Search

Behavioral pattern is the characteristic ways a person acts and has been recognized as a cause of many home accidents (h-accd). This study reviewed the types and prevalence of injuries among women in domestic works and proposes a model using Artificial Neural Network (ANN) function to forecast the safety level of women in domestic duty. The study was conducted in some parts of Western Nigeria among 340 subjects (171 married and 169 unmarried) using questionnaire. SPSS was used for data analysis. The ANN function was developed in MATLAB 2015a using the subjects' behavioral patterns and the model was used to predict safety in domestic duties (d-duties) among some women. 'Cuts/laceration' (40%) and 'skin contact with hot substance' (35.6%) were commonly reported. Carelessness (26.5%) and distraction (22.1%) were the main leading factors across the groups. Marital status and h-accd (Chi-square =4.323 and p= .038); 'hours spent on domestic works' and 'the h-accd' were both significant among other tested groups variables. With the developed ANN function, the results of the MSE was 0.33626 indicating that the function predicted the exact value. The result of the predicted h-accd (safety= -0.5445, hazards= 1.0228) in d-duties of the tested variables with the ANN function, showed a very low level of safety. The article concludes that the developed model is reliable and a recommended ergonomic tool useful in all homes, most especially where women perform most domestic works. Keywords: Women, domestic work, work, safety, behavior, hazards forecasting, western Nigeria, women's work., Introduction Domestic work (DomWrk) includes the physically, mentally and spiritually inclined work that one is daily engaged in as home activities [1] and is almost universally unremunerated. DomWrk may be [...]

Accurate prediction of recoverable train delay can support the train dispatchers' decision-making with timetable rescheduling and improving service reliability. In this paper, we present the results of an effort aimed to develop primary delay recovery (PDR) predictor model using train operation records from Wuhan-Guangzhou (W-G) high-speed railway. To this end, we first identified the main variables that contribute to delay, including dwell buffer time, running buffer time, magnitude of primary delay time, and individual sections' influence. Different models are applied and calibrated to predict the PDR. The validation results on test datasets indicate that the random forest regression (RFR) model outperforms the other three alternative models, namely, multiple linear regression (MLR), support vector machine (SVM), and artificial neural networks (ANN) regarding prediction accuracy measure. Specifically, the evaluation results show that when the prediction tolerance is less than 1 min, the RFR model can achieve up to 80.4% of prediction accuracy, while the accuracy level is 44.4%, 78.5%, and 78.5% for MLR, SVM, and ANN models, respectively. Key words: high-speed railway, primary delay recovery, buffer time, data-driven approaches, random forest model. La prediction precise de retard de train recuperable peut appuyer la prise de decisions des regulateurs de train en matiere de revision d'horaire et d'amelioration de la fiabilite du service. Dans le cadre de cette etude, nous presentons les resultats d'un effort ayant pour but de developper un modele predictif du retablissement de retard primaire (RRP) utilisant des donnees d'exploitation des trains de la liaison ferroviaire a grande vitesse du Wuhan-Guangzhou (W-G). A cette fin, nous avons d'abord identifie les principales variables qui contribuent aux retards, y compris la periode tampon inactive, la periode tampon de parcours, l'ampleur du retard primaire et l'incidence des sections individuelles. Differents modeles sont appliques et calibres en vue de prevoir le retablissement du retard primaire. Les resultats de validation faite sur des ensembles de donnees d'essai indiquent que le modele de regression par foret d'arbres decisionnels (RFAD) surpasse les trois autres modeles de rechange, a savoir la regression lineaire multiple (RLM), la machine a vecteurs de support (MVS) et les reseaux de neurones artificiels (RNA) quant a la mesure de l'exactitude de prediction. Specifiquement, les resultats de l'evaluation montrent que lorsque la tolerance de prediction est moins de 1 minute, le modele de regression par foret d'arbres decisionnels (RFAD) peut realiser une exactitude de prediction jusqu'a 80,4 %, tandis que le niveau d'exactitude est de 44,4, 78,5 et 78,5 % pour les modeles RLM, MVS et RNA, respectivement. [Traduit par la Redaction] Mots-cles : liaison ferroviaire a grande vitesse, retablissement du retard primaire, periode tampon, approches fondees sur les donnees, modele de foret d'arbres decisionnels., Introduction Despite the availability of advanced communication and control technologies, high-speed railway (HSR) lines could still be subject to significant delay in their daily operations due to various unexpected disruptions [...]

Technology that translates neural activity into speech would be transformative for people who are unable to communicate as a result of neurological impairments. Decoding speech from neural activity is challenging because speaking requires very precise and rapid multi-dimensional control of vocal tract articulators. Here we designed a neural decoder that explicitly leverages kinematic and sound representations encoded in human cortical activity to synthesize audible speech. Recurrent neural networks first decoded directly recorded cortical activity into representations of articulatory movement, and then transformed these representations into speech acoustics. In closed vocabulary tests, listeners could readily identify and transcribe speech synthesized from cortical activity. Intermediate articulatory dynamics enhanced performance even with limited data. Decoded articulatory representations were highly conserved across speakers, enabling a component of the decoder to be transferrable across participants. Furthermore, the decoder could synthesize speech when a participant silently mimed sentences. These findings advance the clinical viability of using speech neuroprosthetic technology to restore spoken communication. A neural decoder uses kinematic and sound representations encoded in human cortical activity to synthesize audible sentences, which are readily identified and transcribed by listeners., Author(s): Gopala K. Anumanchipalli [sup.1] [sup.2] , Josh Chartier [sup.1] [sup.2] [sup.3] , Edward F. Chang [sup.1] [sup.2] [sup.3] Author Affiliations: (1) Department of Neurological Surgery, University of California San [...]

Asphalt endurance limit is a strain value if experienced by asphalt pavement layer, no accumulated damage will occur and is directly related to asphalt healing. Therefore, if the pavement experiences this value of strain, or lower, no fatigue damage would accumulate within that pavement section. Beam fatigue test data conducted under the NCHRP Project 9-44A were extracted and utilized to create an artificial neural network predictive model (ANN) to determine the endurance limit strain values for conventional asphalt concrete pavements. The developed ANN model architecture as well as how to utilize it to predict the endurance limit were demonstrated and discussed in detail. Also, a stand- alone equation that is capable in the prediction of the endurance limit strain value, separate from the ANN model environment, was derived utilizing the eclectic extraction approach. The model training and validation data included 934 beam fatigue laboratory data points, as extracted from NCHRP Project 9-44A report. The developed model was able to determine the endurance limit strain value as a function of the stiffness ratio, number of cycles to failure, initial stiffness and rest period, and had a reasonable coefficient of determination ([R.sup.2]) value, which indicates the reliability of both the developed ANN model and the stand- alone equation. Furthermore, a correlation between the endurance limit strain values, as predicted utilizing the generated regression model under the NCHRP project 944-A, and the endurance limit strain values predicted utilizing the stand-alone ANN derived equation was found with a high [R.sup.2] value.Key words: Endurance limit, healing, fatigue, artificial neural network, rest period.La limite d'endurance de l'asphalte est une valeur de deformation, si soutenue par la couche d'asphalte, alors aucun dommage accumule ne se produira, et celle-ci est directement liee a l'autoreparation de l'asphalte. Par consequent, si la chaussee subit cette valeur de deformation, ou une valeur moindre, aucun dommage de fatigue ne s'accumulera dans ce tron^on de chaussee. Les donnees provenant des essais de fatigue de poutre realises dans le cadre du Projet NCHRP 9-44A ont ete extraites et utilisees en vue de creer un modele de prediction de reseau de neurones artificiels (RNA) afin de determiner les valeurs de deformation de la limite d'endurance de chaussees de beton d'asphalte traditionnelles. L'architecture du modele de RNA developpe et comment l'utiliser pour predire la limite d'endurance ont ete expliques et on en a discute en detail. Aussi, une equation autonome apte a predire la valeur de deformation de limite d'endurance, independamment de l'environnement du modele de RNA, a ete derivee au moyen d'une approche d'extraction eclectique. L'entrainement du modele et les donnees de validation comprenaient 934 points de donnees de laboratoire de fatigue de poutre, telles qu'extraites du rapport du Projet NCHRP 9-44A. Le modele developpe etait en mesure de determiner la valeur de deformation de limite d'endurance en fonction du rapport de rigidite, du nombre de cycles jusqu'a la defaillance, de la rigidite initiale et de la periode de repos et avait une valeur raisonnable du coefficient de determination ([R.sup.2]), ce qui indique la fiabilite tant du modele RNA developpe que de l'equation autonome. En outre, une correlation entre les valeurs de deformation de limite d'endurance, comme predites au moyen du modele de regression du projet NCHRP 944-A, et les valeurs de deformation de limite d'endurance predites utilisant l'equation autonome derivee du RNA s'est averee avoir une valeur elevee du [R.sup.2]. [Traduit par la Redaction]Mots-cles: limite d'endurance, autoreparation, fatigue, reseau de neurones artificiels, periode de repos., 1. IntroductionFatigue cracking is one of the major challenges in the flexible pavement design. Fatigue cracking is defined as the longitudinal or interconnect cracks that propagates from the bottom to [...]

Computerized electrocardiogram (ECG) interpretation plays a critical role in the clinical ECG workflow.sup.1. Widely available digital ECG data and the algorithmic paradigm of deep learning.sup.2 present an opportunity to substantially improve the accuracy and scalability of automated ECG analysis. However, a comprehensive evaluation of an end-to-end deep learning approach for ECG analysis across a wide variety of diagnostic classes has not been previously reported. Here, we develop a deep neural network (DNN) to classify 12 rhythm classes using 91,232 single-lead ECGs from 53,549 patients who used a single-lead ambulatory ECG monitoring device. When validated against an independent test dataset annotated by a consensus committee of board-certified practicing cardiologists, the DNN achieved an average area under the receiver operating characteristic curve (ROC) of 0.97. The average F.sub.1 score, which is the harmonic mean of the positive predictive value and sensitivity, for the DNN (0.837) exceeded that of average cardiologists (0.780). With specificity fixed at the average specificity achieved by cardiologists, the sensitivity of the DNN exceeded the average cardiologist sensitivity for all rhythm classes. These findings demonstrate that an end-to-end deep learning approach can classify a broad range of distinct arrhythmias from single-lead ECGs with high diagnostic performance similar to that of cardiologists. If confirmed in clinical settings, this approach could reduce the rate of misdiagnosed computerized ECG interpretations and improve the efficiency of expert human ECG interpretation by accurately triaging or prioritizing the most urgent conditions. Analysis of electrocardiograms using an end-to-end deep learning approach can detect and classify cardiac arrhythmia with high accuracy, similar to that of cardiologists., Author(s): Awni Y. Hannun [sup.1] , Pranav Rajpurkar [sup.1] , Masoumeh Haghpanahi [sup.2] , Geoffrey H. Tison [sup.3] , Codie Bourn [sup.2] , Mintu P. Turakhia [sup.4] [sup.5] , Andrew [...]

Autonomous vehicles could be fixtures on streets and highways as soon as 2025 as developers finalize the technologies and platforms that enable their performance. While issues relating to vehicle operation [...]

Neural-network training can be slow and energy intensive, owing to the need to transfer the weight data for the network between conventional digital memory chips and processor chips. Analogue non-volatile memory can accelerate the neural-network training algorithm known as backpropagation by performing parallelized multiply-accumulate operations in the analogue domain at the location of the weight data. However, the classification accuracies of such in situ training using non-volatile-memory hardware have generally been less than those of software-based training, owing to insufficient dynamic range and excessive weight-update asymmetry. Here we demonstrate mixed hardware-software neural-network implementations that involve up to 204,900 synapses and that combine long-term storage in phase-change memory, near-linear updates of volatile capacitors and weight-data transfer with 'polarity inversion' to cancel out inherent device-to-device variations. We achieve generalization accuracies (on previously unseen data) equivalent to those of software-based training on various commonly used machine-learning test datasets (MNIST, MNIST-backrand, CIFAR-10 and CIFAR-100). The computational energy efficiency of 28,065 billion operations per second per watt and throughput per area of 3.6 trillion operations per second per square millimetre that we calculate for our implementation exceed those of today's graphical processing units by two orders of magnitude. This work provides a path towards hardware accelerators that are both fast and energy efficient, particularly on fully connected neural-network layers.Analogue-memory-based neural-network training using non-volatile-memory hardware augmented by circuit simulations achieves the same accuracy as software-based training but with much improved energy efficiency and speed., Author(s): Stefano Ambrogio [sup.1] , Pritish Narayanan [sup.1] , Hsinyu Tsai [sup.1] , Robert M. Shelby [sup.1] , Irem Boybat [sup.2] [sup.3] , Carmelo di Nolfo [sup.1] [sup.3] , Severin [...]

The present numerical study is carried out for mixed convection in a nanofluid-filled lid-driven triangular cavity. The base wall of the cavity is in a caterpillar shape, which is assumed as a hot wall while the side and inclined walls are considered as cold walls. The finite volume method along with the SIMPLE algorithm is used to discretize the governing equations. The study is evaluated for constrained parameters, such as volume fraction of the nanoparticles, sliding direction of the side wall, Richardson number, and Grashof number. Fluid flow and heat transfer are presented in terms of streamlines and isotherms and rate of enhancement has been shown by local and average Nusselt number. It is observed from the study that the heat transfer rate is enhanced for each volume fraction of nanoparticles, for both directions of sliding wall, Richardson number, and Grashof number. The obtained numerical results are validated with the predicted results of artificial neural network (ANN). Good agreement is reported between the numerical results and the predicted results. Key words: mixed convection, triangular enclosure, nanofluid, Richardson number, ANN. L'etude numerique presentee ici vise la convexion mixte dans une cavite triangulaire entrainee remplie de nanofluide. Le mur de base de la cavite a la forme d'une chenille et suppose chaud, alors que les murs de cote et incline sont consideres froids. Nous utilisons la methode aux volumes finis avec l'algorithme SIMPLE pour discretiser les equations directrices. L'etude est faite a l'interieur de certaines contraintes touchant la fraction volumique des nanoparticules, la direction de glissement du mur de cote, le nombre de Richardson et le nombre de Grashof. L'ecoulement du fluide et le transfert de chaleur sont presentes en termes de lignes de courant et d'isothermes, alors que le taux d'augmentation est indique par le nombre de Nusselt moyen et local. Nous observons a partir de cette etude que le transfert de chaleur augmente avec chaque fraction volumique de nanoparticules, les deux directions du mur mobile, le nombre de Richardson et le nombre de Grashof. Les resultats numeriques obtenus sont valides par comparaison avec les resultats predits par un reseau neuronal artificiel (RNA/ANN). Un bon accord est observe entre les deux. [Traduit par la Redaction] Mots-cles: convexion mixte, enceinte triangulaire, nanofluide, nombre de Richardson, reseau neuronal artificiel (RNA)., 1. Introduction The free convection heat transfer in an enclosure has contributed a lot in industrial applications [1-4]. The day-to-day miniaturization of electronic components in the present scenario requires a [...]

Short-term travel time prediction is an essential input to intelligent transportation systems. Timely and accurate traffic forecasting is necessary for advanced traffic management systems and advanced traveler information systems. Despite several short-term travel time prediction approaches have been proposed in the past decade, especially for hybrid models that consist of machine learning models and statistical models, few studies focus on the over-fitting problem brought by hybrid models. The over-fitting problem deteriorates the prediction accuracy especially during peak hours. This paper proposes a hybrid model that embraces wavelet neural network (WNN), Markov chain (MAR), and the volatility (VOA) model for short-term travel time prediction in a freeway system. The purpose of this paper is to provide deeper insights into underlining dynamic traffic patterns and to improve the prediction accuracy and robustness. The method takes periodical analysis, error correction, and noise extraction into consideration and improve the forecasting performance in peak hours. The proposed methodology predicts travel time by decomposing travel time data into three components: a periodic trend presented by a modified WNN, a residual part modeled by Markov chain, and the volatility part estimated by the modified generalized autoregressive conditional heteroscedasticity model. Forecasting performance is investigated with freeway travel time data from Houston, Texas and examined by three measures: mean absolute error, mean absolute percentage error, and root mean square error. The results show that the travel times predicted by the WNN-MAR-VOA method are robust and accurate. Meanwhile, the proposed method is able to capture the underlying periodic characteristics and volatility nature of travel time data. Key words: travel time, short-term prediction, wavelet neural network, multi- regime based model, Markov chain, volatility analysis. La prediction a court terme du temps de deplacement constitue un apport essentiel aux systemes de transport intelligent. La prevision opportune et precise du trafic est necessaire au niveau des systemes avances de gestion de la circulation et des systemes avances d'information des voyageurs. Bien qu'on ait propose plusieurs approches de prediction a court terme du temps de deplacement au cours de la derniere decennie, particulierement pour les modeles hybrides qui comportent des modeles d'apprentissage automatique et des modeles statistiques, peu d'etudes se concentrent sur le probleme de surapprentissage cause par les modeles hybrides. Le probleme de surapprentissage diminue l'exactitude de prediction particulierement durant les heures de pointe. Dans le cadre de cette etude, on propose un modele hybride qui comprend un reseau neuronal en ondelettes (WNN), la chaine de Markov (MAR) et le modele de volatilite (VOA) aux fins de prediction a court terme du temps de deplacement des usagers d'un systeme d'autoroutes. Le but de cette etude vise a apporter des connaissances plus approfondies par rapport aux modeles de trafic dynamiques sous-jacents et a ameliorer l'exactitude et la robustesse de prediction. Dans cette methode, on tient compte de l'analyse periodique, de la correction d'erreurs et de l'extraction sonore et on ameliore les predictions aux heures de pointe. La methodologie proposee prevoit le temps de deplacement en decomposant les donnees sur le temps de deplacement en trois volets: l'analyse periodique presentee par le reseau neuronal en ondelettes modifie, une partie residuelle modelisee par chaine de Markov et la partie volatilite evaluee par le modele autoregressif conditionnellement heteroscedastique generalise modifie. On examine l'efficacite de prediction en utilisant les donnees du temps de deplacement sur les autoroutes de Houston au Texas et en l'analysant au moyen de trois mesures: l'erreur moyenne absolue, le pourcentage d'erreur absolue moyenne et l'ecart-type. Les resultats montrent que les temps de deplacement prevus par la methode WNNMAR-VOA sont robustes et precis. Entre-temps, la methode proposee est en mesure de refleter les caracteristiques periodiques sous-jacentes et la nature de volatilite des donnees sur le temps de deplacement. [Traduit par la Redaction] Mots-cles: temps de deplacement, prediction a court terme, reseau neuronal en ondelettes, modele a multiples regimes, chaine de Markov, analyse de volatilite., Introduction Short-term travel time prediction is a fundamental input for intelligent transportation systems (ITS). Traffic congestion is a thorny problem in many cities, especially in those megacities because of the [...]

Advocates of fundamental analysis depreciate technical analysis as a superficial study of trends and patterns depicted by charts without any conclusive proof of efficacy. However, technical trading is one of the ancient trading techniques and the advancements in technical trading are growing exponentially in the age of superfast computers. Predicting the movements of stock prices precisely using sophisticated techniques needs continuous improvement to capture trends. Technical trading techniques using fuzzy models are gaining prominence in predicting non-linear trends in stock markets because of the capability of extracting meaningful information from a large set of data. Artificial neural network (ANN) integrated models are serving the needs of learning non-linear patterns and helping in making better predictions. This research paper focuses on designing models using the architecture of ANN techniques, specifically Error Back Propagation Network (EBPN) and Radial Basis Function Network (RBFN), from Multi Input Multi Output (MIMO) and Multi Input Single Output (MISO) perspectives. The tests of the models developed in this study were performed using the key variable of open, close, high and low prices of DOW30 and NASDAQ100. We used two measures of predictability: Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). Based on the results, we observed that EBPN outperformed RBFN in predicting the future prices. The results of MIMO approach were also precise than MISO for both systems. Keywords: Artificial Neural Network (ANN), Error Back Propagation Network (EBPN), Radial Basis Function (RBFN), Multi Input Single Output (MISO), Multi Input Multi Output (MIMO)., INTRODUCTION Volatility has several aspects for trading including predicting the stock market direction for investing. The prediction of the market gives an idea of the direction of the economy. Therefore, [...]

A probabilistic seismic demand model that relates ground motion intensity measures (IMs) to the structural demand measures is a useful tool for reliability analysis of structures. It is common to utilize the scalar seismic parameters or a vector of a few seismic parameters to reveal ground motion uncertainty. However, for the qualification of an IM for representing the ground motion uncertainty, a larger vector of greater seismic component is required. This study aims to use more parameters as vector IMs in the demand model to achieve better estimation of the ground motion uncertainty. In this study, three-layer feed forward neural network was used to predict the seismic demand model of the mid-rise reinforced concrete buildings for pulse-like ground motions. The results indicate that due to the complexity of the relationship between seismic response of structures and seismic intensity parameters, using artificial neural networks method is more suitable than numerical methods to show uncertainties. Key words: seismic demand model, near-fault earthquakes, artificial neural network. Un modele de demande sismique probabiliste qui etablit un rapport entre les mesures d'intensite (MI) de mouvement du sol et les mesures de demande structurale constitue un outil utile aux fins de l'analyse de fiabilite des structures. Il est courant d'utiliser des parametres sismiques scalaires ou un vecteur de quelques parametres sismiques pour representer l'incertitude de mouvement du sol. Cependant, aux fins de l'homologation d'une MI pour representer l'incertitude de mouvement du sol, un plus grand vecteur de composant sismique important est requis. Cette etude a pour but d'utiliser plus de parametres comme MI vectorielle dans le modele de demande afin de realiser une meilleure evaluation de l'incertitude de mouvement du sol. Dans le cadre de cette etude, on a utilise un reseau neuronal non recurrent a trois couches afin de prevoir le modele de demande sismique de batiments de moyenne hauteur en beton arme en ce qui concerne les mouvements du sol par impulsions. Les resultats indiquent qu'en raison de la complexite de la relation entre la reponse sismique de structures et les parametres d'intensite sismiques, l'emploi de la methode de reseaux neuronaux est plus approprie que l'utilisation de methodes numeriques pour exposer les incertitudes. [Traduit par la Redaction] Mots-cles: modele de demande sismique, tremblements de terre a proximite de faille, reseau de neurones artificiels., Introduction Assessment of the structural seismic damage under earthquake ground motions is an important issue in the seismic risk analysis in the context of seismic performance levels (Kostinakis and Athanatopoulou [...]

The oscillation of asphalt mix composition on a daily basis significantly affects the achieved properties of the asphalt during production, thus resulting in conducting expensive laboratory tests to determine existing properties and predicting the future results. To decrease the amount of such tests, a development of artificial neural network and multiple linear regression models in the prediction process of predetermined dependent variables air void and soluble binder content is presented. The input data were obtained from a single laboratory and consists of testing 386 mixes of hot mix asphalt (HMA). It was found that it is possible and desirable to apply such models in the prediction process of the HMA properties. The final aim of the research was to compare results of the prediction models on an independent dataset and analyze them through the boundary conditions of technical regulations and the standard EN13108-21. Key words: hot mix asphalt, artificial neural networks, multiple linear regression, prediction process. La variation quotidienne dans la composition d'asphalte melange a un effet significatif sur les proprietes de l'asphalte pendant la production, entrainant ainsi la necessite de realiser des essais en laboratoire couteux afin de determiner les proprietes existantes et de prevoir les resultats a venir. Dans le but de diminuer la quantite de tels essais, on presente l'elaboration d'un reseau neuronal artificiel et des modeles de regression lineaire multiple mis en reuvre dans le processus de prediction de variables dependantes predeterminees, la teneur en vides interstitiels et en liant soluble. Les donnees d'entree ont ete obtenues d'un seul laboratoire et comptent 386 melanges d'asphalte melange a chaud. Il s'est avere qu'il est possible et desirable de mettre en reuvre de tels modeles dans le processus de prediction des proprietes d'asphalte melange a chaud. Le but de la recherche etait de comparer les resultats des modeles de prediction utilises sur un ensemble de donnees independant et les analyser sous des conditions limites de reglements techniques et de la norme EN 13108-21. [Traduit par la Redaction] Mots-cles: asphalte melange a chaud, reseaux de neurones artificiels, regression lineaire multiple, processus de prediction., Introduction Bituminous mixtures contain in its composition mineral mixture, filler, binder and various additives that are mixed at a temperature whose value depends on the properties of the used binder. [...]

Aftershocks are a response to changes in stress generated by large earthquakes and represent the most common observations of the triggering of earthquakes. The maximum magnitude of aftershocks and their temporal decay are well described by empirical laws (such as Bath's law.sup.1 and Omori's law.sup.2), but explaining and forecasting the spatial distribution of aftershocks is more difficult. Coulomb failure stress change.sup.3 is perhaps the most widely used criterion to explain the spatial distributions of aftershocks.sup.4-8, but its applicability has been disputed.sup.9-11. Here we use a deep-learning approach to identify a static-stress-based criterion that forecasts aftershock locations without prior assumptions about fault orientation. We show that a neural network trained on more than 131,000 mainshock-aftershock pairs can predict the locations of aftershocks in an independent test dataset of more than 30,000 mainshock-aftershock pairs more accurately (area under curve of 0.849) than can classic Coulomb failure stress change (area under curve of 0.583). We find that the learned aftershock pattern is physically interpretable: the maximum change in shear stress, the von Mises yield criterion (a scaled version of the second invariant of the deviatoric stress-change tensor) and the sum of the absolute values of the independent components of the stress-change tensor each explain more than 98 per cent of the variance in the neural-network prediction. This machine-learning-driven insight provides improved forecasts of aftershock locations and identifies physical quantities that may control earthquake triggering during the most active part of the seismic cycle.Neural networks trained on data from about 130,000 aftershocks from around 100 large earthquakes improve predictions of the spatial distribution of aftershocks and suggest physical quantities that may control earthquake triggering., Author(s): Phoebe M. R. DeVries [sup.1] [sup.2] , Fernanda Viégas [sup.3] , Martin Wattenberg [sup.3] , Brendan J. Meade [sup.1] Author Affiliations:(1) Department of Earth and Planetary Sciences, Harvard University, [...]

Software implementations of brain-inspired computing underlie many important computational tasks, from image processing to speech recognition, artificial intelligence and deep learning applications. Yet, unlike real neural tissue, traditional computing architectures physically separate the core computing functions of memory and processing, making fast, efficient and low-energy computing difficult to achieve. To overcome such limitations, an attractive alternative is to design hardware that mimics neurons and synapses. Such hardware, when connected in networks or neuromorphic systems, processes information in a way more analogous to brains. Here we present an all-optical version of such a neurosynaptic system, capable of supervised and unsupervised learning. We exploit wavelength division multiplexing techniques to implement a scalable circuit architecture for photonic neural networks, successfully demonstrating pattern recognition directly in the optical domain. Such photonic neurosynaptic networks promise access to the high speed and high bandwidth inherent to optical systems, thus enabling the direct processing of optical telecommunication and visual data. An optical version of a brain-inspired neurosynaptic system, using wavelength division multiplexing techniques, is presented that is capable of supervised and unsupervised learning., Author(s): J. Feldmann [sup.1] , N. Youngblood [sup.2] , C. D. Wright [sup.3] , H. Bhaskaran [sup.2] , W. H. P. Pernice [sup.1] Author Affiliations: (1) Institute of Physics, University [...]

Background For virtually every patient with colorectal cancer (CRC), hematoxylin-eosin (HE)-stained tissue slides are available. These images contain quantitative information, which is not routinely used to objectively extract prognostic biomarkers. In the present study, we investigated whether deep convolutional neural networks (CNNs) can extract prognosticators directly from these widely available images. Methods and findings We hand-delineated single-tissue regions in 86 CRC tissue slides, yielding more than 100,000 HE image patches, and used these to train a CNN by transfer learning, reaching a nine-class accuracy of >94% in an independent data set of 7,180 images from 25 CRC patients. With this tool, we performed automated tissue decomposition of representative multitissue HE images from 862 HE slides in 500 stage I-IV CRC patients in the The Cancer Genome Atlas (TCGA) cohort, a large international multicenter collection of CRC tissue. Based on the output neuron activations in the CNN, we calculated a 'deep stroma score,' which was an independent prognostic factor for overall survival (OS) in a multivariable Cox proportional hazard model (hazard ratio [HR] with 95% confidence interval [CI]: 1.99 [1.27-3.12], p = 0.0028), while in the same cohort, manual quantification of stromal areas and a gene expression signature of cancer-associated fibroblasts (CAFs) were only prognostic in specific tumor stages. We validated these findings in an independent cohort of 409 stage I-IV CRC patients from the 'Darmkrebs: Chancen der Verhütung durch Screening' (DACHS) study who were recruited between 2003 and 2007 in multiple institutions in Germany. Again, the score was an independent prognostic factor for OS (HR 1.63 [1.14-2.33], p = 0.008), CRC-specific OS (HR 2.29 [1.5-3.48], p = 0.0004), and relapse-free survival (RFS; HR 1.92 [1.34-2.76], p = 0.0004). A prospective validation is required before this biomarker can be implemented in clinical workflows. Conclusions In our retrospective study, we show that a CNN can assess the human tumor microenvironment and predict prognosis directly from histopathological images., Author(s): Jakob Nikolas Kather 1,2,3,4,*, Johannes Krisam 5, Pornpimol Charoentong 1,3, Tom Luedde 4, Esther Herpel 6,7, Cleo-Aron Weis 8, Timo Gaiser 8, Alexander Marx 8, Nektarios A. Valous 1,3, [...]

Background Magnetic resonance imaging (MRI) of the knee is the preferred method for diagnosing knee injuries. However, interpretation of knee MRI is time-intensive and subject to diagnostic error and variability. An automated system for interpreting knee MRI could prioritize high-risk patients and assist clinicians in making diagnoses. Deep learning methods, in being able to automatically learn layers of features, are well suited for modeling the complex relationships between medical images and their interpretations. In this study we developed a deep learning model for detecting general abnormalities and specific diagnoses (anterior cruciate ligament [ACL] tears and meniscal tears) on knee MRI exams. We then measured the effect of providing the model's predictions to clinical experts during interpretation. Methods and findings Our dataset consisted of 1,370 knee MRI exams performed at Stanford University Medical Center between January 1, 2001, and December 31, 2012 (mean age 38.0 years; 569 [41.5%] female patients). The majority vote of 3 musculoskeletal radiologists established reference standard labels on an internal validation set of 120 exams. We developed MRNet, a convolutional neural network for classifying MRI series and combined predictions from 3 series per exam using logistic regression. In detecting abnormalities, ACL tears, and meniscal tears, this model achieved area under the receiver operating characteristic curve (AUC) values of 0.937 (95% CI 0.895, 0.980), 0.965 (95% CI 0.938, 0.993), and 0.847 (95% CI 0.780, 0.914), respectively, on the internal validation set. We also obtained a public dataset of 917 exams with sagittal T1-weighted series and labels for ACL injury from Clinical Hospital Centre Rijeka, Croatia. On the external validation set of 183 exams, the MRNet trained on Stanford sagittal T2-weighted series achieved an AUC of 0.824 (95% CI 0.757, 0.892) in the detection of ACL injuries with no additional training, while an MRNet trained on the rest of the external data achieved an AUC of 0.911 (95% CI 0.864, 0.958). We additionally measured the specificity, sensitivity, and accuracy of 9 clinical experts (7 board-certified general radiologists and 2 orthopedic surgeons) on the internal validation set both with and without model assistance. Using a 2-sided Pearson's chi-squared test with adjustment for multiple comparisons, we found no significant differences between the performance of the model and that of unassisted general radiologists in detecting abnormalities. General radiologists achieved significantly higher sensitivity in detecting ACL tears (p-value = 0.002; q-value = 0.019) and significantly higher specificity in detecting meniscal tears (p-value = 0.003; q-value = 0.019). Using a 1-tailed t test on the change in performance metrics, we found that providing model predictions significantly increased clinical experts' specificity in identifying ACL tears (p-value < 0.001; q-value = 0.006). The primary limitations of our study include lack of surgical ground truth and the small size of the panel of clinical experts. Conclusions Our deep learning model can rapidly generate accurate clinical pathology classifications of knee MRI exams from both internal and external datasets. Moreover, our results support the assertion that deep learning models can improve the performance of clinical experts during medical imaging interpretation. Further research is needed to validate the model prospectively and to determine its utility in the clinical setting., Author(s): Nicholas Bien 1,*, Pranav Rajpurkar 1, Robyn L. Ball 2, Jeremy Irvin 1, Allison Park 1, Erik Jones 1, Michael Bereket 1, Bhavik N. Patel 3, Kristen W. Yeom [...]

Background Pneumothorax can precipitate a life-threatening emergency due to lung collapse and respiratory or circulatory distress. Pneumothorax is typically detected on chest X-ray; however, treatment is reliant on timely review of radiographs. Since current imaging volumes may result in long worklists of radiographs awaiting review, an automated method of prioritizing X-rays with pneumothorax may reduce time to treatment. Our objective was to create a large human-annotated dataset of chest X-rays containing pneumothorax and to train deep convolutional networks to screen for potentially emergent moderate or large pneumothorax at the time of image acquisition. Methods and findings In all, 13,292 frontal chest X-rays (3,107 with pneumothorax) were visually annotated by radiologists. This dataset was used to train and evaluate multiple network architectures. Images showing large- or moderate-sized pneumothorax were considered positive, and those with trace or no pneumothorax were considered negative. Images showing small pneumothorax were excluded from training. Using an internal validation set (n = 1,993), we selected the 2 top-performing models; these models were then evaluated on a held-out internal test set based on area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and positive predictive value (PPV). The final internal test was performed initially on a subset with small pneumothorax excluded (as in training; n = 1,701), then on the full test set (n = 1,990), with small pneumothorax included as positive. External evaluation was performed using the National Institutes of Health (NIH) ChestX-ray14 set, a public dataset labeled for chest pathology based on text reports. All images labeled with pneumothorax were considered positive, because the NIH set does not classify pneumothorax by size. In internal testing, our 'high sensitivity model' produced a sensitivity of 0.84 (95% CI 0.78-0.90), specificity of 0.90 (95% CI 0.89-0.92), and AUC of 0.94 for the test subset with small pneumothorax excluded. Our 'high specificity model' showed sensitivity of 0.80 (95% CI 0.72-0.86), specificity of 0.97 (95% CI 0.96-0.98), and AUC of 0.96 for this set. PPVs were 0.45 (95% CI 0.39-0.51) and 0.71 (95% CI 0.63-0.77), respectively. Internal testing on the full set showed expected decreased performance (sensitivity 0.55, specificity 0.90, and AUC 0.82 for high sensitivity model and sensitivity 0.45, specificity 0.97, and AUC 0.86 for high specificity model). External testing using the NIH dataset showed some further performance decline (sensitivity 0.28-0.49, specificity 0.85-0.97, and AUC 0.75 for both). Due to labeling differences between internal and external datasets, these findings represent a preliminary step towards external validation. Conclusions We trained automated classifiers to detect moderate and large pneumothorax in frontal chest X-rays at high levels of performance on held-out test data. These models may provide a high specificity screening solution to detect moderate or large pneumothorax on images collected when human review might be delayed, such as overnight. They are not intended for unsupervised diagnosis of all pneumothoraces, as many small pneumothoraces (and some larger ones) are not detected by the algorithm. Implementation studies are warranted to develop appropriate, effective clinician alerts for the potentially critical finding of pneumothorax, and to assess their impact on reducing time to treatment., Author(s): Andrew G. Taylor 1,2,*, Clinton Mielke 2, John Mongan 1,2 Introduction Pneumothorax can constitute a medical emergency since the presence of air within the pleural space outside the lung [...]

Background There is interest in using convolutional neural networks (CNNs) to analyze medical imaging to provide computer-aided diagnosis (CAD). Recent work has suggested that image classification CNNs may not generalize to new data as well as previously believed. We assessed how well CNNs generalized across three hospital systems for a simulated pneumonia screening task. Methods and findings A cross-sectional design with multiple model training cohorts was used to evaluate model generalizability to external sites using split-sample validation. A total of 158,323 chest radiographs were drawn from three institutions: National Institutes of Health Clinical Center (NIH; 112,120 from 30,805 patients), Mount Sinai Hospital (MSH; 42,396 from 12,904 patients), and Indiana University Network for Patient Care (IU; 3,807 from 3,683 patients). These patient populations had an age mean (SD) of 46.9 years (16.6), 63.2 years (16.5), and 49.6 years (17) with a female percentage of 43.5%, 44.8%, and 57.3%, respectively. We assessed individual models using the area under the receiver operating characteristic curve (AUC) for radiographic findings consistent with pneumonia and compared performance on different test sets with DeLong's test. The prevalence of pneumonia was high enough at MSH (34.2%) relative to NIH and IU (1.2% and 1.0%) that merely sorting by hospital system achieved an AUC of 0.861 (95% CI 0.855-0.866) on the joint MSH-NIH dataset. Models trained on data from either NIH or MSH had equivalent performance on IU (P values 0.580 and 0.273, respectively) and inferior performance on data from each other relative to an internal test set (i.e., new data from within the hospital system used for training data; P values both Conclusion Pneumonia-screening CNNs achieved better internal than external performance in 3 out of 5 natural comparisons. When models were trained on pooled data from sites with different pneumonia prevalence, they performed better on new pooled data from these sites but not on external data. CNNs robustly identified hospital system and department within a hospital, which can have large differences in disease burden and may confound predictions., Author(s): John R. Zech 1, Marcus A. Badgeley 2, Manway Liu 2, Anthony B. Costa 3, Joseph J. Titano 4, Eric Karl Oermann 3,* Introduction There is significant interest in [...]