Biology subjects

Rokem, A.

Publications and source records attributed to Rokem, A..

14 recordsLinked to original sources

bioRxiv2018

A model of ganglion axon pathways accounts for percepts elicited by retinal implants

Retinal prostheses, now implanted in over 250 patients worldwide, electrically stimulate surviving cells in order to evoke neuronal responses that are interpreted by the brain as visual percepts ( phosphenes). However, instead of seeing focal spots of light, current implant users perceive highly distorted phosphenes that vary in shape both across subjects and electrodes. We characterized these distortions by asking users of the Argus retinal prosthesis system (Second Sight Medical Products) to draw elicited percepts on a touchscreen. We found that phosphene shape could be accurately predicted by simulating the topographic organization of nerve fiber bundles in each subjects retina. Our model shows that activation of ganglion axons contributes to a rich repertoire of phosphene shapes, successfully replicating percepts ranging from blobs to oriented streaks and wedges depending on electrode location. This work provides a first step towards future devices that incorporate stimulation strategies tailored to each individual patients retinal neurophysiology.\n\nOne Sentence SummaryWe show that the perceptual experience of retinal implant users can be accurately predicted using a computational model that simulates the topographic organization of each individual patients retinal ganglion axon pathways.

neuroscience↗

bioRxiv2018

Combining citizen science and deep learning to amplify expertise in neuroimaging

Research in many fields has become increasingly reliant on large and complex datasets. \"Big Data\" holds untold promise to rapidly advance science by tackling new questions that cannot be answered with smaller datasets. While powerful, research with Big Data poses unique challenges, as many standard lab protocols rely on experts examining each one of the samples. This is not feasible for large-scale datasets because manual approaches are time-consuming and hence difficult to scale. Meanwhile, automated approaches lack the accuracy of examination by highly trained scientists and this may introduce major errors, sources of noise, and unforeseen biases into these large and complex datasets. Our proposed solution is to 1) start with a small, expertly labelled dataset, 2) amplify labels through web-based tools that engage citizen scientists, and 3) train machine learning on amplified labels to emulate expert decision making. As a proof of concept, we developed a system to quality control a large dataset of three-dimensional magnetic resonance images (MRI) of human brains. An initial dataset of 200 brain images labeled by experts were amplified by citizen scientists to label 722 brains, with over 80,000 ratings done through a simple web interface. A deep learning algorithm was then trained to predict data quality, based on a combination of the citizen scientist labels that accounts for differences in the quality of classification by different citizen scientists. In an ROC analysis (on left out test data), the deep learning network performed as well as a state-of-the-art, specialized algorithm (MRIQC) for quality control of T1-weighted images, each with an area under the curve of 0.99. Finally, as a specific practical application of the method, we explore how brain image quality relates to the replicability of a well established relationship between brain volume and age over development. Combining citizen science and deep learning can generalize and scale expert decision making; this is particularly important in emerging disciplines where specialized, automated tools do not already exist.

neuroscience↗

bioRxiv2018

Applying biophysical models to understand the role of white matter in cognitive development

Diffusion MRI (dMRI) holds great promise for illuminating the biological changes that underpin cognitive development. The diffusion of water molecules probes the cellular structure of brain tissue, and biophysical modeling of the diffusion signal can be used to make inferences about specific tissue properties that vary over development or predict cognitive performance. However, applying these models to study development requires that the parameters can be reliably estimated given the constraints of data collection with children. Here we collect repeated scans using a multi-shell diffusion MRI protocol in a group of children (ages 7-12) and use two popular biophysical models to characterize axonal properties. We first assess the scan-rescan reliability of model parameters and show that axon water faction can be reliably estimated from a relatively fast acquisition, without applying spatial smoothing or de-noising. We then investigate developmental changes in the white matter, and individual differences in white matter that correlate with reading skill. Specifically, we test the hypothesis that previously reported correlations between reading skill and diffusion anisotropy in the corpus callosum reflect increased axon density in poor readers. Both models support this interpretation, highlighting the utility of biophysical models for testing specific hypotheses about cognitive development.

neuroscience↗

bioRxiv2018

Multilabel multiclass classification of OCT images augmented with age, gender and visual acuity data

Optical Coherence Tomography (OCT) imaging of the retina is in widespread clinical use to diagnose a wide range of retinal pathologies and several previous studies have used deep learning to create systems that can accurately classify retinal OCT as indicative of one of these pathologies. However, patients often exhibit multiple pathologies concurrently. Here, we designed a novel neural network algorithm that performs multiclass and multilabel classification of retinal images from OCT images in four common retinal pathologies: epiretinal membrane, diabetic macular edema, dry age-related macular degeneration and neovascular age-related macular degeneration. Furthermore, clinicians often also use additional information about the patient for diagnosis. Second contribution of this study is improvement of multiclass, multilabel classification augmented with information about the patient: age, visual acuity and gender. We compared two training strategies: a network pre-trained with ImageNet was used for transfer learning, or the network was trained from randomly initialized weights. Transfer learning does not perform better in this case, because many of the low-level filters are tuned to colors, and the OCT images are monochromatic. Finally, we provide a transparent and interpretable diagnosis by highlighting the regions recognized by the neural network.

physiology↗

bioRxiv2018

Forecasting Future Humphrey Visual Fields Using Deep Learning

PurposeTo determine if deep learning networks could be trained to forecast a future 24-2 Humphrey Visual Field (HVF).\n\nDesignRetrospective database study.\n\nParticipantsAll patients who obtained a HVF 24-2 at the University of Washington.\n\nMethodsAll datapoints from consecutive 24-2 HVFs from 1998 to 2018 were extracted from a University of Washington database. Ten-fold cross validation with a held out test set was used to develop the three main phases of model development: model architecture selection, dataset combination selection, and time-interval model training with transfer learning, to train a deep learning artificial neural network capable of generating a point-wise visual field prediction.\n\nMain outcome measuresMean absolute error (MAE) and difference in Mean Deviation (MD) between predicted and actual future HVF.\n\nResultsMore than 1.7 million perimetry points were extracted to the hundredth decibel from 32,443 24-2 HVFs. The best performing model with 20 million trainable parameters, CascadeNet-5, was selected. The overall MAE for the test set was 2.47 dB (95% CI: 2.45 dB to 2.48 dB). The 100 fully trained models were able to successfully predict progressive field loss in glaucomatous eyes up to 5.5 years in the future with a correlation of 0.92 between the MD of predicted and actual future HVF (p < 2.2 = 10-16) and an average difference of 0.41 dB.\n\nConclusionsUsing unfiltered real-world datasets, deep learning networks show an impressive ability to not only learn spatio-temporal HVF changes but also to generate predictions for future HVFs up to 5.5 years, given only a single HVF.

bioinformatics↗

bioRxiv2018

MRI2MRI: A deep convolutional network that accurately transforms between brain MRI contrasts

Different brain MRI contrasts represent different tissue properties and are sensitive to different artifacts. The relationship between different contrasts is therefore complex and nonlinear. We developed a deep convolutional network that learns the mapping between different MRI contrasts. Using a publicly available dataset, we demonstrate that this algorithm accurately transforms between T1- and T2-weighted images, proton density images, time-of-flight angiograms, and diffusion MRI images. We demonstrate that these transformed images can be used to improve spatial registration between MR images of different contrasts.

neuroscience↗

bioRxiv2018

White matter plasticity and reading instruction: Widespread anatomical changes track the learning process

White matter tissue properties correlate with childrens performance across domains ranging from reading, to math, to executive function. We use a longitudinal intervention design to examine experience-dependent growth in reading skills and white matter in a group of grade school aged, struggling readers. Diffusion MRI data were collected at regular intervals during an 8-week, intensive reading intervention. These measurements reveal large-scale changes throughout a collection of white matter tracts, in concert with growth in reading skill. Additionally, we identify tracts whose properties predict reading skill but remain fixed throughout the intervention, suggesting that some anatomical properties may stably predict the ease with which a child learns to read, while others dynamically reflect the effects of experience. These results underscore the importance of considering recent experience when interpreting cross-sectional anatomy-behavior correlations. Widespread changes throughout the white matter may be a hallmark of rapid plasticity associated with an intensive learning experience.

neuroscience↗

bioRxiv2017

Assessment of the need for separate test set and number of medical images necessary for deep learning: a sub-sampling study

Deep learning algorithms have tremendous potential utility in the classification of biomedical images. For example, images acquired with retinal optical coherence tomography (OCT) can be used to accurately classify patients with adult macular degeneration (AMD), and distinguish them from healthy control patients. However, previous research has suggested that large amounts of data are required in order to train deep learning algorithms, because of the large number of parameters that need to be fit. Here, we show that a moderate amount of data (data from approximately 1,800 patients) may be enough to reach close-to-maximal performance in the classification of AMD patients from OCT images. These results suggest that deep learning algorithms can be trained on moderate amounts of data, provided that images are relatively homogenous, and the effective number of parameters is sufficiently small. Furthermore, we demonstrate that in this application, cross-validation with a separate test set that is not used in any part of the training does not differ substantially from cross-validation with a validation data-set used to determine the optimal stopping point for training.

neuroscience↗

bioRxiv2017

AFQ-Browser: Supporting reproducible human neuroscience research through browser-based visualization tools

Human neuroscience research faces several challenges with regards to reproducibility. While scientists are generally aware that data sharing is an important component of reproducible research, it is not always clear how to usefully share data in a manner that allows other labs to understand and reproduce published findings. Here we describe a new open source tool, AFQ-Browser, that builds an interactive website as a companion to a published diffusion MRI study. Because AFQ-browser is portable -- it runs in any modern web-browser -- it can facilitate transparency and data sharing. Moreover, by leveraging new web-visualization technologies to create linked views between different dimensions of a diffusion MRI dataset (anatomy, quantitative diffusion metrics, subject metadata), AFQ-Browser facilitates exploratory data analysis, fueling new scientific discoveries based on previously published datasets. In an era where Big Data is playing an increasingly prominent role in scientific discovery, so will browser-based tools for exploring high-dimensional datasets, communicating scientific discoveries, sharing and aggregating data across labs, and publishing data alongside manuscripts.

neuroscience↗

bioRxiv2017

pulse2percept: A Python-based simulation framework for bionic vision

By 2020 roughly 200 million people worldwide will suffer from photoreceptor diseases such as retinitis pigmentosa and age-related macular degeneration, and a variety of retinal sight restoration technologies are being developed to target these diseases. One technology, analogous to cochlear implants, uses a grid of electrodes to stimulate remaining retinal cells. Two brands of retinal prostheses are currently approved for implantation in patients with late stage photoreceptor disease. Clinical experience with these implants has made it apparent that the vision restored by these devices differs substantially from normal sight. To better understand the outcomes of this technology, we developed pulse2percept, an open-source Python implementation of a computational model that predicts the perceptual experience of retinal prosthesis patients across a wide range of implant configurations. A modular and extensible user interface exposes the different building blocks of the software, making it easy for users to simulate novel implants, stimuli, and retinal models. We hope that this library will contribute substantially to the field of medicine by providing a tool to accelerate the development of visual prostheses.

neuroscience↗

bioRxiv2017

Deep-Learning Based, Automated Segmentation Of Macular Edema In Optical Coherence Tomography

Evaluation of clinical images is essential for diagnosis in many specialties and the development of computer vision algorithms to analyze biomedical images will be important. In ophthalmology, optical coherence tomography (OCT) is critical for managing retinal conditions. We developed a convolutional neural network (CNN) that detects intraretinal fluid (IRF) on OCT in a manner indistinguishable from clinicians. Using 1,289 OCT images, the CNN segmented images with a 0.911 cross-validated Dice coefficient, compared with segmentations by experts. Additionally, the agreement between experts and between experts and CNN were similar. Our results reveal that CNN can be trained to perform automated segmentations.

bioinformatics↗

bioRxiv2017

Learning to see again: Biological constraints on cortical plasticity and the implications for sight restoration technologies

1. ABSTRACTThe \"bionic eye\" - so long a dream of the future - is finally becoming a reality with retinal prostheses available to patients in both the US and Europe. However, clinical experience with these implants has made it apparent that the vision provided by these devices differs substantially from normal sight. Consequently, the ability to learn to make use of this abnormal retinal input plays a critical role in whether or not some functional vision is successfully regained. The goal of the present review is to summarize the vast basic science literature on developmental and adult cortical plasticity with an emphasis on how this literature might relate to the field of prosthetic vision. We begin with describing the distortion and information loss likely to be experienced by visual prosthesis users. We then define cortical plasticity and perceptual learning, and describe what is known, and what is unknown, about visual plasticity across the hierarchy of brain regions involved in visual processing, and across different stages of life. We close by discussing what is known about brain plasticity in sight restoration patients and discuss biological mechanisms that might eventually be harnessed to improve visual learning in these patients.

neuroscience↗

bioRxiv2017

Optimization of a free water elimination two-compartment model for diffusion tensor imaging.

Typical diffusion-weighted imaging (DWI) is susceptible to partial volume effects: different types of tissue that reside in the same voxel are inextricably mixed. For instance, in regions near the cerebral ventricles or parenchyma, fractional anisotropy (FA) from diffusion tensor imaging (DTI) may be underestimated, due to partial volumes of cerebral spinal fluid (CSF). Free-water can be suppressed by adding parameters to diffusion MRI models. For example, the DTI model can be extended to separately take into account the contributions of tissue and CSF, by representing the tissue compartment with an anisotropic diffusion tensor and the CSF compartment as an isotropic free water diffusion coefficient. Recently, two procedures were proposed to fit this two-compartment model to diffusion-weighted data acquired for at least two different non-zero diffusion MRI b-values. In this work, the first open-source reference implementation of these procedures is provided. In addition to presenting some methodological improvements that increase model fitting robustness, the free water DTI procedures are re-evaluated using Monte-Carlo multicompartmental simulations. Analogous to previous studies, our results show that the free water elimination DTI model is able to remove confounding effects of fast diffusion for typical FA values of brain white matter. In addition, this study confirms that for a fixed scanning time the fwDTI fitting procedures have better performance when data is acquired for diffusion gradient direction evenly distributed along two b-values of 500 and 1500 s/mm2.

neuroscience↗

bioRxiv2016

The interaction of orientation-specific surround suppression and visual-spatial attention

Orientation selective surround suppression (OSSS) is a reduction in the perceived contrast of a stimulus, which occurs when a collinear grating is placed adjacent to the stimulus. Attention affects performance on many visual tasks, and we asked whether the perceptual effects of OSSS are mitigated through the allocation of voluntary visual-spatial attention. Participants were tested in a contrast discrimination task: at the beginning of each trial, one location on the screen was cued and a subsequent contrast judgment was then more likely (70%) to be performed in that location. Replicating previous results, we found that the point of subjective equality (PSE) was elevated for a collinear, relative to an orthogonal, surround. While the PSE was similar for validly and invalidly cued trials, the just noticeable difference (JND) was larger for invalid cue trials, and for collinear, relative to orthogonal surround, suggesting that while OSSS affects both perceived contrast and sensitivity, voluntary attention affects only perceptual sensitivity. In another experiment no informative cue was provided, and attention was distributed over the entire display. In this case, JND and PSE were shifted depending on the contrast of the distractor, suggesting that OSSS is affected by the allocation of visual-spatial attention, but only under conditions of distributed attention.

neuroscience↗