COVID-19 classification of X-ray images using deep neural networks.

OBJECTIVES: In the midst of the coronavirus disease 2019 (COVID-19) outbreak, chest X-ray (CXR) imaging is playing an important role in diagnosis and monitoring of patients with COVID-19. We propose a deep learning model for detection of COVID-19 from CXRs, as well as a tool for retrieving similar patients according to the model's results on their CXRs. For training and evaluating our model, we collected CXRs from inpatients hospitalized in four different hospitals. METHODS: In this retrospective study, 1384 frontal CXRs, of COVID-19 confirmed patients imaged between March and August 2020, and 1024 matching CXRs of non-COVID patients imaged before the pandemic, were collected and used to build a deep learning classifier for detecting patients positive for COVID-19. The classifier consists of an ensemble of pre-trained deep neural networks (DNNS), specifically, ReNet34, ReNet50¸ ReNet152, and vgg16, and is enhanced by data augmentation and lung segmentation. We further implemented a nearest-neighbors algorithm that uses DNN-based image embeddings to retrieve the images most similar to a given image. RESULTS: Our model achieved accuracy of 90.3%, (95% CI: 86.3-93.7%) specificity of 90% (95% CI: 84.3-94%), and sensitivity of 90.5% (95% CI: 85-94%) on a test dataset comprising 15% (350/2326) of the original images. The AUC of the ROC curve is 0.96 (95% CI: 0.93-0.97). CONCLUSION: We provide deep learning models, trained and evaluated on CXRs that can assist medical efforts and reduce medical staff workload in handling COVID-19. KEY POINTS: • A machine learning model was able to detect chest X-ray (CXR) images of patients tested positive for COVID-19 with accuracy and detection rate above 90%. • A tool was created for finding existing CXR images with imaging characteristics most similar to a given CXR, according to the model's image embeddings.

Optogenetic assessment of horizontal interactions in primary visual cortex.

Columnar organization of orientation selectivity and clustered horizontal connections linking orientation columns are two of the distinctive organizational features of primary visual cortex in many mammalian species. However, the functional role of these connections has been harder to characterize. Here we examine the extent and nature of horizontal interactions in V1 of the tree shrew using optical imaging of intrinsic signals, optogenetic stimulation, and multi-unit recording. Surprisingly, we find the effects of optogenetic stimulation depend primarily on distance and not on the specific orientation domains or axes in the cortex, which are stimulated. In addition, across a wide range of variation in both visual and optogenetic stimulation we find linear addition of the two inputs. These results emphasize that the cortex provides a rich substrate for functional interactions that are not limited to the orientation-specific interactions predicted by the monosynaptic distribution of horizontal connections.

Dendritic end inhibition in large-field visual neurons of the fly.

The extraction of optic flow fields by visual systems is crucial for course stabilization during locomotion, and relies on feedforward and lateral integration of visual inputs. Here we report a novel form of systemic, motion-sensitive lateral suppression in the dendrites of large, flow-field-selective neurons in the fly's visual lobes. Using in vivo Calcium-imaging and intracellular recordings, we demonstrate that responses in dendrites, but not axon terminals, are end inhibited by flanking gratings both in the vertical and horizontal systems. We show evidence for a mechanism involving wide-field dendritic inhibition that exceeds the retinotopic spatial extent of the dendrites. Using compartmental modeling, we point out a possible function in enhancing selectivity for optic flow fields. Our results suggest that lateral suppression is a common element serving similar functional requirements in different visual systems.

Cellular resolution circuit mapping with temporal-focused excitation of soma-targeted channelrhodopsin.

We describe refinements in optogenetic methods for circuit mapping that enable measurements of functional synaptic connectivity with single-neuron resolution. By expanding a two-photon beam in the imaging plane using the temporal focusing method and restricting channelrhodopsin to the soma and proximal dendrites, we are able to reliably evoke action potentials in individual neurons, verify spike generation with GCaMP6s, and determine the presence or absence of synaptic connections with patch-clamp electrophysiological recording.

Becoming a mother-circuit plasticity underlying maternal behavior.

The transition to motherhood is a dramatic event during the lifetime of many animals. In mammals, motherhood is accompanied by hormonal changes in the brain that start during pregnancy, followed by experience dependent plasticity after parturition. Together, these changes prime the nervous system of the mother for efficient nurturing of her offspring. Recent work has described how neural circuits are modified during the transition to motherhood. Here we discuss changes in the auditory cortex during motherhood as a model for maternal plasticity in sensory systems. We compare classical plasticity paradigms with changes that arise naturally in mothers, highlighting current efforts to establish a mechanistic understanding of plasticity and its different components in the context of maternal behavior.

The development of cortical circuits for motion discrimination.

Stimulus discrimination depends on the selectivity and variability of neural responses, as well as the size and correlation structure of the responsive population. For direction discrimination in visual cortex, only the selectivity of neurons has been well characterized across development. Here we show in ferrets that at eye opening, the cortical response to visual stimulation exhibits several immaturities, including a high density of active neurons that display prominent wave-like activity, a high degree of variability and strong noise correlations. Over the next three weeks, the population response becomes increasingly sparse, wave-like activity disappears, and variability and noise correlations are markedly reduced. Similar changes were observed in identified neuronal populations imaged repeatedly over days. Furthermore, experience with a moving stimulus was capable of driving a reduction in noise correlations over a matter of hours. These changes in variability and correlation contribute significantly to a marked improvement in direction discriminability over development.

Different receptive fields in axons and dendrites underlie robust coding in motion-sensitive neurons.

In the visual system of the blowfly Calliphora vicina, neurons of the vertical system (VS cells) integrate wide-field motion information from a retinotopic array of local motion detectors. In vivo calcium imaging reveals two distinct and separate receptive fields in these cells: a narrow dendritic receptive field corresponding to feedforward input from the local motion detectors and a broad axon terminal receptive field that additionally incorporates input from neighboring cells mediated by lateral axo-axonal gap junctions. We show that the axon terminal responses are linear interpolations of the dendritic responses, resulting in a robust population coding of optic flow parameters as predicted by previous modeling studies. Compartmental modeling shows that spatially separating the axonal gap junctions from the conductive load of the dendritic synapses increases the coupling strength of the gap junctions, making this interpolation possible.