Search

This paper aims to apply deep learning to identify autism spectrum disorder (ASD) patients from a large brain imaging dataset based on the patients’ brain activation patterns. The brain images are collected from the ABIDE (Autism Brain Imaging Data Exchange) database. The proposed convolutional neural network (CNN) architecture investigates functional connectivity patterns between different brain areas to identify specifics patterns to diagnose ASD. The enhanced CNN uses blocks of temporal convolutional layers that employ casual convolutions and dilations; hence, it is suitable for sequential data with temporality large receptive fields. Experimental results show that the proposed ECNN achieves an accuracy of up to 80% accuracy. These patterns show an anticorrelation of brain function between anterior and posterior areas of the brain; that is, the disruption in brain connectivity is one primary evidence of ASD.

Visual explanations aim at providing an understanding of the inner behavior of convolutional neural networks. Naturally, it is necessary to explore whether these methods themselves are reasonable and reliable. In this paper, we focus on Class Activation Mapping (CAM), a type of attractive explanations that has been widely applied to model diagnosis and weakly supervised tasks. Our contribution is two-fold. First, we identify an important but neglected issue that affects the reliability of CAM results: there is a misalignment between the effective receptive field and the implicit receptive field, where the former is determined by the model and the input, and the latter is determined by the upsampling function in CAM. Occlusion experiments are designed to empirically testify to its existence. Second, based on this finding, an adversarial marginal attack is proposed to fool the CAM-based method and the CNN model simultaneously. Experimental results demonstrate that the provided saliency map can be completely changed to another shape by only perturbing the area with 1-pixel width. The prototype code of the method is now available at https://github.com/xpf/CAM-Adversarial-Marginal-Attack .

We conduct a comparative evaluation of the visual systems from the retina to the muscles of the mouse and the macaque monkey noting the differences and similarities between these two species. The topics covered include (1) visual-field overlap, (2) visual spatial resolution, (3) V1 cortical point-image [i.e., V1 tissue dedicated to analyzing a unit receptive field], (4) object versus motion encoding, (5) oculomotor range, (6) eye, head, and body movement coordination, and (7) neocortical and cerebellar function. We also discuss blindsight in rodents and primates which provides insights on how the neocortex mediates conscious vision in these species. This review is timely because the field of visuomotor neurophysiology is expanding beyond the macaque monkey to include the mouse; there is therefore a need for a comparative analysis between these two species on how the brain generates visuomotor responses.

Image super-resolution from low-resolution fundus image has valuable applications in clinical practices. The popular methods yield unsatisfactory results when the fundus images are contaminated due to the bleeding or plaques caused by eye diseases. To this end, we propose a style-guided UNet (SGUNet) which incorporates a series of style-guided U-shape block (SUB) for fundus image super-resolution. Each SUB consists of trunk and mask branches. The proposed trunk branch is a U-shape structure that intends to enlarge the receptive field by down-sampling via large-stride convolution, and fuses the complementary information under the different receptive fields. The mask branch then dynamically estimates the relative importance of individual potential styles to reweigh the feature maps according to the significance of the potential styles. To fully leverage the hierarchical features, a dense feature fusion scheme is introduced by concatenating the output of preceding SUBs. We extensively validate the proposed network on low-resolution retina dataset with adversely affected by diseases. The experimental results demonstrate that our SGUNet achieves superior performance with excellent robustness and high accuracy by comparing with six popular methods.

Neurons in the primary visual cortex (V1) are often classified as simple or complex cells, but it is debated whether they are discrete hierarchical classes of neurons or if they represent a continuum of variation within a single class of cells. Herein, we show that simple and complex cells may arise commonly from the feedforward projections from the retina. From analysis of the cortical receptive fields in cats, we show evidence that simple and complex cells originate from the periodic variation of ON–OFF segregation in the feedforward projection of retinal mosaics, by which they organize into periodic clusters in V1. From data in cats, we observed that clusters of simple and complex receptive fields correlate topographically with orientation maps, which supports our model prediction. Our results suggest that simple and complex cells are not two distinct neural populations but arise from common retinal afferents, simultaneous with orientation tuning.

The segmentation of liver and liver tumor is an essential step for computer-aided liver disease diagnosis, treatment and prognosis. Although deep convolutional neural networks have contributed to liver and tumor segmentation, their architectures can not maintain spatial details and long-range context information. Besides, the fixed receptive fields of these networks limit the segmentation performance of livers and tumors with variant sizes and shapes. To address above problems, we propose a deep attention neural network which contains high-resolution branch and multi-scale features aggregation for cascaded liver and tumor segmentation from CT images. To be specific, the high-resolution branch can maintain the resolution of the input image and thus preserves the spatial details. The multi-scale features exchange and fusion enable the receptive fields of the network to adapt to liver and tumor with variant shapes and sizes. The appended attention module evaluates the similarities between every two pixels to model the long-range dependence and context information so that the network can segment liver and tumor areas located in distant regions. Experimental results on the LiTS and the 3D-IRCADb datasets demonstrate that our method can generate satisfying performance.

Retinal processing of a visual scene is an essential step of human visual perception. Although foveal vision is linked to the visual attention, perception is by not means limited to this region. Rather, the retinal field of view ranges from 60 ° nasal to 107 ° temporal, and from 70 ° superior to 80 ° inferior. Whether a scene object is visually perceived depends on both its visual appearance as well as its retinal location. We present a framework to evaluate the visual stimulus of a scene object with regard to different types of retinal receptive fields. Driven by gaze location provided by an eye tracker, the estimated retinal response considers the visual appearance of the object, its eccentricity in the users field of view, and the capabilities at the local retinal region. A desktop experiment shows that, in additional to foveal processing, the estimated retinal response leads to a significant increase in classification accuracy in terms of whether an object is reported as perceived by the user.

When presented with locally paired dots moving in opposite directions, motion selective neurons in the middle temporal cortex (MT) reduce firing while neurons in V1 are unaffected. This physiological effect is known as motion opponency. The current study used psychophysics to investigate the neural circuit underlying motion opponency. We asked whether opposing motion signals could arrive from different eyes into the receptive field of a binocular neuron while still maintaining motion opponency. We took advantage of prior findings that orientation discrimination of the motion axis (along which paired dots oscillate) is harder when dots move counter-phase than in-phase, an effect associated with motion opponency. We found that such an effect disappeared when paired dots originated from different eyes. This suggests that motion opponency, at some point, involves strictly monocular processing. This does not mean that motion opponency is entirely monocular. Further, we found that the effect of a Glass pattern disappeared under similar viewing conditions, suggesting that Glass pattern perception also involves some strictly monocular processing.

Despite the wide success in many vision tasks, it is still challenging for Convolutional Neural Networks (CNNs) to perform saliency detection due to their limited receptive fields and lack...

Cone photoreceptors are the first neurons along the visual pathway that exhibit center-surround antagonistic receptive fields, the basic building blocks for spatial information processing in the visual system. The surround responses in cones are mediated by the horizontal cells (HCs) via multiple feedback synaptic mechanisms. It has been controversial on which mechanisms are responsible for the surround-elicited depolarizing responses in cones (ΔVCone(s)), and whether the surround responses of various types of cones are mediated by the same HC feedback mechanisms. In this report, we studied ΔVCone(s)) of four types of cones in the salamander retina, and found that they are mediated by feedback synapses from A-type, B-type or A- and B-type HCs. ΔVCone(s) are observable in the presence of concomitant center light spots, and surround + center light stimuli of various intensity, size and wavelength differentially activate the feedback synapses from A- and B-type HCs to cones. We found that ΔVCone(s) of the L-cones are mediated by both A- and B-type HCs, those of the P- and S-cones by B-type HCs, and those of the A-cones by the A-type HCs. Moreover, our results suggest that B-type HCs mediate ΔVCone(s) through both GABAergic and GluT-ClC feedback synaptic mechanisms, and A-type HCs mediate ΔVCone(s) via the GluT-ClC feedback mechanism. Feedback synaptic mechanisms that increase calcium influx in cone synaptic terminals play important roles in mediating the antagonistic surround responses in the postsynaptic bipolar cells, but they may not generate enough current to depolarize the cones and significantly contribute to ΔVCone(s).

A modulatory role has been reported for the isoflavone, genistein, on voltage-gated Na+ channels in the trigeminal ganglion in vitro. However, the acute effects of genistein in vivo, particularly on nociceptive transmission in the trigeminal system, remain to be determined. The aim of the present study was to examine whether acute local genistein administration to rats attenuates the excitability of wide-dynamic range (WDR) spinal trigeminal nucleus caudalis (SpVc) neurons in response to nociceptive and non-nociceptive mechanical stimulation in vivo. Extracellular single unit recordings were made from SpVc WDR neurons in response to orofacial non-noxious and noxious mechanical stimulation of pentobarbital-anesthetized rats. The effects of local administration of genistein, lidocaine, and lidocaine with genistein to the receptive field on the discharge frequency of SpVc neurons were evaluated. The mean firing frequency of SpVc WDR neurons in response to both non-noxious and noxious mechanical stimuli was significantly and dose-dependently (0.1−10 mM) inhibited by genistein, and maximum inhibition of the discharge frequency of both non-noxious and noxious mechanical stimuli was seen within 10 min. The inhibitory effect of genistein lasted for 20 min and was reversible. No significant difference was seen between the relative magnitude of inhibition by genistein on the SpVc WDR neuronal discharge frequency for noxious and non-noxious stimulation. The mean magnitude of inhibition by genistein (10 mM) on SpVc neuronal discharge frequency was almost equal to that of the local anesthetic, 1 % lidocaine (37 mM). Local injection of half-dose of lidocaine replaced the half-dose of genistein. These results suggest that local injection of genistein into the peripheral receptive field suppresses the excitability of SpVc neurons, possibly via inhibition of voltage-gated Na+ channels in the nociceptive nerve terminals of trigeminal ganglion. Therefore, administration of genistein as a local anesthetic may provide relief from trigeminal nociceptive pain without side effects, thus contributing to the area of complementary and alternative medicines.

Convolutional neural network (CNN) has achieved promising results in image segmentation recently. However, for the segmentation of synthetic aperture radar (SAR) images with complicated scene, the single receptive field in CNN has a limited ability to effectively capture structural and regional information at the same time. In this paper, we propose a hierarchical fusion CNN (HIFCNN) model for SAR image segmentation. At each convolutional layer, HIFCNN sets several different-sized receptive fields, and thus extracts hierarchical features. Concretely, the larger-sized receptive field captures regional information and is robust against speckle, while the smaller one preserves the structural information well. Then, based on the Dempster-Shafer evidential theory, the proposed hierarchical network, HIFCNN, implements a decision-level fusion to integrate these hierarchical features. In this way, the structural and regional information can be accurately captured by different receptive fields, which is beneficial for edge location, structure preservation and region homogeneity in SAR image segmentation. The effectiveness of HIFCNN model is demonstrated by the application to the segmentation of the simulated images and real SAR images.

Based on visual cortical theory of Rybak, a new heterogeneous Rybak neural network (HRYNN) model is proposed for image enhancement. HRYNN is constructed with several Rybak neural network (RYNN) models proposed, which have different parameters corresponding to different neurons. We show that HRYNN can better represent prior information for edge detail enhancement than the logarithmic domain. To capture different resolution texture features of image, a novel receptive field model is proposed to solve the problem of detail enhancement. HRYNN model has excellent enhancement effect on the edge details of image based on the receptive field’s lateral inhibitory characteristics. Moreover, the experimental enhancement results of the colour images from Berkeley image Dataset show the validity and efficiency of the proposed enhancement method. Finally, three evaluation indicators are employed to measure the enhancement result.

Experience-dependent plasticity within visual cortex is controlled by postnatal maturation of inhibitory circuits, which are both morphologically diverse and precisely connected. Gene-targeted disruption of the voltage-dependent potassium channel Kv3.1 broadens action potentials and reduces net inhibitory function of parvalbumin (PV)-positive GABA subtypes within the neocortex. In mice lacking Kv3.1, the rate of input loss from an eye deprived of vision was slowed two-fold, despite otherwise normal critical period timecourse and receptive field properties. Rapid ocular dominance plasticity was restored by local or systemic enhancement of GABAergic transmission with acute benzodiazepine infusion. Diazepam instead exacerbated a global suppression of slow-wave oscillations during sleep described previously in these mutant mice, which therefore did not account for the rescued plasticity. Rapid ocular dominance shifts closely reflected Kv3.1 gene dosage that prevented prolonged spike discharge of their target pyramidal cells in vivo or the spike amplitude decrement of fast-spiking cells during bouts of high-frequency firing in vitro. Late postnatal expression of this unique channel in fast-spiking interneurons thus subtly regulates the speed of critical period plasticity with implications for mental illnesses.

Haze removal is an important pre-processing step in many computer vision tasks. Convolutional neural networks, especially the U-shaped networks, have shown to be effective in image dehazing. Nevertheless, these networks have three main limitations. First, the relevant haze information, e.g. concentration of haze, is totally ignored. Second, spatial inconsistency and information dilution usually occur when the networks refine the dehazed results with a coarse-to-fine strategy. Third, the receptive field of the network is not large enough to capture structural information. Motivated by these problems, a new attentive U-recurrent encoder-decoder dehazing network is presented, which consists of an attentive recurrent network and a U-recurrent encoder-decoder network. By assuming that haze layers with different depths can be detected by multiple stages, we use an attentive recurrent network to generate the haze attention map for guiding the U-recurrent encoder-decoder network with the concentration of haze to better estimate the clear image. Meanwhile, the features for dehazing are further enhanced and the dehazing results are refined in the U-recurrent encoder-decoder network. This design not only enables spatial consistency but also reduces information dilution with short recurrent pathways. Furthermore, a novel residual pyramid pooling module is also proposed and used in the U-recurrent encoder-decoder network, which provides the network with structural information and with an enlarged receptive field. The experimental results demonstrate that our method outperforms state-of-the-art dehazing algorithms on both synthetic and real hazy images.

Objective: The objective is to develop a new deep learning method for the estimation of respiratory effort from a chest-worn accelerometer during sleep. We evaluate performance, compare it against a state-of-the art method, and assess whether it can differentiate between sleep stages. Methods: In 146 participants undergoing overnight polysomnography data were collected from an accelerometer worn on the chest. The study data were partitioned into train, validation, and holdout (test) sets. We used the train and validation sets to generate and train a convolutional neural network and performed model selection respectively, while we used the holdout set (72 participants) to evaluate performance. Results: A convolutional neural network with 9 layers and 207,855 parameters was automatically generated and trained. The neural network significantly outperformed the best performing conventional method, based on Principal Component Analysis; it reduced the Mean Squared Error from 0.26 to 0.11 and it also performed better in the detection of breaths (Sensitivity 98.4 %, PPV 98.2 %). In addition, the neural network exposed significant differences in characteristics of respiratory effort between sleep stages (p < 0.001). Conclusion: The deep learning method predicts respiratory effort with low error and is sensitive and precise in the detection of breaths. In addition, it reproduces differences between sleep stages, which may enable automatic sleep staging, using just a chest-worn accelerometer.

Recent advances in salient detection have exploited the foreground or background information to assist other saliency cues such as contrast to achieve state-of-the-art results. However the problem remains challenging. For example, human skin color is easily overlooked during saliency detection. Simulating the human visual mechanism to improve the current algorithm, we propose a saliency model based on the color-opponent mechanisms of a certain type of color-sensitive double-opponent (DO) cells in the primary visual cortex (V1) of human visual system. Firstly, DO cells with concentric receptive fields (RFs) can detect region contrast for yielding foreground saliency map. Our approach is intuitive and easy to interpret, and it allows fast implementation. Then, skin saliency map is built in a way that is different from high-level factors such as face detection, combining skin region and spatial Euclidean distance weight in the RGB space, and the significant skin features can be obtained effectively. Finally, a linear fusion strategy is proposed to integrate different saliency maps. Experimental results with three well-known benchmark databases demonstrate that the proposed method can achieve competitive performance when compared to state-of-the-art methods. Saliency regions contain important skin features compared to other traditional methods.

This paper presents a new artificial neuron model capable of learning its receptive field in the topological domain of inputs. The experiments include tests of focusing neuron networks of one or two hidden layers on synthetic and well-known image recognition data sets. The results demonstrated that the focusing neurons can move their receptive fields towards more informative inputs. In the simple two-hidden layer networks, the focusing layers outperformed the dense layers in the classification of the 2D spatial data sets. Moreover, the focusing networks performed better than the dense networks even when 70% of the weights were pruned. The tests on convolutional networks revealed that using focusing layers instead of dense layers for the classification of convolutional features may work better in some data sets. This work was supported by TUBITAK 1001 program (no:118E722), Isik University BAP program (No: 16A202), and NVIDIA hardware donation of a Tesla K40 GPU unit. Publisher's Version

International audience; A fundamental task for the auditory system is to process communication sounds according to their behavioral significance. In many mammalian species, pup calls became more significant for mothers than other conspecific and heterospecific communication sounds. To study the cortical consequences of motherhood on the processing of communication sounds, we recorded neuronal responses in the primary auditory cortex of virgin and mother C57BL/6 mice which had similar ABR thresholds. In mothers, the evoked firing rate in response to pure tones was decreased and the frequency receptive fields were narrower. The responses to pup and adult calls were also reduced but the amount of mutual information (MI) per spike about the pup call's identity was increased in mother mice. The response latency to pup and adult calls was significantly shorter in mothers. Despite similarly decreased responses to guinea pig whistles, the response latency, and the MI per spike did not differ between virgins and mothers for these heterospecific vocalizations. Noise correlations between cortical recordings were decreased in mothers, suggesting that the firing rate of distant neurons was more independent from each other. Together, these results indicate that in the most commonly used mouse strain for behavioral studies, the discrimination of pup calls by auditory cortex neurons is more efficient during motherhood.

Albinism is a congenital disorder where misrouting of the optic nerves at the chiasm gives rise to abnormal visual field representations in occipital cortex. In typical human development, the left occipital cortex receives retinal input predominantly from the right visual field, and vice-versa. In albinism, there is a more complete decussation of optic nerve fibers at the chiasm, resulting in partial representation of the temporal hemiretina (ipsilateral visual field) in the contralateral hemisphere. In this study, we characterize the receptive field properties for these abnormal representations by conducting detailed fMRI population receptive field mapping in a rare subset of participants with albinism and no ocular nystagmus. We find a nasal bias for receptive field positions in the abnormal temporal hemiretina representation. In addition, by modelling responses to bilateral visual field stimulation in the overlap zone, we found evidence in favor of discrete unilateral receptive fields, suggesting a conservative pattern of spatial selectivity in the presence of abnormal retinal input.

Contour detection is a salient operation in many computer vision applications as it extracts features that are important for distinguishing objects in scenes. It is believed to be a primary role of simple cells in visual cortex of the mammalian brain. Many of such cells receive push-pull inhibition or surround suppression. We propose a computational model that exhibits a combination of these two phenomena. It is based on two existing models, which have been proven to be very effective for contour detection. In particular, we introduce a brain-inspired contour operator that combines push-pull and surround inhibition. It turns out that this combination results in a more effective contour detector, which suppresses texture while keeping the strongest responses to lines and edges, when compared to existing models. The proposed model consists of a Combination of Receptive Field (or CORF) model with push-pull inhibition, extended with surround suppression. We demonstrate the effectiveness of the proposed approach on the RuG and Berkeley benchmark data sets of 40 and 500 images, respectively. The proposed push-pull CORF operator with surround suppression outperforms the one without suppression with high statistical significance.

By using the multi-electrode array (MEA) recording technique in conjunction with white-noise checkerboard stimuli and reverse correlation methods, we studied modulatory actions of glycinergic and GABAergic interneurons on spatiotemporal profiles of ganglion cells (GCs) in dark-adapted mouse retinas. We found that application of 2 µM strychnine decreased receptive field center radii of GCs by a mean value of 11%, and shifted the GC receptive field (RF) centers by a mean distance of 28.3 µm. On the other hand, 200 µM picrotoxin + 100 µM bicuculline + 50 µM TPMPA increased GC receptive field center radii by a mean value of 19%, and shifted the GC RF centers by a mean distance of 53.7 µm. Glycinergic neurons in the mouse retina are narrow-field amacrine cells that have been shown to mediate ON-OFF crossover inhibitory synapses within the RGs’ RF center, therefore they may increase the size and shift the location of GC RF center by synergistic addition to bipolar cell inputs to GCs. GABAergic neurons are wide-field amacrine cells and horizontal cells that are known to mediate antagonistic surround responses of GCs, and thus they decrease the GCs’ RF center size. Our results suggest that a major global function of glycinergic and GABAergic interneurons in the mammalian retina is to provide the flexibility for adjusting the size and location of GCs’ RF centers. The apparent shifts of GC RF centers suggest that the synergistic addition by GlyACs and the surround inhibition by GABAergic interneurons are not spatially symmetrical within GC RFs.

Radiating pain is a significant feature of chronic musculoskeletal pain conditions such as radiculopathies, repetitive motion disorders and whiplash associated disorders. It is reported to be caused by the development of mechanically-sensitive ectopic receptive fields along intact nociceptor axons at sites of peripheral neuroinflammation (neuritis). Since inflammation disrupts axonal transport, we have hypothesised that anterogradely-transported mechanically sensitive ion channels accumulate at the site of disruption, which leads to axonal mechanical sensitivity (AMS). In this study, we have characterised the mechanical properties of the ectopic axonal receptive fields in the rat and have examined the contribution of mechanically sensitive ion channels to the development of AMS following neuritis and vinblastine-induced axonal transport disruption. In both models, there was a positive force–discharge relationship and mechanical thresholds were low (∼9 mN/mm2). All responses were attenuated by Ruthenium Red and FM1-43, which block mechanically sensitive ion channels. In both models, the transport of TRPV1 and TRPA1 was disrupted, and intraneural injection of agonists of these channels caused responses in neurons with AMS following neuritis but not vinblastine treatment. In summary, these data support a role for mechanically sensitive ion channels in the development of AMS.

Retinal ganglion cells (GCs) are important visual neurons which carry complex spatiotemporal information from the retina to higher visual centers in the brain. By taking advantage of pathway-specific knockout/mutant mice and multi-electrode array (MEA) recording techniques, we analyze contributions of rod and cone pathways to responsiveness, kinetics and receptive field profiles of GCs under scotopic and photopic conditions. Our data suggest: (1) Scotopic responses of some GCs require all three rod pathways, some require only the secondary and tertiary rod pathways, and others require only the tertiary rod pathway. (2) There are more responsive GCs in photopic conditions than responsive GCs in scotopic conditions. (3) Gap junctions slow down GCs’ scotopic light responses and increase GCs’ ratio of antagonistic to center inputs. (4) Cone pathways do not affect the kinetics but alter the ratio of antagonistic to center inputs of scotopic GC responses, and they speed up GCs photopic responses and alter the ratio of GCs’ antagonistic to center synaptic inputs and receptive field profiles. (5) Rod bipolar cells shorten response latency of ON GCs and increase the ratio of GCs’ antagonistic to center synaptic inputs. (6) Light adaptation speeds up GCs’ temporal processing and tunes GC photopic responses to higher frequencies, and the tertiary rod pathway plays a significant role in adaptation-induced TTP changes in some GCs. (7) GC RF center sizes are partially mediated by AIIACs and GC-GC coupling. (8) Connexin36 gap junctions and cone pathways alter synaptic circuits underlying antagonistic surround inputs to GCs in photopic conditions.

Most binocular neurons in the mammalian visual cortex show matched selectivity for light stimuli presented through either eye. A recent study tracked the responses of individual neurons in early visual cortex over time, revealing that matched binocular selectivity develops through major rearrangements of binocular visual circuits.

Group III/IV striated muscle afferents are small diameter sensory neurons that play important roles in reflexes and sensation. To date, the morphological features of physiologically characterised group III/IV muscular afferents have not been identified. Here, the electrophysiological and morphological characteristics of sensory neurons innervating striated muscles of the mouse abdominal wall were investigated, ex vivo. Extracellular recordings were made from subcostal nerve trunks innervating the muscles. A distinctive class of mechanosensitive afferents was identified by a combination of physiological features including sensitivity to local compression, saturating response to graded stretch and, in most cases, absence of spontaneous firing. Studies were restricted to these distinctive units. These units had conduction velocities averaging 14 ± 4 m/s (range: 8–20 m/s, n = 7); within the range of group III fibres in mice. Von Frey hairs were used to map receptive fields, which covered an area of 0.36 ± 0.18 mm2 (n = 7). In 7 preparations, biotinamide filling of recorded nerve trunks revealed a single axon in the marked receptive field, with distinctive axonal branching and terminations meandering through the connective tissue sandwiched between two closely associated muscle layers. These axons were not immunoreactive for CGRP (n = 7) and were not activated by application of capsaicin (1 µM, n = 14). All of these afferents were strongly activated by a “metabolite mix” containing lactate, adenosine triphosphate and reduced pH. Responses to mechanical stimuli and to metabolites were additive. We have characterised a distinctive class of mechano- and chemo-sensitive group III afferent endings associated with connective tissue close to muscle fibres.

Rainy images severely degrade visibility and make many computer vision algorithms invalid. Hence, it is necessary to remove rain streaks from a single image. In this paper, we propose a novel end-to-end deep learning based deraining method. Previous methods neglect the correlation between different layers with different receptive fields that loss a lot of important information. To better solve the problem, we develop a multi-level guided residual block that is the basic unit of our network. In this block, we utilize multi-level dilation convolutions to obtain different receptive fields and the layer with smaller receptive fields to guide the learning of larger receptive fields. Moreover, in order to reduce the model sizes, the parameters are shared among all multi-level guided residual blocks. Experiments illustrate that guided learning improves the deraining performance and the shared parameters strategy is also feasible. Quantitative and qualitative experimental results demonstrate the superiority of the proposed method compared with several state-of-the-art deraining methods.

Recent studies show that research on single image super-resolution (SISR) has achieved great success by using deep convolutional neural network (CNN). Different types of features obtained in deep CNN have different contribution. However, most of the previous models ignore the distinction between different features and deal with them in the same way, which affects the representational capacity of the models. On the other hand, receptive fields with different size capture diverse features from the input. Based on the above considerations, we propose a dual residual attention module (DRAM) network which concentrates on recovering the high-frequency details and sharing the information between two receptive fields of different sizes. We construct local information integration (LFI) module as the basic module to make full use of the local information. The LFI module is a cascade of several dual residual attention fusion (DRAF) blocks with a dense connection structure. The feature modulation can focus on important features and suppress unimportant ones. The evaluation results on five benchmark datasets demonstrate the superiority of our DRAM network against the state-of-the-art algorithms.

There is substantial anatomical segregation in the organization of the rodent barrel system – each whisker on the mystacial pad sends input to TC cells within a dedicated thalamic barreloid, which in turn innervates a corresponding cortical barrel, and RS cells within a barrel respond primarily to deflections of the corresponding whisker at the beginning of the dedicated transmission line (the principal whisker, PW). However, it is also well-established that barrel cells exhibit multiwhisker receptive fields (RFs), and display lower amplitude, longer latency responses to deflections of non-PWs (or adjacent whiskers, AWs). There is considerable controversy regarding the origin of such multiwhisker RFs; three possibilities include: (i) TC cells within a barreloid respond to multiple whiskers, and barrel RS cells simply inherit multiwhisker responses from their aligned barreloid; (ii) TC cells respond only to the PW, but individual barreloids innervate multiple barrels; (iii) multiwhisker responses of barrel cells arise from lateral corticocortical (barrel-to-barrel) synaptic transmission. Ablation studies attempting to pinpoint the source of RS cell AW responses are often contradictory (though experimental work tends to suggest possibilities (i) or (iii) to be most plausible), and hence it is important to carefully evaluate these hypotheses in terms of available physiological data on barreloid and barrel response dynamics. In this work, I employ a biologically detailed model of the rat barrel cortex to evaluate possibility (i) , and I show that, within the model, hypothesis (i) is capable of explaining a broad range of the available physiological data on responses to single (PW or AW) deflections and paired whisker deflections (AW deflection followed by PW deflection), as well as the dependence of such responses on the angular direction of whisker deflection. In particular, the model shows that barrel RS cells can exhibit AW direction tuning despite the fact that barreloid to barrel wiring has no systematic dependence on the AW direction preference of TC cells. Future modeling work will examine the other possibilities for the generation of multiwhisker RS cell RFs, and compare and contrast the different possible mechanisms within the context of available experimental data.

Due to the complicated structure and tough working environment of planetary gearboxes, intelligent identification of the health states based on the raw vibration signal is still a huge challenge in equipment maintenance. Aiming at this issue, an enhanced convolutional neural network (ECNN) with enlarged receptive fields was proposed in this paper. First, a one-dimensional convolutional layer was applied to enlarge receptive field preliminarily and capture the fault information within each group of adjacent points in the vibration signal. Then, several fused dilated convolutional layers were constructed to enlarge the receptive field further and capture the long distance dependencies of the raw signal comprehensively. At last, the raw vibration signals were directly fed into the developed ECNN to train the fault diagnosis model, and evaluate the ECNN model with the testing data. The experimental results demonstrated that the developed method can enhance the fault feature learning ability by enlarging the receptive fields twice, and achieved higher diagnosis accuracies than the traditional deep learning methods in fault diagnosis of planetary gearboxes.

Behavioral studies document a functional hand proximity effect: objects near the palm, but not the back of the hand, affect visual processing. Although visuotactile bimodal neurons integrate visual and haptic inputs, their receptive fields in monkey cortex encompass the whole hand, not just the palm. Using ERPs, we investigated whether hand function influenced the topology of integrated space around the hand. In a visual detection paradigm, target and non-target stimuli appeared equidistantly in front or in back of the hand. Equivalent N1 amplitudes were found for both conditions. P3 target versus non-target amplitude differences were greater for palm conditions. Hand proximity biased processing of visual targets equidistant from the hand early in processing. However, hand function biases emerged later when targets were selected for potential action. Thus, early hand proximity effects on object processing depend on sensory-reliant neural responses, whereas later multisensory integration depend more on the hand's functional expertise.

Synchronous gamma oscillations are believed to play a prominent role in the information processing of biological neural systems. Experimental observations have found that the frequency and power of gamma oscillations in the primary visual cortex (V1 zone) are regulated by the illumination contrast of visual stimulus. However, the underlying mechanism of how the synchronous oscillations depend on the illumination contrast has not been well explained. We propose a local excitatory/inhibitory (E/I) neuronal network of integrate-and-fire (IAF) neurons with the difference-of-Gaussians (DOG) receptive field to unveil this mechanism. Simulation results demonstrate that the higher the illumination contrast, the higher the frequency of gamma oscillations. The power of gamma oscillations also increases with the increase of illumination contrast. These results are consistent with the experimental findings. Keywords: Gamma oscillations, Illumination contrast, E/I network, Receptive field

Artificial vision using computational models that can mimic biological vision is an area of ongoing research. One of the main themes within this research is the study of the retina and in particular, retinal ganglion cells which are responsible for encoding the visual stimuli. A common approach to modelling the internal processes of retinal ganglion cells is the use of a linear - non-linear cascade model, which models the cell's response using a linear filter followed by a static non-linearity. However, the resulting model is generally restrictive as it is often a poor estimator of the neuron's response. In this paper we present an alternative to the linear - non-linear model by modelling retinal ganglion cells using a number of machine learning techniques which have a proven track record for learning complex non-linearities in many different domains. A comparison of the model predicted spike rate shows that the machine learning models perform better than the standard linear - non-linear approach in the case of temporal white noise stimuli.

Motion vision circuits of the fly are organized into parallel ON and OFF pathways tuned to the direction and speed of motion. Directionally selective small-field columnar neurons innervate the third optic ganglion, the lobula plate, where their outputs are integrated by large dendritic fields that encode wide-field background motion. Wide-field lobula plate cells, in turn, coordinate gaze stabilization. By contrast, behavioral responses to visual features is believed to begin within the circuits of the fourth optic ganglion, the lobula. We investigated 19 lobula columnar neurons (LCs) with stimuli that are behaviorally salient to the fly. A comprehensive analysis of receptive field properties showed that LC15 encodes moving bars, whereas LC12 is relatively agnostic to object dimensions. Neither responds to optic flow, rather object responses in both LCs are suppressed by background motion. Further contrasting neurons of the motion pathway, LC12 and LC15 are both small-field cells indifferent to ON-OFF contrast polarity. LC15 responds equally to four directions of motion, whereas LC12 is selective for motion in the horizontal plane. Both are equally sensitive to object speeds ranging from 22 to 132° per second. Small objects strongly stimulate presynaptic fibers of lobula layer 2, which corresponds to the postsynaptic domains of object-detecting LCs. Finally, in direct contrast to the entire motion pathway, we were surprised to find that octopamine nullifies the inhibition of object responses generated by background motion; LC12 and LC15 remain sensitive to objects appearing against a moving background.

SummaryModels of the cerebral cortex propose a canonical columnar architecture preserved across the brain. Evidence for this model primarily stems from observations in primary sensory areas, such as the visual cortex, where neurons in an anatomical column respond preferentially to the same stimulus features. Similar columnar feature selectivity has been described for visual area V2, visual area V3 [9, 10], the middle temporal visual area (area V5), and inferotemporal cortex. The fourth visual area (V4) features color selective neurons, organized in clusters across its surface. However, whether these color-sensitive neurons are organized in a columnar fashion is debated and uncertain due to technical limitations. Here, we utilize laminar multielectrode recordings along V4 columns to resolve rival hypotheses. Linear electrode arrays were introduced orthogonally into area V4, spanning all cortical layers. Monkeys viewed color stimuli presented to the visual receptive field of the column of neurons. We found that preference for a specific color was conserved along depth but varied across columns. Interestingly, while color preference (e.g. red or green) was consistent along depth, color selectivity was significantly greater in upper layers relative to the input layer and was weakest in the lower layers. These findings demonstrate that area V4 is organized in functional columns like other visual areas.

Variation in the neural code contributes to making each individual unique. We probed neural code variation using ∼100 neural population recordings from major ganglion cell types in the macaque retina, combined with an interpretable computational representation of individual variability using machine learning. This representation captured individual variation and covariation in properties such as nonlinearity, temporal dynamics, and spatial receptive field size, while preserving invariances, such as asymmetries between ON and OFF cells. The covariation of response properties in different cell types was associated with the proximity of lamination of their synaptic inputs. Surprisingly, male retinas exhibited higher firing rates and faster temporal integration than female retinas. Exploiting data from previously recorded macaque retinas enabled efficient characterization of a new macaque retina, and of a human retina. Simulations indicated that combining a vast dataset of healthy macaque recordings with behavioral feedback could be used to identify the neural code and thus improve retinal implants for vision restoration.

UNet adopting an encoder-decoder structure has been used widely in medical image segmentation tasks for its outstanding performance. However, in our work, we find that UNet has the worse segmentation performance of small masses. The reason behind this is that the sizes of receptive fields are limited. In this work, to address this issue, we develop a novel end-to-end model, Adaptive Receptive Field Network (ARF-Net), for the precise breast mass segmentation in whole mammographic images and ultrasound images. ARF-Net composes of an encoder network and a corresponding decoder network, followed by a pixel-wise classifier. In ARF-Net, a Selective Receptive Filed Module (SRFM) is proposed to allocate the suitable sizes of receptive fields to the breast masses of different sizes. SRFM consists of a Multiple Receptive Field Module (MRFM) for generating multiple receptive fields of different sizes and a Multi-Scale Selection Module (MSSM) for selecting the suitable sizes of receptive fields based on the objects’ size. The proposed ARF-Net achieves the dice index of 86.1 % , 85.75 % , and 88.12 % on the two mammographic databases (INbreast and CBIS-DDSM) and one ultrasonic database (UDIAT), respectively. Moreover, extensive ablation experiments show that ARF-Net transcends several state-of-the-art segmentation networks, and the developed MSSM exceeds several counterparts.

Retinal ganglion cells (RGCs) were one of the first classes of sensory neurons to be described in terms of a receptive field (RF). Over the last six decades, our understanding of the diversity of RGC types and the nuances of their response properties has grown exponentially. We will review the current understanding of RGC receptive fields (RFs) mostly from studies in mammals, but including work from other vertebrates as well. We will argue for a new paradigm that embraces the fluidity of RGC RFs with an eye toward the neuroethology of vision. Specifically, we will focus on (1) different methods for measuring RGC RFs, (2) RF models, (3) feature selectivity and the distinction between fluid and stable RF properties, and (4) ideas about the future of understanding RGC RFs.

The "blue-on" and "blue-off" receptive fields in retina and dorsal lateral geniculate nucleus (LGN) of diurnal primates combine signals from short-wavelength sensitive (S) cone photoreceptors with signals from medium/long wavelength sensitive (ML) photoreceptors. Three questions about this combination remain unresolved. Firstly, is the combination of S and ML signals in these cells linear or non-linear? Secondly, how does the timing of S and ML inputs to these cells influence their responses? Thirdly, is there spatial antagonism within S and ML subunits of the receptive field of these cells? We measured contrast sensitivity and spatial frequency tuning for four types of drifting sine gratings: S cone isolating, ML cone isolating, achromatic (S + ML), and counterphase chromatic (S - ML), in extracellular recordings from LGN of marmoset monkeys. We found that responses to stimuli which modulate both S and ML cones are well predicted by a linear sum of S and ML signals, followed by a saturating contrast-response relation. Differences in sensitivity and timing (i.e. vector combination) between S and ML inputs are needed to explain the amplitude and phase of responses to achromatic (S + ML) and counterphase chromatic (S - ML) stimuli. Best-fit spatial receptive fields for S and/or ML subunits in most cells (>80%) required antagonistic surrounds, usually in the S subunit. The surrounds were however generally weak and had little influence on spatial tuning. The sensitivity and size of S and ML subunits were correlated on a cell-by-cell basis, adding to evidence that blue-on and blue-off receptive fields are specialised to signal chromatic but not spatial contrast.

The center-surround receptive field structure, ubiquitous in the visual system, is hypothesized to be evolutionarily advantageous in image processing tasks. We address the potential functional benefits and shortcomings of spatial localization and center-surround antagonism in the context of an integrate-and-fire neuronal network model with image-based forcing. Utilizing the sparsity of natural scenes, we derive a compressive-sensing framework for input image reconstruction utilizing evoked neuronal firing rates. We investigate how the accuracy of input encoding depends on the receptive field architecture, and demonstrate that spatial localization in visual stimulus sampling facilitates marked improvements in natural scene processing beyond uniformly-random excitatory connectivity. However, for specific classes of images, we show that spatial localization inherent in physiological receptive fields combined with information loss through nonlinear neuronal network dynamics may underlie common optical illusions, giving a novel explanation for their manifestation. In the context of signal processing, we expect this work may suggest new sampling protocols useful for extending conventional compressive sensing theory.

Visual processing in cortex relies on feedback projections contextualising feedforward information flow. Primary visual cortex (V1) has small receptive fields and processes feedforward information at a fine-grained spatial scale, whereas higher visual areas have larger, spatially invariant receptive fields. Therefore, feedback could provide coarse information about the global scene structure or alternatively recover fine-grained structure by targeting small receptive fields in V1. We tested if feedback signals generalise across different spatial frequencies of feedforward inputs, or if they are tuned to the spatial scale of the visual scene. Using a partial occlusion paradigm, functional magnetic resonance imaging (fMRI) and multivoxel pattern analysis (MVPA) we investigated whether feedback to V1 contains coarse or fine-grained information by manipulating the spatial frequency of the scene surround outside an occluded image portion. We show that feedback transmits both coarse and fine-grained information as it carries information about both low (LSF) and high spatial frequencies (HSF). Further, feedback signals containing LSF information are similar to feedback signals containing HSF information, even without a large overlap in spatial frequency bands of the HSF and LSF scenes. Lastly, we found that feedback carries similar information about the spatial frequency band across different scenes. We conclude that cortical feedback signals contain information which generalises across different spatial frequencies of feedforward inputs.

The relevance of human primary motor cortex (M1) for motor actions has long been established. However, it is still unknown how motor actions are represented, and whether M1 contains an ordered somatotopy at the mesoscopic level. In the current study we show that a detailed within-limb somatotopy can be obtained in M1 during finger movements using Gaussian population Receptive Field (pRF) models. Similar organizations were also obtained for primary somatosensory cortex (S1), showing that individual finger representations are interconnected throughout sensorimotor cortex. The current study additionally estimates receptive field sizes of neuronal populations, showing differences between finger digit representations, between M1 and S1, and additionally between finger digit flexion and extension. Using the Gaussian pRF approach, the detailed somatotopic organization of M1 can be obtained including underlying characteristics, allowing for the in-depth investigation of cortical motor representation and sensorimotor integration.

In vivo calcium (Ca2+) imaging using two-photon microscopy allows activity to be monitored simultaneously from hundreds of individual neurons within a local population. While this allows us to gain important insights into how cortical neurons represent sensory information, factors such as photo-bleaching of the Ca2+ indicator limit imaging duration (and thus the numbers of stimuli that can be tested), which in turn hampers the full characterization of neuronal response properties. Here, we demonstrate that using an encoding model combined with presentation of natural movies results in detailed characterization of receptive field (RF) properties despite the relatively short time for data collection. During presentation of natural movie clips to macaque monkeys, we recorded fluorescence signals from primary visual cortex (V1) neurons that had been loaded with a Ca2+ indicator. For each recorded neuron, we constructed an encoding model that comprised an array of motion-energy filters that tiled over the RFs. We optimized the weight of each filter's output so that the linear sum of the outputs across the filters mimicked the neuron's Ca2+-signal responses. These models were able to predict the neural responses to a different set of natural movies with a significant degree of accuracy. Moreover, the orientation tunings of neurons simulated by the model were highly correlated with those experimentally obtained when grating stimuli were presented to the monkeys. The model predictions were also consistent with what is known about spatial frequency tunings, the structure of excitatory subfields of RFs (i.e., classical RFs), and functional maps for these RF properties in V1. Further analysis revealed a new aspect of V1 functional architecture; the extent and distribution of suppressive RF subfields varied among nearby neurons, while those for excitatory subfields were shared. Thus, applying our encoding-model analysis to two-photon Ca2+ imaging of neuronal responses to natural movies provides a reliable and efficient means of analyzing a wide range of RF properties in multiple neurons imaged in a local region.

Summary: For a proper understanding of neural circuit function, it is important to know which signals neurons relay to their downstream partners. Calcium imaging with genetically encoded calcium sensors like GCaMP has become the default approach for mapping these responses. How well such measurements represent the true neurotransmitter output of any given cell, however, remains unclear. Here, we demonstrate the viability of the glutamate sensor iGluSnFR for 2-photon in vivo imaging in Drosophila melanogaster and prove its usefulness for estimating spatiotemporal receptive fields in the visual system. We compare the results obtained with iGluSnFR with the ones obtained with GCaMP6f and find that the spatial aspects of the receptive fields are preserved between indicators. In the temporal domain, however, measurements obtained with iGluSnFR reveal the underlying response properties to be much faster than those acquired with GCaMP6f. Our approach thus offers a more accurate description of glutamatergic neurons in the fruit fly. : Optical Imaging; Sensory Neuroscience; Techniques in Neuroscience Subject Areas: Optical Imaging, Sensory Neuroscience, Techniques in Neuroscience

Gait recognition is an important issue currently. In this paper, we propose to combine deep features and hand-crafted representations into a globally trainable deep model. Specifically, a set of deep feature vectors are firstly extracted by a pre-trained CNN model from the input sequences. Then, a kernel function with respect to the fully connected vector is trained as the guiding weight of the respective receptive fields of the input sequences. Therefore, the hand-crafted features are extracted based on the guiding weight. Finally, the hand-crafted features and the deep features are combined into a unified deep network to complete classification. The optimized gait descriptor, termed as deep convolutional location weight descriptor (DLWD), is capable of effectively revealing the importance of different body parts to gait recognition accuracy. Experiments on two gait data sets (i.e., CASIA-B, OU-ISIR) show that our method outperforms the other existing methods for gait recognition.

Intracellular recording allows measurement and perturbation of the membrane potential of identified neurons with sub-millisecond and sub-millivolt precision. This gives intracellular recordings a unique capacity to provide rich information about individual cells (e.g., high-resolution characterization of inputs, outputs, excitability, and structure). Hence, such recordings can elucidate the mechanisms that underlie fundamental phenomena, such as brain state, sparse coding, gating, gain modulation, and learning. Technical developments have increased the range of behaviors during which intracellular recording methods can be employed, such as in freely moving animals and head-fixed animals actively performing tasks, including in virtual environments. Such advances, and the combination of intracellular recordings with genetic and imaging techniques, have enabled investigation of the mechanisms that underlie neural computations during natural and trained behaviors.

Optical aberrations of a lens introduce lens blur to photographed images. Lens blur is non-stationary with the amount and characteristics of blur varying depending on spatial pixel locations in an image. This work presents non-stationary deep networks for the restoration of non-stationary lens blur. Deep networks have relatively larger receptive fields. However, the receptive fields of stationary deep networks are not wide enough for the networks to cope with the non-stationarity of lens blur that span the entire image. We use spatial pixel locations as an additional input to networks to let the network utilize location dependent features to handle the non-stationarity. Experimental results show that even shallower non-stationary networks provide better performance than deeper stationary networks. The non-stationary networks are trained from pairs of images photographed at different aperture settings, eliminating the necessity of estimation or measurement of pixel-wise variant non-stationary lens blur.

Simple cells in the cat primary visual cortex usually have elongated receptive fields (RFs), and their orientation selectivity can be largely predicted by their RFs. As to the relay cells in cats' lateral geniculate nucleus (LGN), they also have weak but significant orientation bias (OB). It is thus of interest to investigate the fine spatiotemporal receptive field (STRF) properties in LGN, compare them with the dynamics of orientation tuning, and examine the dynamic relationship between STRF and orientation sensitivity in LGN. We mapped the STRFs of the LGN neurons in cats with white noise and characterized the dynamics of the orientation tuning by flashing gratings. We found that most of the LGN neurons showed elongated RFs and that the elongation axes were consistent with the preferred orientations. STRFs and the dynamics of orientation tuning were closely correlated temporally: the elongation of RFs and OB emerged, peaked and decayed at the same pace, with unchanged elongation axis of RF and preferred orientation but consistently changing aspect ratio of RF and OB strength across time. Importantly, the above consistency between RF and orientation tuning was not influenced by the ablation of the primary visual cortex. Furthermore, biased orientation tuning emerged 20-30 ms earlier than those in the primary visual cortex. These data demonstrated that similar to the primary visual cortex, the orientation sensitivity was closely reflected by the RF properties in LGN. However, the elongated RF and OB in LGN did not originate from the primary visual cortex feedback.

Most animals that have compound eyes determine object distances by using monocular cues, especially motion parallax. In artificial compound eye imaging systems inspired by natural compound eyes, object depths are typically estimated by measuring optic flow; however, this requires mechanical movement of the compound eyes or additional acquisition time. In this paper, we propose a method for estimating object depths in a monocular compound eye imaging system based on the computational compound eye (COMPU-EYE) framework. In the COMPU-EYE system, acceptance angles are considerably larger than interommatidial angles, causing overlap between the ommatidial receptive fields. In the proposed depth estimation technique, the disparities between these receptive fields are used to determine object distances. We demonstrate that the proposed depth estimation technique can estimate the distances of multiple objects.

In the visual system retinal space is compressed such that acuity decreases further from the fovea. Different forms of memory may rely on a compressed representation of time, manifested as decreased accuracy for events that happened further in the past. Neurophysiologically, "time cells" show receptive fields in time. Analogous to the compression of visual space, time cells show less acuity for events further in the past. Behavioral evidence suggests memory can be accessed by scanning a compressed temporal representation, analogous to visual search. This suggests a common computational language for visual attention and memory retrieval. In this view, time functions like a scaffolding that organizes memories in much the same way that retinal space functions like a scaffolding for visual perception.