Receptive Field Size Research Articles

Single transient intraocular pressure elevations cause prolonged retinal ganglion cell dysfunction and retinal capillary abnormalities in mice

Transient intraocular pressure (IOP) elevations are likely to occur in certain forms of glaucoma and after intravitreal injections to treat various retinal diseases. However, the impact of these transient IOP elevations on the physiology of individual retinal ganglion cells (RGCs) is unknown. In this report, we explore how transient IOP elevations in mice affect RGC physiology, RGC anatomy, and retinal arteriole and capillary structure. Transient IOP elevation was induced in 12-week old wild type C57BL6J mice by injecting sodium hyaluronate into the anterior chamber. IOP was measured immediately after the injection and again 1 and 7 days later. Average peak IOP after injection was ~50 mmHg and subsequent IOPs returned to normal. RGC physiology was assessed with a multielectrode array (MEA) by calculating a spike triggered average (STA) at the same time points. RGC counts and retinal vascular structure were assessed 14 days after injection with immunohistochemistry to label RGCs and blood vessels. Transient IOP elevation caused a marked reduction of scotopic STA presence and delayed center and surround STA peak times that did not recover. Transient IOP elevation also caused a reduced photopic receptive field size and spontaneous firing rate, both of which showed some recovery with time. Transient IOP elevation also induced vascular remodeling: the number of capillary branches was decreased within the superficial and intermediate vascular plexi. RGC counts, retinal arteriole diameter, and deep capillary plexus branching were unaffected. These previously unappreciated findings suggest that transient IOP elevation may cause unrecognized and potentially long-term pathology to RGCs and associated neurovascular units which should be accounted for in clinical practice.

Mouse visual cortex contains a region of enhanced spatial resolution

The representation of space in mouse visual cortex was initially thought to be relatively uniform, with no strong biases towards any particular region of space. This contrasts with the primate visual cortex with its over-representation of the fovea, placing potential limits on the translation of research in mice to humans. Here we reveal a previously unsuspected organization of the visual cortex of mice that resembles that fovea-centric organization of human visual cortex. Using population receptive-field (pRF) mapping techniques, which allow estimates to be made of aggregate receptive field sizes, we found that mouse visual cortex contains a region in which pRFs are considerably smaller. This region, the ‘focea’, represents a location in space directly in front of, and slightly above, the mouse. Using two-photon imaging we show that the smaller pRFs are due to a more orderly representation of space and an over-representation of binocular regions of the visual scene. We also show that RFs of single neurons in areas LM, AL and AM are smaller at the focea. Mice have improved visual resolution in this region of space and freely-moving mice make compensatory eye-movements to hold this region in front of them. Our results indicate that the representation of space in mouse visual cortex is non-uniform and mice have spatial biases in their visual processing. The presence of a focea has important implications for the use of the mouse model of vision.

DFD-Net: lung cancer detection from denoised CT scan image using deep learning

The availability of pulmonary nodules in CT scan image of lung does not completely specify cancer. The noise in an image and morphology of nodules, like shape and size has an implicit and complex association with cancer, and thus, a careful analysis should be mandatory on every suspected nodules and the combination of information of every nodule. In this paper, we introduce a “denoising first” two-path convolutional neural network (DFD-Net) to address this complexity. The introduced model is composed of denoising and detection part in an end to end manner. First, a residual learning denoising model (DR-Net) is employed to remove noise during the preprocessing stage. Then, a two-path convolutional neural network which takes the denoised image by DR-Net as an input to detect lung cancer is employed. The two paths focus on the joint integration of local and global features. To this end, each path employs different receptive field size which aids to model local and global dependencies. To further polish our model performance, in different way from the conventional feature concatenation approaches which directly concatenate two sets of features from different CNN layers, we introduce discriminant correlation analysis to concatenate more representative features. Finally, we also propose a retraining technique that allows us to overcome difficulties associated to the image labels imbalance. We found that this type of model easily first reduce noise in an image, balances the receptive field size effect, affords more representative features, and easily adaptable to the inconsistency among nodule shape and size. Our intensive experimental results achieved competitive results.

Long-term performance of Utah slanted electrode arrays and intramuscular electromyographic leads implanted chronically in human arm nerves and muscles

Objective. We explore the long-term performance and stability of seven percutaneous Utah Slanted Electrode Arrays (USEAs) and intramuscular recording leads (iEMGs) implanted chronically in the residual arm nerves and muscles of three human participants as a means to permanently restore sensorimotor function after transradial amputations. Approach. We quantify the number of functional recording and functional stimulating electrodes over time. We also calculate the signal-to-noise ratio (SNR) of USEA and iEMG recordings and quantify the stimulation current necessary to evoke detectable sensory percepts. Furthermore, we quantify the consistency of the sensory modality, receptive field location, and receptive field size of USEA-evoked percepts. Main results. In the most recent subject, involving USEAs with technical improvements, neural recordings persisted for 502 d (entire implant duration) and the number of functional recording electrodes for one USEA increased over time. However, for six out of seven USEAs across the three participants, the number of functional recording electrodes decreased within the first 2 months after implantation. The SNR of neural recordings and electromyographic recordings stayed relatively consistent over time. Sensory percepts were consistently evoked over the span of 14 months, were not significantly different in size, and highlighted the nerves’ fascicular organization. The percentage of percepts with consistent modality or consistent receptive field location between sessions (∼1 month apart) varied between 0%–86.2% and 9.1%–100%, respectively. Stimulation thresholds and electrode impedances increased initially but then remained relatively stable over time. Significance. This work demonstrates improved performance of USEAs, and provides a basis for comparing the longevity and stability of USEAs to that of other neural interfaces. USEAs provide a rich repertoire of neural recordings and sensory percepts. Although their performance still generally declines over time, functionality can persist long-term. Future work should leverage the results presented here to further improve USEA design or to develop adaptive algorithms that can maintain a high level of performance.

The optimisation of deep neural networks for segmenting multiple knee joint tissues from MRIs

Automated semantic segmentation of multiple knee joint tissues is desirable to allow faster and more reliable analysis of large datasets and to enable further downstream processing e.g. automated diagnosis.In this work, we evaluate the use of conditional Generative Adversarial Networks (cGANs) as a robust and potentially improved method for semantic segmentation compared to other extensively used convolutional neural network, such as the U-Net. As cGANs have not yet been widely explored for semantic medical image segmentation, we analysed the effect of training with different objective functions and discriminator receptive field sizes on the segmentation performance of the cGAN. Additionally, we evaluated the possibility of using transfer learning to improve the segmentation accuracy. The networks were trained on i) the SKI10 dataset which comes from the MICCAI grand challenge “Segmentation of Knee Images 2010″, ii) the OAI ZIB dataset containing femoral and tibial bone and cartilage segmentations of the Osteoarthritis Initiative cohort and iii) a small locally acquired dataset (Advanced MRI of Osteoarthritis (AMROA) study) consisting of 3D fat-saturated spoiled gradient recalled-echo knee MRIs with manual segmentations of the femoral, tibial and patellar bone and cartilage, as well as the cruciate ligaments and selected peri-articular muscles. The Sørensen–Dice Similarity Coefficient (DSC), volumetric overlap error (VOE) and average surface distance (ASD) were calculated for segmentation performance evaluation.DSC ≥ 0.95 were achieved for all segmented bone structures, DSC ≥ 0.83 for cartilage and muscle tissues and DSC of ≈0.66 were achieved for cruciate ligament segmentations with both cGAN and U-Net on the in-house AMROA dataset. Reducing the receptive field size of the cGAN discriminator network improved the networks segmentation performance and resulted in segmentation accuracies equivalent to those of the U-Net. Pretraining not only increased segmentation accuracy of a few knee joint tissues of the fine-tuned dataset, but also increased the network’s capacity to preserve segmentation capabilities for the pretrained dataset.cGAN machine learning can generate automated semantic maps of multiple tissues within the knee joint which could increase the accuracy and efficiency for evaluating joint health.

Lightweight Stacked Hourglass Network for Human Pose Estimation

Human pose estimation is a problem that continues to be one of the greatest challenges in the field of computer vision. While the stacked structure of an hourglass network has enabled substantial progress in human pose estimation and key-point detection areas, it is largely used as a backbone network. However, it also requires a relatively large number of parameters and high computational capacity due to the characteristics of its stacked structure. Accordingly, the present work proposes a more lightweight version of the hourglass network, which also improves the human pose estimation performance. The new hourglass network architecture utilizes several additional skip connections, which improve performance with minimal modifications while still maintaining the number of parameters in the network. Additionally, the size of the convolutional receptive field has a decisive effect in learning to detect features of the full human body. Therefore, we propose a multidilated light residual block, which expands the convolutional receptive field while also reducing the computational load. The proposed residual block is also invariant in scale when using multiple dilations. The well-known MPII and LSP human pose datasets were used to evaluate the performance using the proposed method. A variety of experiments were conducted that confirm that our method is more efficient compared to current state-of-the-art hourglass weight-reduction methods.

Application of binocular disparity and receptive field dynamics: A biologically-inspired model for contour detection

Neurophysiological evidence demonstrates that classical receptive field responses in the primary visual cortex can be modulated by the non-classical receptive field. Although models based on the non-classical receptive field have been proposed, which has not employed the two following characteristics: dynamic regulation of the receptive field under external stimuli and depth determination by binocular cells. We propose a model that includes processing of gray-scale images and depth images. Luminance of grayscale image response was preprocessed using weighted least squares filtering; luminance of grayscale image and disparity information of depth image responses were obtained by V1 neuron responses. The receptive field size of each pixel used to calculate the luminance and disparity response is determined by the depth information. Luminance and disparity information responses were combined based on the optimal orientation, and robust contour maps were developed. Experimental results show that the proposed contour detection model outperforms biologically inspired models.

A DNN-based semantic segmentation for detecting weed and crop

Weed control is a global issue, and has attracted great attention in recent years. Deploying autonomous robots for weed removal has great potential in terms of constructing environment-friendly agriculture, and saving manpower. In this paper, we propose a weed/crop segmentation network that provides better performance for precisely recognizing the weed with arbitrary shape in complex environment condition, and offers great support for autonomous robots to successfully reduce the density of weed. Our deep neural network (DNN)-based segmentation model obtains persistent improvements by integrating four additional components. i) Hybrid dilated convolution and DropBlock are introduced into the classification backbone network, where the hybrid dilated convolution enlarges the receptive field, while DropBlock regularizes the weight parameters to learn robust features by random drops contiguous regions. ii) A universal function approximation block is added to the front-end of the backbone network, which adaptively converts the existing RGB-NIR bands into optimized (RGB + NIR)-based indices to increase the classification performance. iii) The bridge attention block is exploited, in order to make the network “globally” refer to the correlated region, regardless of the distance for capturing the rich long-range contextual information. iv) The spatial pyramid refinement block is inserted to fuse multi-scale feature maps with different size of receptive fields to provide the precise localization of segmentation result, by maintaining the consistency of feature maps. We evaluate our network performance on two challenging Stuttgart and Bonn datasets. The state-of-the-art performance on the two datasets shows that each added component has notable potential to boost the segmentation accuracy.

A Deep Multiscale Spatiotemporal Network for Assessing Depression From Facial Dynamics

Recently, deep learning models have been successfully employed in many video-based affective computing applications (e.g., detecting pain, stress, and Alzheimer’s disease). One key application is automatic depression recognition – recognition of facial expressions associated with depressive behaviour. State-of-the-art deep learning algorithms to recognize depression typically explore spatial and temporal information individually, by using 2D convolutional neural networks (CNNs) to analyze appearance information, and then by either mapping facial feature variations or averaging the depression level over video frames. This approach has limitations in terms of its ability to represent dynamic information that can help to accurately discriminate between depression levels. In contrast, models based on 3D CNNs allow to directly encode the spatio-temporal relationships, although these models rely on temporal information with fixed range and single receptive field. This approach limits the ability to capture variations of facial expression with diverse ranges, and the exploitation of diverse facial areas. In this article, a novel 3D CNN architecture – the Multiscale Spatiotemporal Network (MSN) – is introduced to effectively represent facial information related to depressive behaviours from videos. The basic structure of the model is composed of parallel convolutional layers with different temporal depths and sizes of receptive field, which allows the MSN to explore a wide range of spatio-temporal variations in facial expressions. Experimental results on two benchmark datasets show that our MSN architecture is effective, outperforming state-of-the-art methods in automatic depression recognition.

Learnable pooling weights for facial expression recognition

Abstract Pooling layers are spatial down-sampling layers used in convolutional neural networks (CNN) to gradually downscale the feature map, increase the receptive field size and reduce the number of the parameters in the model. The use of pooling layers leads to less computing complexity and memory consumption reduction but also introduces invariance to certain filter distortions which may induce subtle detail loss. This behaviour is undesired for some fine-grained recognition tasks such as facial expression recognition (FER) which highly relies on specific regional distortion detection. In this paper, we introduce a more filter distortion aware pooling layer based on kernel functions. The proposed pooling reduces the feature map dimensions while keeping track of the majority of the information fed to the next layer instead of ignoring part of them. The experiments on RAF, FER2013 and ExpW databases demonstrate the benefits of such layer and show that our model achieves competitive results with respect to the state-of-the-art approaches.

Accurate and fast single shot multibox detector

With the development of deep learning, the performance of object detection has made great progress. However, there are still some challenging problems, such as the detection accuracy of small objects and the efficiency of the detector. This study proposes an accurate and fast single shot multibox detector, which includes context comprehensive enhancement (CCE) module and feature enhancement module (FEM). To integrate more efficient information when aggregating context information, the conv4_3 and fc_7 feature maps are merged to design the CCE module. To obtain more fine‐grained feature information, this study presents a FEM and special feature enhancement module (FEM‐s) module that can fuse different receptive field sizes to better adapt to the scale change of the object. Compared to existing methods based on deep learning, the proposed method helps to gradually produce more detailed feature maps with better performance. Under the premise of ensuring real‐time speed, the authors network can achieve 81.2 mean average precision on the PASCAL VOC 2007 test with an input size of 320 × 320 on a single Nvidia 2080Ti GPU.

Arbitrary-shaped text detection with adaptive convolution and path enhancement pyramid network

Recently, scene text detection has become an active research field, which is an essential component of scene text reading. Especially, segmentation-based methods are commonly used, since the segmentation results can describe text of arbitrary shape. However, curve texts have a diversity of shapes, scales and orientations, which are difficult to locate, so the detector requires to adjust the local receptive fields size adaptively, which can aggregate multi-scale spatial information to accurately locate the curve text instance. Moreover, the low-level features are critical for localizing large text instances. When using Feature Pyramid Network (FPN) for multi-scale feature fusion, it will prevent the flow of accurate localization signals due to the long path from low-level to top-level. In order to solve these two problems, this paper proposes an Adaptive Convolution and Path Enhancement Pyramid Network (ACPEPNet), which can more accurately locate the text instances with arbitrary shapes. Firstly, an Adaptive Convolution Unit is introduced to improve the ability of backbone to aggregate multi-scale spatial information at the same stage. Specially, this unit is a lightweight component and without the cost of computations, based on this component we present a backbone network for text features extraction. Secondly, the original FPN structure is redesigned to build a short path from the low-level to top-level, in this way, we modify the path from one-way flow to two-way flow and add original features to the final stage of information fusion. Experiments on CTW1500, Total-Text, ICDAR 2015 and MSRA-TD500 validate the robustness of the proposed method. When there is no bells and whistles, this method achieves an F-measure of 80.8% without external training data on CTW1500.

Interocular ND filter suppression: Eccentricity and luminance polarity effects.

The depth and extent of interocular suppression were measured in binocularly normal observers who unilaterally adapted to neutral density (ND) filters (0, 1.5, 2, and 3 ND). Suppression was measured by dichoptically matching sectors of a ring presented to the adapted eye to a fixed contrast contiguous ring presented to the non-adapted eye. Other rings of alternating polarity were viewed binocularly. Rings were defined by luminance (L), luminance with added dynamic binary luminance noise (LM), and contrast modulating the same noise (CM). Interocular suppression depth increased with increasing ND, nearing significance (p = 0.058) for 1.5 ND. For L and LM stimuli, suppression depth across eccentricity (±12° visual field) differed for luminance increment (white) versus luminance decrement (black) stimuli, potentially confounding eccentricity results. Suppression for increment-only (white) luminance stimuli was steeper centrally and extended across the visual field, but was deeper for L than for LM stimuli. Suppression for decrement-only (black) luminance stimuli revealed only central suppression. Suppression was deeper with CM than LM stimuli, suggesting that CM stimuli are extracted in areas receiving predominantly binocular input which may be more sensitive to binocular disruption. Increment (white) luminance stimuli demonstrate deeper interocular suppression in the periphery than decrement (black) stimuli, so they are more sensitive to changes in peripheral suppression. Asymmetry of suppression in the periphery for opposite polarity luminance stimuli may be due to interocular receptive field size mismatch as a result of dark adaptation separately affecting ON and OFF pathways. Clinically, measurement of suppression with CM stimuli may provide the best information about post-combination binocularity.

Investigating the Reliability of Population Receptive Field Size Estimates Using fMRI.

In functional MRI (fMRI), population receptive field (pRF) models allow a quantitative description of the response as a function of the features of the stimuli that are relevant for each voxel. The most popular pRF model used in fMRI assumes a Gaussian shape in the features space (e.g., the visual field) reducing the description of the voxel’s pRF to the Gaussian mean (the pRF preferred feature) and standard deviation (the pRF size). The estimation of the pRF mean has been proven to be highly reliable. However, the estimate of the pRF size has been shown not to be consistent within and between subjects. While this issue has been noted experimentally, here we use an optimization theory perspective to describe how the inconsistency in estimating the pRF size is linked to an inherent property of the Gaussian pRF model. When fitting such models, the goodness of fit is less sensitive to variations in the pRF size than to variations in the pRF mean. We also show how the same issue can be considered from a bias-variance perspective. We compare different estimation procedures in terms of the reliability of their estimates using simulated and real fMRI data in the visual (using the Human Connectome Project database) and auditory domain. We show that, the reliability of the estimate of the pRF size can be improved considering a linear combination of those pRF models with similar goodness of fit or a permutation based approach. This increase in reliability of the pRF size estimate does not affect the reliability of the estimate of the pRF mean and the prediction accuracy.

Effect of electrical stimulation of receptive fields in people with lower limb amputation on variables of gait

People with amputation may perceive phantom limb sensations or pain in the amputated body part when ipsilateral body-regions are stimulated. These body-regions are called receptive fields. This study assessed whether receptive fields change in size and position over the course of one month in people with trans-tibial amputation and whether electrical stimulation of these fields in synchrony with walking affects phantom sensations and variables of gait. Thirty-one subjects participated in this study. Receptive fields were mapped seven times over a one month period. Thereafter, the effect of electrical stimulation in synchrony with walking was compared to placebo stimulation in an acute setting with a randomized, single-blind gait analysis in 18 participants. Results showed that receptive field size and position presented an adequate degree of consistency (difference in point of first response position of 4.9±4.8cm and overlap of total receptive field area of 54.3±35.0 %) for future use of electrical stimulation. Gait parameters for everyday activities (speed, gait width, % stance and swing phase) as well as perception of phantom pain were not altered to a clinically relevant degree by electrical stimulation and no negative effects were reported. In conclusion: Location and size of receptive fields are consistent enough for daily electrical stimulation without laborious daily assessment. If applied acutely, no significant effect on gait or pain could be detected. However, results are promising enough to test chronic application of electrical stimulation during gait in a long-term setting.

Deep neural networks capture texture sensitivity in V2.

Deep convolutional neural networks (CNNs) trained on visual objects have shown intriguing ability to predict some response properties of visual cortical neurons. However, the factors (e.g., if the model is trained or not, receptive field size) and computations (e.g., convolution, rectification, pooling, normalization) that give rise to such ability, at what level, and the role of intermediate processing stages in explaining changes that develop across areas of the cortical hierarchy are poorly understood. We focused on the sensitivity to textures as a paradigmatic example, since recent neurophysiology experiments provide rich data pointing to texture sensitivity in secondary (but not primary) visual cortex (V2). We initially explored the CNN without any fitting to the neural data and found that the first two layers of the CNN showed qualitative correspondence to the first two cortical areas in terms of texture sensitivity. We therefore developed a quantitative approach to select a population of CNN model neurons that best fits the brain neural recordings. We found that the CNN could develop compatibility to secondary cortex in the second layer following rectification and that this was improved following pooling but only mildly influenced by the local normalization operation. Higher layers of the CNN could further, though modestly, improve the compatibility with the V2 data. The compatibility was reduced when incorporating random rather than learned weights. Our results show that the CNN class of model is effective for capturing changes that develop across early areas of cortex, and has the potential to help identify the computations that give rise to hierarchical processing in the brain (code is available in GitHub).

OS-LFFD: a light and fast face detector with Ommateum structure

Face detection has been deployed on edge devices as the basis for face applications, but the devices cannot store large-scale models and have low computing power. The existing anchor-based face detection schemes cannot cover face images over a continuous size range, and their performance is not satisfactory. Obviously, good performances are accompanied by increased storage and lower speed. We find that the feature points in different layers correspond to a specific size range of RFs (receptive fields). According to the survey, the predictable range of RFs with the same size is the face on a continuous scale. Therefore, we argue that RFs are inherent anchors. A Light and Fast Face Detector with an Ommateum Structure (OS-LFFD) is proposed in this paper. By analyzing the correlation between the effective receptive field (ERF) and face sizes, a 4-branch network is designed to cover the objective range of face sizes. Each branch involves an ommateum block with a similar structure and shared parameters. It reduces the number of model parameters (8 M), which makes it much smaller than most face detectors. Experiments on the popular benchmarks WIDER FACE and FDDB using multiple hardware platforms demonstrate that the proposed scheme can considerably balance the accuracy and running speed.

A Simplified Model for Simulating Population Responses of Tactile Afferents and Receptors in the Skin.

Tactile information about an object can only be extracted from population responses of tactile receptors and their afferents. Thus, to best control tactile information in robots, neuroprostheses or haptic devices, inputs should represent responses from full populations of afferents. Here, we describe a simplified model that recreates afferent population responses of thousands of tactile afferents in a personal computer. The whole model includes a resistance network model to simplify the skin mechanics and an improved version of a single unit model that we have previously described. The whole model was implemented by short and efficient python code. The parameters of the model were fit based on a simple vibrating stimulus, but the simulated outputs generalize to match receptive field sizes, edge enhancement, and neurophysiological responses to dot textures, embossed letters and curved surfaces. We discuss how to use this work to model haptic perception and provide guidance in designing and controlling highly realistic tactile interfaces in robots, neural prostheses and haptic devices.

Kernel Attention Network for Single Image Super-Resolution

Recently, attention mechanisms have shown a developing tendency toward convolutional neural network (CNN), and some representative attention mechanisms, i.e., channel attention (CA) and spatial attention (SA) have been fully applied to single image super-resolution (SISR) tasks. However, the existing architectures directly apply these attention mechanisms to SISR without much consideration of the nature characteristic, resulting in less strong representational power. In this article, we propose a novel kernel attention module (KAM) for SISR, which enables the network to adjust its receptive field size corresponding to various scales of input by dynamically selecting the appropriate kernel. Based on this, we stack multiple kernel attention modules with group and residual connection to constitute a novel architecture for SISR, which enables our network to learn more distinguishing representations through filtering the information under different receptive fields. Thus, our network is more sensitive to multi-scale features, which enables our single network to deal with multi-scale SR task by predefining the upscaling modules. Besides, other attention mechanisms in super-resolution are also investigated and illustrated in detail in this article. Thanks to the kernel attention mechanism, the extensive benchmark evaluation shows that our method outperforms the other state-of-the-art methods.

Realistic Mathematical Model of Retinal Outer Plexiform Layer for Edge Detection.

Edge information is essential for object recognition and motion detection. It is reported that photoreceptors, horizontal cells and bipolar cells in the outer retina involved for edge detection. Moreover, it is known that the center and the surround receptive field structure found in the bipolar cell layer is thought to be related to an initial process of edge detection. In the present study, we constructed retinal network models including photoreceptors, horizontal cells, and bipolar cells using single-compartment neurons to investigate those contributions for edge detection. We simulate fixation of a natural image with changing the size of the horizontal cell receptive field and confirmed that the constructed model successfully extracts edges in the image. Furthermore, most of the edge in the scene is extracted when the size of the horizontal cell receptive field matched with that reported in anatomical evidence. To evaluate the performance of edge detection, we compare the result of edge detection with the Canny algorithm. As a result, we conformed that the model well detects fine edges similar to the Canny edge detection.