split brain monkeys reaching to visual targets revealed that reaching is always controlled by the hemisphere contralateral to the moving forelimb whether it is visually intact or ‘blind'. Two supplementary mechanisms compensate for the ‘blindness' of the hemisphere controlling the moving forelimb. First, the information about the location of the target is derived from head and eye movements and is sent to the ‘blind' hemisphere via inferior parietal cortical areas, while the information about the forelimb position is derived from proprioceptive mechanisms and is sent via the somatosensory and superior parietal cortices. Second, the cerebellar hemispheric extensions of vermian lobules V, VI and VIII, ipsilateral to the moving forelimb, combine visual and oculomotor information about the target position, relayed by the ‘seeing' cerebral hemisphere, with sensorimotor information concerning cortical intended and peripheral actual movements of the forelimb, and then send this integrated information back to the motor cortex of the ‘blind' hemisphere, thus enabling it to guide the contralateral forelimb to the target.     

Smooth-pursuit eye movements elicited by first-order and second-order motion

 The perception of the displacement of luminance-defined contours (i.e., first-order motion) is an important and well-examined function of the visual system. It can be explained, for example, by the operation of elementary motion detectors (EMDs), which cross-correlate the spatiotemporal luminance distribution. More recent studies using second-order motion stimuli, i.e., shifts of the distribution of features such as contrast, texture, flicker, or motion, extended classic concepts of motion perception by including nonlinear or hierarchical processing in the EMD. Smooth-pursuit eye movements can be used as a direct behavioral probe for motion processing. The ability of the visual system to extract motion signals from the spatiotemporal changes of the retinal image can be addressed by analyzing the elicited eye movements. We measured the eye movement response to moving objects defined by two different types of first-order motion and two different types of second-order motion. Our results clearly showed that the direction of smooth-pursuit eye movements was always determined by the direction of object motion. In particular, in the case of second-order motion stimuli, smooth-pursuit did not follow the retinal image motion. The latency of the initial saccades during pursuit of second-order stimuli was slightly but significantly increased, compared with the latency of saccades elicited by first-order motion. The processing of second-order motion in the peripheral visual field was less exact than the processing of first-order motion in the peripheral field. Steady state smooth-pursuit eye speed did not reflect the velocity of second-order motion as precisely as that of first-order motion, and the resulting retinal error was compensated by saccades. Interestingly, for slow second-order stimuli we observed that the eye could move faster than the target, leading to small, corrective saccades in the opposite direction to the ongoing smooth-pursuit eye movement. We conclude from our results that both visual perception and the control of smooth-pursuit eye movements have access to processing mechanisms extracting first- and second-order motion. Received: 26 August 1996 / Accepted: 8 November 1996     

Primary visual cortex neurons that contribute to resolve the aperture problem

It is traditional to believe that neurons in primary visual cortex are sensitive only or principally to stimulation within a spatially restricted receptive field (classical receptive field). It follows from this that they should only be capable of encoding the direction of stimulus movement orthogonal to the local contour, since this is the only information available in their classical receptive field "aperture." This direction is not necessarily the same as the motion of the entire object, as the direction cue within an aperture is ambiguous to the global direction of motion, which can only be derived by integrating with unambiguous components of the object. Recent results, however, show that primary visual cortex neurons can integrate spatially and temporally distributed cues outside the classical receptive field, and so we reexamined whether primary visual cortex neurons suffer the "aperture problem." With the stimulation of an optimally oriented bar drifting across the classical receptive field in different global directions, here we show that a subpopulation of primary visual cortex neurons (25/81) recorded from anesthetized and paralyzed marmosets is capable of integrating informative unambiguous direction cues presented by the bar ends, well outside their classical receptive fields, to encode global motion direction. Although the stimuli within the classical receptive field were identical, their directional responses were significantly modulated according to the global direction of stimulus movement. Hence, some primary visual cortex neurons are not local motion energy filters, but may encode signals that contribute directly to global motion processing.     

Olfaction and its dynamic influence on word and face processing: cross-modal integration

The article specifies several important aspects related to the sense of smell in vertebrates. The idea that odors exert effects in the human brain though being not consciously perceived is introduced. Functional aspects related to cross-modal sensory interaction between olfaction and vision are highlighted. In particular, studies making use of electrophysiological methods providing high temporal resolution reveal an early processing stage around 300 ms and a later stage around 700 ms after stimulus onset. The early stage has been associated with subconscious olfactory information processing, whereas the later stage most likely reflects conscious odor perception. Specific interactions are described between olfaction and language and between olfaction and face processing in correlation with both stages of olfactory information processing. A consciously perceived odor can negatively affect language and face processing if these stimuli are presented and associated simultaneously, whereas simultaneous subconscious odor processing has the potential to improve memory formation in other stimulus modalities. Strikingly, the subconscious effect seems not to depend on odor valence. Besides a better understanding of the sense of olfaction itself, these findings on cross-modal integration support the idea that neural representations exist for semantic contents (object meaning) independent from particular sensory modalities. These representations can be referred to as meta representations because the information they contain is derived from a great variety of sensory information integrated into a semantic representation of an object. It is suggested that such meta representations represent the basic units for cognition and that they provide inputs during dreaming.     

Functional impact of interneuronal inhibition in the cerebral cortex of behaving animals

This paper reviews recent progress in understanding the functional roles of inhibitory interneurons in behaving animals and how they affect information processing in cortical microcircuits. Multiple studies have shown that the morphological subtypes of inhibitory cells show distinct electrophysiological properties, as well as different molecular and neurochemical identities, providing a large mosaic of inhibitory mechanisms for the dynamic processing of information in the cortex. However, it is only recently that some specific functions of different interneuronal subtypes have been described in behaving animals. In this regard, influential results have been obtained using the known differences of interneurons and pyramidal cells recorded extracellularly to dissociate the functional roles that these two classes of neurons may play in the cortical microcircuits during various behaviors. Neurons can be segregated into fast-spiking (FS) cells that show short action potentials, high discharge rates, and correspond to putative interneurons; and regular-spiking (RS) cells that show larger action potentials and correspond to pyramidal neurons. Using this classification strategy, it has been found that cortical inhibition is involved in sculpting the tuning to different stimulus or behavioral features across a wide variety of sensory, association, and motor areas. Recent studies have suggested that the increase in high-frequency synchronization during information processing and spatial attention may be mediated by FS activation. Finally, FS are active during motor planning and movement execution in different motor areas, supporting the notion that inhibitory interneurons are involved in shaping the motor command but not in gating the cortical output.     

Corticocortical connections to the motor cortex from the posterior parietal lobe (areas 5a, 5b, 7) in the cat demonstrated by the retrograde axonal transport of horseradish peroxidase

Summary Neurons in the parietal region of the cerebral cortex, projecting to the ipsilateral distal forelimb area of the motor cortex (area 4) were identified in the cat brain using the horseradish peroxidase (HRP) retrograde tracing method. After making microinjections of HRP into the distal forelimb area of the motor cortex, clusters of HRP-labeled cell bodies were observed in different regions of the ipsilateral parietal cortex. In particular these clusters of labeled cells were found in areas 5a, 5b and 7. The area 5a cluster is formed from closely packed irregularly-shaped cells, the area 5b cluster is made up of dispersed medium-sized pyramidal cells, while area 7 contains a cluster of widely dispersed small pyramidal cells. Typically, labeled cell bodies were found in lamina III of cortex. Labeled cell bodies were neither observed in the contralateral cortex nor in the visual cortex (areas 17, 18 and 19). Since parietal cortex receives projections from primary somatosensory and visual cortex, the projections from parietal to motor cortex may well form the neural substrate for the processing of convergent sensory information used in voluntary movements.     

卷积神经网络在肝癌病理切片图像分类中的应用

目的探讨基于卷积神经网络的肝脏组织切片图像正常和病变性分类方法的可行性及应用价值。方法使用一种能够自动学习图像特征并分类的方法,先利用原始的Inception V3模型在肝脏组织切片数据集上进行训练,然后在原始模型的基础上通过微调得到改进的Inception V3模型,最后用改进的模型来实现肝脏组织切片图像正常和病变性两种类型的分类。结果改进后的Inception V3模型对肝脏切片图像的分类结果较佳,平均分类准确率达到99.2%。结论卷积神经网络的肝脏组织切片图像正常和病变性分类方法可行、合理,改进的Inception V3模型的分类效果较好。     

Retrograde transneuronal transport of wheat-germ agglutinin to the retina from visual cortex in the cat

Summary The stability of visual perception despite eye movements suggests the existence, in the visual system, of neural elements able to recognize whether a movement of an image occurring in a particular part of the retina is the consequence of an actual movement that occurred in the visual field, or self-induced by an ocular movement while the object was still in the field of view. Recordings from single neurons in area V3A of awake macaque monkeys were made to check the existence of such a type of neurons (called real-motion cells; see Galletti et al. 1984, 1988) in this prestriate area of the visual cortex. A total of 119 neurons were recorded from area V3A. They were highly sensitive to the orientation of the visual stimuli, being on average more sensitive than V1 and V2 neurons. Almost all of them were sensitive to a large range of velocities of stimulus movement and about one half to the direction of it. In order to assess whether they gave different responses to the movement of a stimulus and to that of its retinal image alone (self-induced by an eye movement while the stimulus was still), a comparison was made between neuronal responses obtained when a moving stimulus swept a stationary receptive field (during steady fixation) and when a moving receptive field swept a stationary stimulus (during tracking eye movement). The receptive field stimulation at retinal level was physically the same in both cases, but only in the first was there actual movement of the visual stimulus. Control trials, where the monkeys performed tracking eye movements without any intentional receptive field stimulation, were also carried out. For a number of neurons, the test was repeated in darkness and against a textured visual background. Eighty-seven neurons were fully studied to assess whether they were real-motion cells. About 48% of them (42/87) showed significant differences between responses to stimulus versus eye movement. The great majority of these cells (36/42) were real-motion cells, in that they showed a weaker response to visual stimulation during tracking than to the actual stimulus movement during steady fixation. On average, the reduction in visual response during eye movement was 64.0 ± 15.7% (SD). Data obtained with a uniform visual background, together with those obtained in darkness and with textured background, indicate that real-motion cells receive an eye-motion input, either retinal or extraretinal in nature, probably acting presynaptically on the cell's visual input. In some cases, both retinal and extraretinal eye-motion inputs converge on the same real-motion cell. No correlation was observed between the real-motion behaviour and the sensitivity to either orientation or direction of movement of the visual stimulus used to activate the receptive field, nor with the retinotopic location of the receptive field. We suggest that the visual system uses real-motion cells in order to distinguish real from self-induced movements of retinal images, hence to recognize the actual movement in the visual field. Based on psychophysical data, the hypothesis has been advanced of an internal representation of the field of view, stable despite eye movement (cf. MacKay 1973). The real-motion cells may be neural elements of this network and we suggest that the visual system uses the output of this network to properly interpret the large number of sensory changes resulting from exploratory eye movements in a stable visual world.     

Projections from the medial nucleus of the inferior pulvinar complex to the middle temporal area of the visual cortex

Medial, central and posterior nuclei have been previously identified within the inferior pulvinar complex of owl monkeys (Lin & Kaas, 1979). In the present experiments injections of [³H]-proline into the posterior thalamus, including the medial nucleus, produced densely labeled terminals in the middle temporal area of the visual cortex (Allman & Kaas, 1971). Injections of the retrograde tracer, horseradish peroxidase, into the middle temporal area labeled most of the neurons in the medial nucleus and only occasional neurons in the central and posterior nuclei. The posterior nucleus appeared to project to the cortex rostral to the middle temporal area in the temporal lobe, and the central nucleus projects to the visual cortex caudal to the middle temporal area. We conclude that the middle temporal area is the major cortical target of the medial nucleus in the inferior pulvinar complex and that the central and posterior nuclei have other cortical targets.Thus, these findings support the view that the inferior pulvinar complex consists of three distinct nuclei, which should lead to further progress in understanding its connections and functions.