Search

Temporal hierarchies have been widely used during the past few years as they are capable to provide more accurate coherent forecasts at different planning horizons. However, they still display some limitations, being mainly subject to the forecasting methods used for generating the base forecasts and the particularities of the examined series. This paper deals with such limitations by considering three different strategies: (i) combining forecasts of multiple methods, (ii) applying bias adjustments and (iii) selectively implementing temporal hierarchies to avoid seasonal shrinkage. The proposed strategies can be applied either separately or simultaneously, being complements to the method considered for reconciling the base forecasts and completely independent from each other. Their effect is evaluated using the monthly series of the M and M3 competitions. The results are very promising, displaying lots of potential for improving the performance of temporal hierarchies, both in terms of accuracy and bias. [ABSTRACT FROM AUTHOR]

Most cetacean species are wide-ranging and highly mobile, creating significant challenges for researchers by limiting the scope of data that can be collected and leaving large areas un-surveyed. Aerial surveys have proven an effective way to locate and study cetacean movements but are costly and limited in spatial extent. Here we present a semi-automated pipeline for whale detection from very high-resolution (sub-meter) satellite imagery that makes use of a convolutional neural network (CNN). We trained ResNet, and DenseNet CNNs using down-scaled aerial imagery and tested each model on 31 cm-resolution imagery obtained from the WorldView-3 sensor. Satellite imagery was tiled and the trained algorithms were used to classify whether or not a tile was likely to contain a whale. Our best model correctly classified 100% of tiles with whales, and 94% of tiles containing only water. All model architectures performed well, with learning rate controlling performance more than architecture. While the resolution of commercially-available satellite imagery continues to make whale identification a challenging problem, our approach provides the means to efficiently eliminate areas without whales and, in doing so, greatly accelerates ocean surveys for large cetaceans. [ABSTRACT FROM AUTHOR]

Background: The Sequential Organ Failure Assessment (SOFA) score is commonly used in ICUs around the world, designed to assess the severity of the patient's clinical state based on function/dysfunction of six major organ systems. The goal of this work is to build a computational model to predict mortality based on a series of SOFA scores. In addition, we examined the possibility of improving the prediction by incorporating a new component designed to measure the performance of the gastrointestinal system, added to the other six components. Methods: In this retrospective study, we used patients’ three latest SOFA scores recorded during an individual ICU stay as input to different machine learning models and ensemble learning models. We added three validated parameters representing gastrointestinal failure. Among others, we used classification models such as Support Vector Machines (SVMs), Neural Networks, Logistic Regression and a penalty function used to increase model robustness in regard to certain extreme cases, which may be found in ICU population. We used the Area under Curve (AUC) performance metric to examine performance. Results: We found an ensemble model of linear and logistic regression achieves a higher AUC compared related works in past years. After incorporating the gastrointestinal failure score along with the penalty function, our best performing ensemble model resulted in an additional improvement in terms of AUC metrics. We implemented and compared 36 different models that were built using both the information from the SOFA score as well as that of the gastrointestinal system. All compared models have approximately similar and relatively large AUC (between 0.8645 and 0.9146) with the best results are achieved by incorporating the gastrointestinal parameters into the prediction models. Conclusions: Our findings indicate that gastrointestinal parameters carry significant information as a mortality predictor in addition to the conventional SOFA score. This information improves the predictive power of machine learning models by extending the SOFA to include information related to gastrointestinal organ system. The described method improves mortality prediction by considering the dynamics of the extended SOFA score. Although tested on a limited data set, the results' stability across different models suggests robustness in real-time use. [ABSTRACT FROM AUTHOR]

Objective: The Centers for Disease Control and Prevention (CDC) coordinates a labor-intensive process to measure the prevalence of autism spectrum disorder (ASD) among children in the United States. Random forests methods have shown promise in speeding up this process, but they lag behind human classification accuracy by about 5%. We explore whether more recently available document classification algorithms can close this gap. Materials and methods: Using data gathered from a single surveillance site, we applied 8 supervised learning algorithms to predict whether children meet the case definition for ASD based solely on the words in their evaluations. We compared the algorithms’ performance across 10 random train-test splits of the data, using classification accuracy, F1 score, and number of positive calls to evaluate their potential use for surveillance. Results: Across the 10 train-test cycles, the random forest and support vector machine with Naive Bayes features (NB-SVM) each achieved slightly more than 87% mean accuracy. The NB-SVM produced significantly more false negatives than false positives (P = 0.027), but the random forest did not, making its prevalence estimates very close to the true prevalence in the data. The best-performing neural network performed similarly to the random forest on both measures. Discussion: The random forest performed as well as more recently available models like the NB-SVM and the neural network, and it also produced good prevalence estimates. NB-SVM may not be a good candidate for use in a fully-automated surveillance workflow due to increased false negatives. More sophisticated algorithms, like hierarchical convolutional neural networks, may not be feasible to train due to characteristics of the data. Current algorithms might perform better if the data are abstracted and processed differently and if they take into account information about the children in addition to their evaluations. Conclusion: Deep learning models performed similarly to traditional machine learning methods at predicting the clinician-assigned case status for CDC’s autism surveillance system. While deep learning methods had limited benefit in this task, they may have applications in other surveillance systems. [ABSTRACT FROM AUTHOR]

Neural correlates of mind wandering: The ability to detect mind wandering as it occurs is an important step towards improving our understanding of this phenomenon and studying its effects on learning and performance. Current detection methods typically rely on observable behaviour in laboratory settings, which do not capture the underlying neural processes and may not translate well into real-world settings. We address both of these issues by recording electroencephalography (EEG) simultaneously from 15 participants during live lectures on research in orthopedic surgery. We performed traditional group-level analysis and found neural correlates of mind wandering during live lectures that are similar to those found in some laboratory studies, including a decrease in occipitoparietal alpha power and frontal, temporal, and occipital beta power. However, individual-level analysis of these same data revealed that patterns of brain activity associated with mind wandering were more broadly distributed and highly individualized than revealed in the group-level analysis. Mind wandering detection: To apply these findings to mind wandering detection, we used a data-driven method known as common spatial patterns to discover scalp topologies for each individual that reflects their differences in brain activity when mind wandering versus attending to lectures. This approach avoids reliance on known neural correlates primarily established through group-level statistics. Using this method for individual-level machine learning of mind wandering from EEG, we were able to achieve an average detection accuracy of 80–83%. Conclusions: Modelling mind wandering at the individual level may reveal important details about its neural correlates that are not reflected when using traditional observational and statistical methods. Using machine learning techniques for this purpose can provide new insight into the varieties of neural activity involved in mind wandering, while also enabling real-time detection of mind wandering in naturalistic settings. [ABSTRACT FROM AUTHOR]

Filter or screening methods are often used as a preprocessing step for reducing the number of variables used by a learning algorithm in obtaining a classification or regression model. While there are many such filter methods, there is a need for an objective evaluation of these methods. Such an evaluation is needed to compare them with each other and also to answer whether they are at all useful, or a learning algorithm could do a better job without them. For this purpose, many popular screening methods are partnered in this paper with three regression learners and five classification learners and evaluated on ten real datasets to obtain accuracy criteria such as R-square and area under the ROC curve (AUC). The obtained results are compared through curve plots and comparison tables in order to find out whether screening methods help improve the performance of learning algorithms and how they fare with each other. Our findings revealed that the screening methods were useful in improving the prediction of the best learner on two regression and two classification datasets out of the ten datasets evaluated. [ABSTRACT FROM AUTHOR]

Accurately predicting and testing the types of Pulmonary arterial hypertension (PAH) of each patient using cost-effective microarray-based expression data and machine learning algorithms could greatly help either identifying the most targeting medicine or adopting other therapeutic measures that could correct/restore defective genetic signaling at the early stage. Furthermore, the prediction model construction processes can also help identifying highly informative genes controlling PAH, leading to enhanced understanding of the disease etiology and molecular pathways. In this study, we used several different gene filtering methods based on microarray expression data obtained from a high-quality patient PAH dataset. Following that, we proposed a novel feature selection and refinement algorithm in conjunction with well-known machine learning methods to identify a small set of highly informative genes. Results indicated that clusters of small-expression genes could be extremely informative at predicting and differentiating different forms of PAH. Additionally, our proposed novel feature refinement algorithm could lead to significant enhancement in model performance. To summarize, integrated with state-of-the-art machine learning and novel feature refining algorithms, the most accurate models could provide near-perfect classification accuracies using very few (close to ten) low-expression genes. [ABSTRACT FROM AUTHOR]

While there have been many studies using machine learning (ML) algorithms to predict process outcomes and device performance in semiconductor manufacturing, the extensively developed technology computer-aided design (TCAD) physical models should play a more significant role in conjunction with ML. While TCAD models have been effective in predicting the trends of experiments, a machine learning statistical model is more capable of predicting the anomalous effects that can be dependent on the chambers, machines, fabrication environment, and specific layouts. In this paper, we use an analytics-statistics mixed training (ASMT) approach using TCAD. Under this method, the TCAD models are incorporated into the machine learning training procedure. The mixed dataset with the experimental and TCAD results improved the prediction in terms of accuracy. With the application of ASMT to the BOSCH process, we show that the mean square error (MSE) can be effectively decreased when the analytics-statistics mixed training (ASMT) scheme is used instead of the classic neural network (NN) used in the baseline study. In this method, statistical induction and analytical deduction can be combined to increase the prediction accuracy of future intelligent semiconductor manufacturing. [ABSTRACT FROM AUTHOR]

The article offers information on the challenging scientific problems faced by humanity. Topics discussed include information on the efforts to empower Africans to join the global neuroscience community; discussions on the BRAIN initiative in the U.S. and the Human Brain Project in Europe; and the information on the limiting factor for computational neuroscience development in Africa.

A key goal of precision medicine is predicting the best drug therapy for a specific patient from genomic information. In oncology, cancers that appear similar pathologically can vary greatly in how they respond to the same drug. Fortunately, data from high-throughput screening programs often reveal important relationships between genomic variability of cancer cells and their response to drugs. Nevertheless, many current computational methods to predict compound activity against cancer cells require large quantities of genomic, epigenomic, and additional cellular data to develop and to apply. Here we integrate recent screening data and machine learning to train classification models that predict the activity/inactivity of compounds against cancer cells based on the mutational status of only 145 oncogenes and a set of compound structural descriptors. Using IC50 values of 1 μM as activity cutoffs, our predictive models have sensitivities of 87%, specificities of 87%, and yield an area under the receiver operating characteristic curve equal to 0.94. We also develop regression models to predict log(IC50) values of compounds for cancer cells; the models achieve a Pearson correlation coefficient of 0.86 for cross-validation and up to 0.65–0.73 against blind test sets. Predictive performance remains strong when as few as 50 oncogenes are included. Finally, even when 40% of experimental IC50 values are missing from screening data, they can be imputed with sufficient reliability that classification accuracy is not diminished. The presented models are fast to generate and may serve as easily implemented screening tools for personalized oncology medicine, drug repurposing, and drug discovery. [ABSTRACT FROM AUTHOR]

Alzheimer's disease and related dementias (ADRD) are highly prevalent conditions, and prior efforts to develop predictive models have relied on demographic and clinical risk factors using traditional logistical regression methods. We hypothesized that machine-learning algorithms using administrative claims data may represent a novel approach to predicting ADRD. Using a national de-identified dataset of more than 125 million patients including over 10,000 clinical, pharmaceutical, and demographic variables, we developed a cohort to train a machine learning model to predict ADRD 4–5 years in advance. The Lasso algorithm selected a 50-variable model with an area under the curve (AUC) of 0.693. Top diagnosis codes in the model were memory loss (780.93), Parkinson’s disease (332.0), mild cognitive impairment (331.83) and bipolar disorder (296.80), and top pharmacy codes were psychoactive drugs. Machine learning algorithms can rapidly develop predictive models for ADRD with massive datasets, without requiring hypothesis-driven feature engineering. [ABSTRACT FROM AUTHOR]

Microbial communities are ubiquitous and often influence macroscopic properties of the ecosystems they inhabit. However, deciphering the functional relationship between specific microbes and ecosystem properties is an ongoing challenge owing to the complexity of the communities. This challenge can be addressed, in part, by integrating the advances in DNA sequencing technology with computational approaches like machine learning. Although machine learning techniques have been applied to microbiome data, use of these techniques remains rare, and user-friendly platforms to implement such techniques are not widely available. We developed a tool that implements neural network and random forest models to perform regression and feature selection tasks on microbiome data. In this study, we applied the tool to analyze soil microbiome (16S rRNA gene profiles) and dissolved organic carbon (DOC) data from a 44-day plant litter decomposition experiment. The microbiome data includes 1709 total bacterial operational taxonomic units (OTU) from 300+ microcosms. Regression analysis of predicted and actual DOC for a held-out test set of 51 samples yield Pearson’s correlation coefficients of.636 and.676 for neural network and random forest approaches, respectively. Important taxa identified by the machine learning techniques are compared to results from a standard tool (indicator species analysis) widely used by microbial ecologists. Of 1709 bacterial taxa, indicator species analysis identified 285 taxa as significant determinants of DOC concentration. Of the top 285 ranked features determined by machine learning methods, a subset of 86 taxa are common to all feature selection techniques. Using this subset of features, prediction results for random permutations of the data set are at least equally accurate compared to predictions determined using the entire feature set. Our results suggest that integration of multiple methods can aid identification of a robust subset of taxa within complex communities that may drive specific functional outcomes of interest. [ABSTRACT FROM AUTHOR]

As a promising research direction in recent decades, active learning allows an oracle to assign labels to typical examples for performance improvement in learning systems. Existing works mainly focus on designing criteria for screening examples of high value to be labeled in a handcrafted manner. Instead of manually developing strategies of querying the user to access labels for the desired examples, we utilized the reinforcement learning algorithm parameterized with the neural network to automatically explore query strategies in active learning when addressing stream-based one-shot classification problems. With the involvement of cross-entropy in the loss function of Q-learning, an efficient policy to decide when and where to predict or query an instance is learned through the developed framework. Compared with a former influential work, the advantages of our method are demonstrated experimentally with two image classification tasks, and it exhibited better performance, quick convergence, relatively good stability and fewer requests for labels. [ABSTRACT FROM AUTHOR]

A brain-computer interface (BCI), based on motor imagery EEG, uses information extracted from the electroencephalography signals generated by a person who intends to perform any action. One of the most important issues of current research is how to detect automatically whether the user intends to send some message to a certain device. This study presents a proposal, based on a hierarchical structured system, for recognising intentional and non-intentional mental tasks on a BCI system by applying machine learning techniques to the EEG signals. First-level clustering is performed to distinguish between intentional control (IC) and non-intentional control (NC) state patterns. Then, the patterns recognised as IC are passed on to a second stage where supervised learning techniques are used to classify them. In BCI applications, it is critical to correctly classify NC states with a low false positive rate (FPR) to avoid undesirable effects. According to the literature, we selected a maximum FPR of 10%. Under these conditions, our proposal achieved an average test accuracy of 66.6%, with an 8.2% FPR, for the BCI competition IIIa dataset. The main contribution of this paper is the hierarchical approach, based on machine learning paradigms, which performs intentional and non-intentional discrimination and, depending on the case, classifies the intended command selected by the user. [ABSTRACT FROM AUTHOR]

We have attempted to reproduce the results in Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs, published in JAMA 2016; 316(22), using publicly available data sets. We re-implemented the main method in the original study since the source code is not available. The original study used non-public fundus images from EyePACS and three hospitals in India for training. We used a different EyePACS data set from Kaggle. The original study used the benchmark data set Messidor-2 to evaluate the algorithm's performance. We used another distribution of the Messidor-2 data set, since the original data set is no longer available. In the original study, ophthalmologists re-graded all images for diabetic retinopathy, macular edema, and image gradability. We have one diabetic retinopathy grade per image for our data sets, and we assessed image gradability ourselves. We were not able to reproduce the original study's results with publicly available data. Our algorithm's area under the receiver operating characteristic curve (AUC) of 0.951 (95% CI, 0.947-0.956) on the Kaggle EyePACS test set and 0.853 (95% CI, 0.835-0.871) on Messidor-2 did not come close to the reported AUC of 0.99 on both test sets in the original study. This may be caused by the use of a single grade per image, or different data. This study shows the challenges of reproducing deep learning method results, and the need for more replication and reproduction studies to validate deep learning methods, especially for medical image analysis. Our source code and instructions are available at: . [ABSTRACT FROM AUTHOR]

DIBR-3D technology has evolved over the past few years with the demands of consumers increasing in recent times for future free-view 3D videos on their home televisions. The main issue in 3D technology is the lack of 3D content available to watch using the traditional TV systems. Although, some sophisticated devices like stereoscopic cameras have been used to fill the gap between the 3D content demand and 3D content supply. But the content generated through these sophisticated devices can not be displayed on the traditional TV systems, so there needs to be some mechanism which is inline with the traditional TV. Furthermore, the huge collection of existing 2D content should be converted to 3D using depth image-based rendering techniques. This conversion technique can highly contribute in overcoming the shortage problem of the 3D content. This paper presents a novel approach for converting 2D degraded image for DIBR 3D-TV view. This degraded or noisy/blur image is enhanced through image dehazing and Directional Filter Bank (DFB). This enhancement is necessary because of the occlusion effect or hole filling problem that occurs due to imperfect depth map. The enhanced image is then segmented into the foreground image and the background image. After the segmentation, the depth map is generated using image profiles. Moreover, Stereoscopic images are finally produced using the DIBR procedure which is based on the 2D input image and the corresponding depth map. We have verified the results of the proposed approach by comparing the results with the existing state-of-the-art techniques. [ABSTRACT FROM AUTHOR]

Electroencephalogram (EEG) is a common base signal used to monitor brain activities and diagnose sleep disorders. Manual sleep stage scoring is a time-consuming task for sleep experts and is limited by inter-rater reliability. In this paper, we propose an automatic sleep stage annotation method called SleepEEGNet using a single-channel EEG signal. The SleepEEGNet is composed of deep convolutional neural networks (CNNs) to extract time-invariant features, frequency information, and a sequence to sequence model to capture the complex and long short-term context dependencies between sleep epochs and scores. In addition, to reduce the effect of the class imbalance problem presented in the available sleep datasets, we applied novel loss functions to have an equal misclassified error for each sleep stage while training the network. We evaluated the performance of the proposed method on different single-EEG channels (i.e., Fpz-Cz and Pz-Oz EEG channels) from the Physionet Sleep-EDF datasets published in 2013 and 2018. The evaluation results demonstrate that the proposed method achieved the best annotation performance compared to current literature, with an overall accuracy of 84.26%, a macro F1-score of 79.66% and κ = 0.79. Our developed model can be applied to other sleep EEG signals and aid the sleep specialists to arrive at an accurate diagnosis. The source code is available at . [ABSTRACT FROM AUTHOR]

Objective: To compare axonal loss in ganglion cells detected with swept-source optical coherence tomography (SS-OCT) in eyes of patients with multiple sclerosis (MS) versus healthy controls using different machine learning techniques. To analyze the capability of machine learning techniques to improve the detection of retinal nerve fiber layer (RNFL) and the complex Ganglion Cell Layer–Inner plexiform layer (GCL+) damage in patients with multiple sclerosis and to use the SS-OCT as a biomarker to early predict this disease. Methods: Patients with relapsing-remitting MS (n = 80) and age-matched healthy controls (n = 180) were enrolled. Different protocols from the DRI SS-OCT Triton system were used to obtain the RNFL and GCL+ thicknesses in both eyes. Macular and peripapilar areas were analyzed to detect the zones with higher thickness decrease. The performance of different machine learning techniques (decision trees, multilayer perceptron and support vector machine) for identifying RNFL and GCL+ thickness loss in patients with MS were evaluated. Receiver-operating characteristic (ROC) curves were used to display the ability of the different tests to discriminate between MS and healthy eyes in our population. Results: Machine learning techniques provided an excellent tool to predict MS disease using SS-OCT data. In particular, the decision trees obtained the best prediction (97.24%) using RNFL data in macular area and the area under the ROC curve was 0.995, while the wide protocol which covers an extended area between macula and papilla gave an accuracy of 95.3% with a ROC of 0.998. Moreover, it was obtained that the most significant area of the RNFL to predict MS is the macula just surrounding the fovea. On the other hand, in our study, GCL+ did not contribute to predict MS and the different machine learning techniques performed worse in this layer than in RNFL. Conclusions: Measurements of RNFL thickness obtained with SS-OCT have an excellent ability to differentiate between healthy controls and patients with MS. Thus, the use of machine learning techniques based on these measures can be a reliable tool to help in MS diagnosis. [ABSTRACT FROM AUTHOR]

Antenna selection in Multiple-Input Multiple-Output (MIMO) systems has attracted increasing attention due to the challenge of keeping a balance between communication performance and computational complexity. Recently, deep learning based methods have achieved promising performance in many application fields. This paper proposed a deep learning (DL) based antenna selection technique. First, we generated the label of training antenna systems by maximizing the channel capacity. Then, we adopted the deep convolutional neural network (CNN) on the channel matrices to explicitly exploit the massive latent cues of attenuation coefficients. Finally, we used the adopted CNN to assign the class label and then select the optimal antenna subset. Experimental results demonstrate that our method can achieve better performance than the state-of-the-art baselines for data-driven based antenna selection. [ABSTRACT FROM AUTHOR]

An extreme learning machine (ELM) is a novel training method for single-hidden layer feedforward neural networks (SLFNs) in which the hidden nodes are randomly assigned and fixed without iterative tuning. ELMs have earned widespread global interest due to their fast learning speed, satisfactory generalization ability and ease of implementation. In this paper, we extend this theory to hypercomplex space and attempt to simultaneously consider multisource information using a hypercomplex representation. To illustrate the performance of the proposed hypercomplex extreme learning machine (HELM), we have applied this scheme to the task of multispectral palmprint recognition. Images from different spectral bands are utilized to construct the hypercomplex space. Extensive experiments conducted on the PolyU and CASIA multispectral databases demonstrate that the HELM scheme can achieve competitive results. The source code together with datasets involved in this paper can be available for free download at . [ABSTRACT FROM AUTHOR]

Background: The postmortem microbiome can provide valuable information to a death investigation and to the human health of the once living. Microbiome sequencing produces, in general, large multi-dimensional datasets that can be difficult to analyze and interpret. Machine learning methods can be useful in overcoming this analytical challenge. However, different methods employ distinct strategies to handle complex datasets. It is unclear whether one method is more appropriate than others for modeling postmortem microbiomes and their ability to predict attributes of interest in death investigations, which require understanding of how the microbial communities change after death and may represent those of the once living host. Methods and findings: Postmortem microbiomes were collected by swabbing five anatomical areas during routine death investigation, sequenced and analyzed from 188 death cases. Three machine learning methods (boosted algorithms, random forests, and neural networks) were compared with respect to their abilities to predict case attributes: postmortem interval (PMI), location of death, and manner of death. Accuracy depended on the method used, the numbers of anatomical areas analyzed, and the predicted attribute of death. Conclusions: All algorithms performed well but with distinct features to their performance. Xgboost often produced the most accurate predictions but may also be more prone to overfitting. Random forest was the most stable across predictions that included more anatomic areas. Analysis of postmortem microbiota from more than three anatomic areas appears to yield limited returns on accuracy, with the eyes and rectum providing the most useful information correlating with circumstances of death in most cases for this dataset. [ABSTRACT FROM AUTHOR]

Ex-ante online risk assessment for building emergency evacuation is essential to protect human life and property. Current risk assessment methods are limited by the tradeoff between accuracy and efficiency. In this paper, we propose an online method that overcomes this tradeoff based on multimedia data (e.g. videos data from surveillance cameras) and deep learning. The method consists of two parts. The first estimates the evacuee position as input for training the assessment model to then perform risk assessment in real scenarios. The second considers a social force model based on the evacuation simulation for the output of training model. We verify the proposed method in simulation and real scenarios. Model sensitivity analyses and large-scale tests demonstrate the usability and superiority of the proposed method. By the method, the computation time of risk assessment could be decreased from 10 minutes (by traditional simulation method) to 2.18 s. [ABSTRACT FROM AUTHOR]

Restoration of communication in people with complete motor paralysis—a condition called complete locked-in state (CLIS)—is one of the greatest challenges of brain-computer interface (BCI) research. New findings have recently been presented that bring us one step closer to this goal. However, the validity of the evidence has been questioned: independent reanalysis of the same data yielded significantly different results. Reasons for the failure to replicate the findings must be of a methodological nature. What is the best practice to ensure that results are stringent and conclusive and analyses replicable? Confirmation bias and the counterintuitive nature of probability may lead to an overly optimistic interpretation of new evidence. Lack of detail complicates replicability. [ABSTRACT FROM AUTHOR]

Purpose: To determine if deep learning networks could be trained to forecast future 24–2 Humphrey Visual Fields (HVFs). Methods: All data points from consecutive 24–2 HVFs from 1998 to 2018 were extracted from a university 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. The point-wise mean absolute error (PMAE) and difference in Mean Deviation (MD) between predicted and actual future HVF were calculated. Results: More 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 point-wise PMAE for the test set was 2.47 dB (95% CI: 2.45 dB to 2.48 dB), and deep learning showed a statistically significant improvement over linear models. The 100 fully trained models successfully predicted future HVFs 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 and an average difference of 0.41 dB. Conclusions: Using unfiltered real-world datasets, deep learning networks show the 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. [ABSTRACT FROM AUTHOR]

Objectives: We propose a bottom-up, machine-learning approach, for the objective vestibular and balance diagnostic data of concussion patients, to provide insight into the differences in patients’ phenotypes, independent of existing diagnoses (unsupervised learning). Methods: Diagnostic data from a battery of validated balance and vestibular assessments were extracted from the database of the Swiss Concussion Center. The desired number of clusters within the patient database was estimated using Calinski-Harabasz criteria. Complex (self-organizing map, SOM) and standard (k-means) clustering tools were used, and the formed clusters were compared. Results: A total of 96 patients (81.3% male, age (median [IQR]): 25.0[10.8]) who were expected to suffer from sports-related concussion or post-concussive syndrome (52[140] days between diagnostic testing and the concussive episode) were included. The cluster evaluation indicated dividing the data into two groups. Only the SOM gave a stable clustering outcome, dividing the patients in group-1 (n = 38) and group-2 (n = 58). A large significant difference was found for the caloric summary score for the maximal speed of the slow phase, where group-1 scored 30.7% lower than group-2 (27.6[18.2] vs. 51.0[31.0]). Group-1 also scored significantly lower on the sensory organisation test composite score (69.0[22.3] vs. 79.0[10.5]) and higher on the visual acuity (-0.03[0.33] vs. -0.14[0.12]) and dynamic visual acuity (0.38[0.84] vs. 0.20[0.20]) tests. The importance of caloric, SOT and DVA, was supported by the PCA outcomes. Group-1 tended to report headaches, blurred vision and balance problems more frequently than group-2 (>10% difference). Conclusion: The SOM divided the data into one group with prominent vestibular disorders and another with no clear vestibular or balance problems, suggesting that artificial intelligence might help improve the diagnostic process. [ABSTRACT FROM AUTHOR]

Identifying university students’ weaknesses results in better learning and can function as an early warning system to enable students to improve. However, the satisfaction level of existing systems is not promising. New and dynamic hybrid systems are needed to imitate this mechanism. A hybrid system (a modified Recurrent Neural Network with an adapted Grey Wolf Optimizer) is used to forecast students’ outcomes. This proposed system would improve instruction by the faculty and enhance the students’ learning experiences. The results show that a modified recurrent neural network with an adapted Grey Wolf Optimizer has the best accuracy when compared with other models. [ABSTRACT FROM AUTHOR]

In a recidivism prediction context, there is no consensus on which modeling strategy should be followed for obtaining an optimal prediction model. In previous papers, a range of statistical and machine learning techniques were benchmarked on recidivism data with a binary outcome. However, two important tree ensemble methods, namely gradient boosting and random forests were not extensively evaluated. In this paper, we further explore the modeling potential of these techniques in the binary outcome criminal prediction context. Additionally, we explore the predictive potential of classical statistical and machine learning methods for censored time-to-event data. A range of statistical manually specified statistical and (semi-)automatic machine learning models is fitted on Dutch recidivism data, both for the binary outcome case and censored outcome case. To enhance generalizability of results, the same models are applied to two historical American data sets, the North Carolina prison data. For all datasets, (semi-) automatic modeling in the binary case seems to provide no improvement over an appropriately manually specified traditional statistical model. There is however evidence of slightly improved performance of gradient boosting in survival data. Results on the reconviction data from two sources suggest that both statistical and machine learning should be tried out for obtaining an optimal model. Even if a flexible black-box model does not improve upon the predictions of a manually specified model, it can serve as a test whether important interactions are missing or other misspecification of the model are present and can thus provide more security in the modeling process. [ABSTRACT FROM AUTHOR]

Background and purpose: Architecture of the cerebral network has been shown to associate with IQ in children with epilepsy. However, subject-level prediction on this basis, a crucial step toward harnessing network analyses for the benefit of children with epilepsy, has yet to be achieved. We compared two network normalization strategies in terms of their ability to optimize subject-level inferences on the relationship between brain network architecture and brain function. Materials and methods: Patients with epilepsy and resting state fMRI were retrospectively identified. Brain network nodes were defined by anatomic parcellation, first in patient space (nodes defined for each patient) and again in template space (same nodes for all patients). Whole-brain weighted graphs were constructed according to pair-wise correlation of BOLD-signal time courses between nodes. The following metrics were then calculated: clustering coefficient, transitivity, modularity, path length, and global efficiency. Metrics computed on graphs in patient space were normalized to the same metric computed on a random network of identical size. A machine learning algorithm was used to predict patient IQ given access to only the network metrics. Results: Twenty-seven patients (8–18 years) comprised the final study group. All brain networks demonstrated expected small world properties. Accounting for intrinsic population heterogeneity had a significant effect on prediction accuracy. Specifically, transformation of all patients into a common standard space as well as normalization of metrics to those computed on a random network both substantially outperformed the use of non-normalized metrics. Conclusion: Normalization contributed significantly to accurate subject-level prediction of cognitive function in children with epilepsy. These findings support the potential for quantitative network approaches to contribute clinically meaningful information in children with neurological disorders. [ABSTRACT FROM AUTHOR]

Based on morphology it is often challenging to distinguish between the many different soft tissue sarcoma subtypes. Moreover, outcome of disease is highly variable even between patients with the same disease. Machine learning on transcriptome sequencing data could be a valuable new tool to understand differences between and within entities. Here we used machine learning analysis to identify novel diagnostic and prognostic markers and therapeutic targets for soft tissue sarcomas. Gene expression data was used from the Cancer Genome Atlas, the Genotype-Tissue Expression data and the French Sarcoma Group. We identified three groups of tumors that overlap in their molecular profile as seen with unsupervised t-Distributed Stochastic Neighbor Embedding clustering and a deep neural network. The three groups corresponded to subtypes that are morphologically overlapping. Using a random forest algorithm, we identified novel diagnostic markers for soft tissue sarcoma that distinguished between synovial sarcoma and MPNST, and that we validated using qRT-PCR in an independent series. Next, we identified prognostic genes that are strong predictors of disease outcome when used in a k-nearest neighbor algorithm. The prognostic genes were further validated in expression data from the French Sarcoma Group. One of these, HMMR, was validated in an independent series of leiomyosarcomas using immunohistochemistry on tissue micro array as a prognostic gene for disease-free interval. Furthermore, reconstruction of regulatory networks combined with data from the Connectivity Map showed, amongst others, that HDAC inhibitors could be a potential effective therapy for multiple soft tissue sarcoma subtypes. A viability assay with two HDAC inhibitors confirmed that both leiomyosarcoma and synovial sarcoma are sensitive to HDAC inhibition. In this study we identified novel diagnostic markers, prognostic markers and therapeutic leads from multiple soft tissue sarcoma gene expression datasets. Thus, machine learning algorithms are powerful new tools to improve our understanding of rare tumor entities. [ABSTRACT FROM AUTHOR]

Background: In an intensive care units, experts in mechanical ventilation are not continuously at patient’s bedside to adjust ventilation settings and to analyze the impact of these adjustments on gas exchange. The development of clinical decision support systems analyzing patients’ data in real time offers an opportunity to fill this gap. Objective: The objective of this study was to determine whether a machine learning predictive model could be trained on a set of clinical data and used to predict transcutaneous hemoglobin oxygen saturation 5 min (5min SpO2) after a ventilator setting change. Data sources: Data of mechanically ventilated children admitted between May 2015 and April 2017 were included and extracted from a high-resolution research database. More than 776,727 data rows were obtained from 610 patients, discretized into 3 class labels (< 84%, 85% to 91% and c92% to 100%). Performance metrics of predictive models: Due to data imbalance, four different data balancing processes were applied. Then, two machine learning models (artificial neural network and Bootstrap aggregation of complex decision trees) were trained and tested on these four different balanced datasets. The best model predicted SpO2 with area under the curves < 0.75. Conclusion: This single center pilot study using machine learning predictive model resulted in an algorithm with poor accuracy. The comparison of machine learning models showed that bagged complex trees was a promising approach. However, there is a need to improve these models before incorporating them into a clinical decision support systems. One potentially solution for improving predictive model, would be to increase the amount of data available to limit over-fitting that is potentially one of the cause for poor classification performances for 2 of the three class labels. [ABSTRACT FROM AUTHOR]

Place cells in the mammalian hippocampus signal self-location with sparse spatially stable firing fields. Based on observation of place cell activity it is possible to accurately decode an animal’s location. The precision of this decoding sets a lower bound for the amount of information that the hippocampal population conveys about the location of the animal. In this work we use a novel recurrent neural network (RNN) decoder to infer the location of freely moving rats from single unit hippocampal recordings. RNNs are biologically plausible models of neural circuits that learn to incorporate relevant temporal context without the need to make complicated assumptions about the use of prior information to predict the current state. When decoding animal position from spike counts in 1D and 2D-environments, we show that the RNN consistently outperforms a standard Bayesian approach with either flat priors or with memory. In addition, we also conducted a set of sensitivity analysis on the RNN decoder to determine which neurons and sections of firing fields were the most influential. We found that the application of RNNs to neural data allowed flexible integration of temporal context, yielding improved accuracy relative to the more commonly used Bayesian approaches and opens new avenues for exploration of the neural code. [ABSTRACT FROM AUTHOR]

In general, the stock prices of the same industry have a similar trend, but those of different industries do not. When investing in stocks of different industries, one should select the optimal model from lots of trading models for each industry because any model may not be suitable for capturing the stock trends of all industries. However, the study has not been carried out at present. In this paper, firstly we select 424 S&P 500 index component stocks (SPICS) and 185 CSI 300 index component stocks (CSICS) as the research objects from 2010 to 2017, divide them into 9 industries such as finance and energy respectively. Secondly, we apply 12 widely used machine learning algorithms to generate stock trading signals in different industries and execute the back-testing based on the trading signals. Thirdly, we use a non-parametric statistical test to evaluate whether there are significant differences among the trading performance evaluation indicators (PEI) of different models in the same industry. Finally, we propose a series of rules to select the optimal models for stock investment of every industry. The analytical results on SPICS and CSICS show that we can find the optimal trading models for each industry based on the statistical tests and the rules. Most importantly, the PEI of the best algorithms can be significantly better than that of the benchmark index and “Buy and Hold” strategy. Therefore, the algorithms can be used for making profits from industry stock trading. [ABSTRACT FROM AUTHOR]

Annotation of foot-contact and foot-off events is the initial step in post-processing for most quantitative gait analysis workflows. If clean force plate strikes are present, the events can be automatically detected. Otherwise, annotation of gait events is performed manually, since reliable automatic tools are not available. Automatic annotation methods have been proposed for normal gait, but are usually based on heuristics of the coordinates and velocities of motion capture markers placed on the feet. These heuristics do not generalize to pathological gait due to greater variability in kinematics and anatomy of patients, as well as the presence of assistive devices. In this paper, we use a data-driven approach to predict foot-contact and foot-off events from kinematic and marker time series in children with normal and pathological gait. Through analysis of 9092 gait cycle measurements we build a predictive model using Long Short-Term Memory (LSTM) artificial neural networks. The best-performing model identifies foot-contact and foot-off events with an average error of 10 and 13 milliseconds respectively, outperforming popular heuristic-based approaches. We conclude that the accuracy of our approach is sufficient for most clinical and research applications in the pediatric population. Moreover, the LSTM architecture enables real-time predictions, enabling applications for real-time control of active assistive devices, orthoses, or prostheses. We provide the model, usage examples, and the training code in an open-source package. [ABSTRACT FROM AUTHOR]

Early prediction of the potential for neurological recovery after resuscitation from cardiac arrest is difficult but important. Currently, no clinical finding or combination of findings are sufficient to accurately predict or preclude favorable recovery of comatose patients in the first 24 to 48 hours after resuscitation. Thus, life-sustaining therapy is often continued for several days in patients whose irrecoverable injury is not yet recognized. Conversely, early withdrawal of life-sustaining therapy increases mortality among patients who otherwise might have gone on to recover. In this work, we present Canonical Autocorrelation Analysis (CAA) and Canonical Autocorrelation Embeddings (CAE), novel methods suitable for identifying complex patterns in high-resolution multivariate data often collected in highly monitored clinical environments such as intensive care units. CAE embeds sets of datapoints onto a space that characterizes their latent correlation structures and allows direct comparison of these structures through the use of a distance metric. The methodology may be particularly suitable when the unit of analysis is not just an individual datapoint but a dataset, as for instance in patients for whom physiological measures are recorded over time, and where changes of correlation patterns in these datasets are informative for the task at hand. We present a proof of concept to illustrate the potential utility of CAE by applying it to characterize electroencephalographic recordings from 80 comatose survivors of cardiac arrest, aiming to identify patients who will survive to hospital discharge with favorable functional recovery. Our results show that with very low probability of making a Type 1 error, we are able to identify 32.5% of patients who are likely to have a good neurological outcome, some of whom have otherwise unfavorable clinical characteristics. Importantly, some of these had 5% predicted chance of favorable recovery based on initial illness severity measures alone. Providing this information to support clinical decision-making could motivate the continuation of life-sustaining therapies for these patients. [ABSTRACT FROM AUTHOR]

Many real-world systems can be studied in terms of pattern recognition tasks, so that proper use (and understanding) of machine learning methods in practical applications becomes essential. While many classification methods have been proposed, there is no consensus on which methods are more suitable for a given dataset. As a consequence, it is important to comprehensively compare methods in many possible scenarios. In this context, we performed a systematic comparison of 9 well-known clustering methods available in the R language assuming normally distributed data. In order to account for the many possible variations of data, we considered artificial datasets with several tunable properties (number of classes, separation between classes, etc). In addition, we also evaluated the sensitivity of the clustering methods with regard to their parameters configuration. The results revealed that, when considering the default configurations of the adopted methods, the spectral approach tended to present particularly good performance. We also found that the default configuration of the adopted implementations was not always accurate. In these cases, a simple approach based on random selection of parameters values proved to be a good alternative to improve the performance. All in all, the reported approach provides subsidies guiding the choice of clustering algorithms. [ABSTRACT FROM AUTHOR]

Different performance measures are used to assess the behaviour, and to carry out the comparison, of classifiers in Machine Learning. Many measures have been defined on the literature, and among them, a measure inspired by Shannon’s entropy named the Confusion Entropy (CEN). In this work we introduce a new measure, MCEN, by modifying CEN to avoid its unwanted behaviour in the binary case, that disables it as a suitable performance measure in classification. We compare MCEN with CEN and other performance measures, presenting analytical results in some particularly interesting cases, as well as some heuristic computational experimentation. [ABSTRACT FROM AUTHOR]

The high failure rate of therapeutics showing promise in mouse models to translate to patients is a pressing challenge in biomedical science. Though retrospective studies have examined the fidelity of mouse models to their respective human conditions, approaches for prospective translation of insights from mouse models to patients remain relatively unexplored. Here, we develop a semi-supervised learning approach for inference of disease-associated human differentially expressed genes and pathways from mouse model experiments. We examined 36 transcriptomic case studies where comparable phenotypes were available for mouse and human inflammatory diseases and assessed multiple computational approaches for inferring human biology from mouse datasets. We found that semi-supervised training of a neural network identified significantly more true human biological associations than interpreting mouse experiments directly. Evaluating the experimental design of mouse experiments where our model was most successful revealed principles of experimental design that may improve translational performance. Our study shows that when prospectively evaluating biological associations in mouse studies, semi-supervised learning approaches, combining mouse and human data for biological inference, provide the most accurate assessment of human in vivo disease processes. Finally, we proffer a delineation of four categories of model system-to-human “Translation Problems” defined by the resolution and coverage of the datasets available for molecular insight translation and suggest that the task of translating insights from model systems to human disease contexts may be better accomplished by a combination of translation-minded experimental design and computational approaches. [ABSTRACT FROM AUTHOR]

k-nearest neighbour (k-NN) has been shown to be an effective learning algorithm for classification and prediction. However, the application of k-NN for prediction and classification in specific sport is still in its infancy. The present study classified and predicted high and low potential archers from a set of physical fitness variables trained on a variation of k-NN algorithms and logistic regression. 50 youth archers with the mean age and standard deviation of (17.0 ± 0.56) years drawn from various archery programmes completed a one end archery shooting score test. Standard fitness measurements of the handgrip, vertical jump, standing broad jump, static balance, upper muscle strength and the core muscle strength were conducted. Multiple linear regression was utilised to ascertain the significant variables that affect the shooting score. It was demonstrated from the analysis that core muscle strength and vertical jump were statistically significant. Hierarchical agglomerative cluster analysis (HACA) was used to cluster the archers based on the significant variables identified. k-NN model variations, i.e., fine, medium, coarse, cosine, cubic and weighted functions as well as logistic regression, were trained based on the significant performance variables. The HACA clustered the archers into high potential archers (HPA) and low potential archers (LPA). The weighted k-NN outperformed all the tested models at itdemonstrated reasonably good classification on the evaluated indicators with an accuracy of 82.5 ± 4.75% for the prediction of the HPA and the LPA. Moreover, the performance of the classifiers was further investigated against fresh data, which also indicates the efficacy of the weighted k-NN model. These findings could be valuable to coaches and sports managers to recognise high potential archers from a combination of the selected few physical fitness performance indicators identified which would subsequently save cost, time and energy for a talent identification programme. [ABSTRACT FROM AUTHOR]

D-Wave quantum annealers represent a novel computational architecture and have attracted significant interest. Much of this interest has focused on the quantum behavior of D-Wave machines, and there have been few practical algorithms that use the D-Wave. Machine learning has been identified as an area where quantum annealing may be useful. Here, we show that the D-Wave 2X can be effectively used as part of an unsupervised machine learning method. This method takes a matrix as input and produces two low-rank matrices as output—one containing latent features in the data and another matrix describing how the features can be combined to approximately reproduce the input matrix. Despite the limited number of bits in the D-Wave hardware, this method is capable of handling a large input matrix. The D-Wave only limits the rank of the two output matrices. We apply this method to learn the features from a set of facial images and compare the performance of the D-Wave to two classical tools. This method is able to learn facial features and accurately reproduce the set of facial images. The performance of the D-Wave shows some promise, but has some limitations. It outperforms the two classical codes in a benchmark when only a short amount of computational time is allowed (200-20,000 microseconds), but these results suggest heuristics that would likely outperform the D-Wave in this benchmark. [ABSTRACT FROM AUTHOR]

Purpose: To test the ability of machine learning classifiers (MLCs) using optical coherence tomography (OCT) and standard automated perimetry (SAP) parameters to discriminate between healthy and glaucomatous individuals, and to compare it to the diagnostic ability of the combined structure-function index (CSFI), general ophthalmologists and glaucoma specialists. Design: Cross-sectional prospective study. Methods: Fifty eight eyes of 58 patients with early to moderate glaucoma (median value of the mean deviation = −3.44 dB; interquartile range, -6.0 to -2.4 dB) and 66 eyes of 66 healthy individuals underwent OCT and SAP tests. The diagnostic accuracy (area under the ROC curve—AUC) of 10 MLCs was compared to those obtained with the CSFI, 3 general ophthalmologists and 3 glaucoma specialists exposed to the same OCT and SAP data. Results: The AUCs obtained with MLCs ranged from 0.805 (Classification Tree) to 0.931 (Radial Basis Function Network, RBF). The sensitivity at 90% specificity ranged from 51.6% (Classification Tree) to 82.8% (Bagging, Multilayer Perceptron and Support Vector Machine Gaussian). The CSFI had a sensitivity of 79.3% at 90% specificity, and the highest AUC (0.948). General ophthalmologists and glaucoma specialists’ grading had sensitivities of 66.2% and 83.8% at 90% specificity, and AUCs of 0.879 and 0.921, respectively. RBF (the best MLC), the CSFI, and glaucoma specialists showed significantly higher AUCs than that obtained by general ophthalmologists (P<0.05). However, there were no significant differences between the AUCs obtained by RBF, the CSFI, and glaucoma specialists (P>0.25). Conclusion: Our findings suggest that both MLCs and the CSFI can be helpful in clinical practice and effectively improve glaucoma diagnosis in the primary eye care setting, when there is no glaucoma specialist available. [ABSTRACT FROM AUTHOR]

Background: Resuscitated cardiac arrest is associated with high mortality; however, the ability to estimate risk of adverse outcomes using existing illness severity scores is limited. Using in-hospital data available within the first 24 hours of admission, we aimed to develop more accurate models of risk prediction using both logistic regression (LR) and machine learning (ML) techniques, with a combination of demographic, physiologic, and biochemical information.Methods and Findings: Patient-level data were extracted from the Australian and New Zealand Intensive Care Society (ANZICS) Adult Patient Database for patients who had experienced a cardiac arrest within 24 hours prior to admission to an intensive care unit (ICU) during the period January 2006 to December 2016. The primary outcome was in-hospital mortality. The models were trained and tested on a dataset (split 90:10) including age, lowest and highest physiologic variables during the first 24 hours, and key past medical history. LR and 5 ML approaches (gradient boosting machine [GBM], support vector classifier [SVC], random forest [RF], artificial neural network [ANN], and an ensemble) were compared to the APACHE III and Australian and New Zealand Risk of Death (ANZROD) predictions. In all, 39,566 patients from 186 ICUs were analysed. Mean (±SD) age was 61 ± 17 years; 65% were male. Overall in-hospital mortality was 45.5%. Models were evaluated in the test set. The APACHE III and ANZROD scores demonstrated good discrimination (area under the receiver operating characteristic curve [AUROC] = 0.80 [95% CI 0.79-0.82] and 0.81 [95% CI 0.8-0.82], respectively) and modest calibration (Brier score 0.19 for both), which was slightly improved by LR (AUROC = 0.82 [95% CI 0.81-0.83], DeLong test, p < 0.001). Discrimination was significantly improved using ML models (ensemble and GBM AUROCs = 0.87 [95% CI 0.86-0.88], DeLong test, p < 0.001), with an improvement in performance (Brier score reduction of 22%). Explainability models were created to assist in identifying the physiologic features that most contributed to an individual patient's survival. Key limitations include the absence of pre-hospital data and absence of external validation.Conclusions: ML approaches significantly enhance predictive discrimination for mortality following cardiac arrest compared to existing illness severity scores and LR, without the use of pre-hospital data. The discriminative ability of these ML models requires validation in external cohorts to establish generalisability. [ABSTRACT FROM AUTHOR]

Purpose: To build a deep learning model to diagnose glaucoma using fundus photography. Design: Cross sectional case study Subjects, Participants and Controls: A total of 1,542 photos (786 normal controls, 467 advanced glaucoma and 289 early glaucoma patients) were obtained by fundus photography. Method: The whole dataset of 1,542 images were split into 754 training, 324 validation and 464 test datasets. These datasets were used to construct simple logistic classification and convolutional neural network using Tensorflow. The same datasets were used to fine tune pre-trained GoogleNet Inception v3 model. Results: The simple logistic classification model showed a training accuracy of 82.9%, validation accuracy of 79.9% and test accuracy of 77.2%. Convolutional neural network achieved accuracy and area under the receiver operating characteristic curve (AUROC) of 92.2% and 0.98 on the training data, 88.6% and 0.95 on the validation data, and 87.9% and 0.94 on the test data. Transfer-learned GoogleNet Inception v3 model achieved accuracy and AUROC of 99.7% and 0.99 on training data, 87.7% and 0.95 on validation data, and 84.5% and 0.93 on test data. Conclusion: Both advanced and early glaucoma could be correctly detected via machine learning, using only fundus photographs. Our new model that is trained using convolutional neural network is more efficient for the diagnosis of early glaucoma than previously published models. [ABSTRACT FROM AUTHOR]

Big data classification is very slow when using traditional machine learning classifiers, particularly when using a lazy and slow-by-nature classifier such as the k-nearest neighbors algorithm (KNN). This paper proposes a new approach which is based on sorting the feature vectors of training data in a binary search tree to accelerate big data classification using the KNN approach. This is done using two methods, both of which utilize two local points to sort the examples based on their similarity to these local points. The first method chooses the local points based on their similarity to the global extreme points, while the second method chooses the local points randomly. The results of various experiments conducted on different big datasets show reasonable accuracy rates compared to state-of-the-art methods and the KNN classifier itself. More importantly, they show the high classification speed of both methods. This strong trait can be used to further improve the accuracy of the proposed methods. [ABSTRACT FROM AUTHOR]

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. [ABSTRACT FROM AUTHOR]

Background: The importance of identifying and evaluating adverse drug reactions (ADRs) has been widely recognized. Many studies have developed algorithms for ADR signal detection using electronic health record (EHR) data. In this study, we propose a machine learning (ML) model that enables accurate ADR signal detection by integrating features from existing algorithms based on inpatient EHR laboratory results. Materials and methods: To construct an ADR reference dataset, we extracted known drug–laboratory event pairs represented by a laboratory test from the EU-SPC and SIDER databases. All possible drug–laboratory event pairs, except known ones, are considered unknown. To detect a known drug–laboratory event pair, three existing algorithms—CERT, CLEAR, and PACE—were applied to 21-year inpatient EHR data. We also constructed ML models (based on random forest, L1 regularized logistic regression, support vector machine, and a neural network) that use the intermediate products of the CERT, CLEAR, and PACE algorithms as inputs and determine whether a drug–laboratory event pair is associated. For performance comparison, we evaluated the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), F1-measure, and area under receiver operating characteristic (AUROC). Results: All measures of ML models outperformed those of existing algorithms with sensitivity of 0.593–0.793, specificity of 0.619–0.796, NPV of 0.645–0.727, PPV of 0.680–0.777, F1-measure of 0.629–0.709, and AUROC of 0.737–0.816. Features related to change or distribution of shape were considered important for detecting ADR signals. Conclusions: Improved performance of ML models indicated that applying our model to EHR data is feasible and promising for detecting more accurate and comprehensive ADR signals. [ABSTRACT FROM AUTHOR]

Background: Prognostication is an essential tool for risk adjustment and decision making in the intensive care unit (ICU). Research into prognostication in ICU has so far been limited to data from admission or the first 24 hours. Most ICU admissions last longer than this, decisions are made throughout an admission, and some admissions are explicitly intended as time-limited prognostic trials. Despite this, temporal changes in prognostic ability during ICU admission has received little attention to date. Current predictive models, in the form of prognostic clinical tools, are typically derived from linear models and do not explicitly handle incremental information from trends. Machine learning (ML) allows predictive models to be developed which use non-linear predictors and complex interactions between variables, thus allowing incorporation of trends in measured variables over time; this has made it possible to investigate prognosis throughout an admission. Methods and findings: This study uses ML to assess the predictability of ICU mortality as a function of time. Logistic regression against physiological data alone outperformed APACHE-II and demonstrated several important interactions including between lactate & noradrenaline dose, between lactate & MAP, and between age & MAP consistent with the current sepsis definitions. ML models consistently outperformed logistic regression with Deep Learning giving the best results. Predictive power was maximal on the second day and was further improved by incorporating trend data. Using a limited range of physiological and demographic variables, the best machine learning model on the first day showed an area under the receiver-operator characteristic curve (AUC) of 0.883 (σ = 0.008), compared to 0.846 (σ = 0.010) for a logistic regression from the same predictors and 0.836 (σ = 0.007) for a logistic regression based on the APACHE-II score. Adding information gathered on the second day of admission improved the maximum AUC to 0.895 (σ = 0.008). Beyond the second day, predictive ability declined. Conclusion: This has implications for decision making in intensive care and provides a justification for time-limited trials of ICU therapy; the assessment of prognosis over more than one day may be a valuable strategy as new information on the second day helps to differentiate outcomes. New ML models based on trend data beyond the first day could greatly improve upon current risk stratification tools. [ABSTRACT FROM AUTHOR]

Mind wandering (MW) is extremely common during driving and is often accompanied by performance losses. This study investigated the use of driving behavior measurements to automatically detect mind wandering state in the driving task. In the experiment, participants (N = 40) performed a car-following task in a driving simulator and reported, upon hearing a tone, whether they were experiencing mind wandering or not. Supervised machine learning techniques were applied to classify MW-absent versus MW-present state, using both driver-independent and driver-dependent modeling methods. In the driver-independent modeling, we separately built models for participants with high or low MW and participants with medium MW. The optimal models can not offer a significant improvement than other models. So building effective driver-independent models with the leave-one-participant-out cross-validation method is challenging. In the driver-dependent modeling, we built models for each participant with medium MW. The best models of some participants were effective. The results indicate the development of mind wandering detecting system should take into account both inter-individual and intra-individual difference. This study provides a step toward minimizing the negative impacts of mindless driving and should benefit other fields of psychological research. [ABSTRACT FROM AUTHOR]

Intrinsic Connectivity Networks, patterns of correlated activity emerging from “resting-state” BOLD time series, are increasingly being associated with cognitive, clinical, and behavioral aspects, and compared with patterns of activity elicited by specific tasks. We study the reconfiguration of brain networks between task and resting-state conditions by a machine learning approach, to highlight the Intrinsic Connectivity Networks (ICNs) which are more affected by the change of network configurations in task vs. rest. To this end, we use a large cohort of publicly available data in both resting and task-based fMRI paradigms. By applying a battery of different supervised classifiers relying only on task-based measurements, we show that the highest accuracy to predict ICNs is reached with a simple neural network of one hidden layer. In addition, when testing the fitted model on resting state measurements, such architecture yields a performance close to 90% for areas connected to the task performed, which mainly involve the visual and sensorimotor cortex, whilst a relevant decrease of the performance is observed in the other ICNs. On one hand, our results confirm the correspondence of ICNs in both paradigms (task and resting) thus opening a window for future clinical applications to subjects whose participation in a required task cannot be guaranteed. On the other hand it is shown that brain areas not involved in the task display different connectivity patterns in the two paradigms. [ABSTRACT FROM AUTHOR]

Multivoxel pattern analysis (MVPA) typically begins with the estimation of single trial activation levels, and several studies have examined how different procedures for estimating single trial activity affect the ultimate classification accuracy of MVPA. Here we show that the currently preferred estimation procedures impart spurious positive correlations between the means of different category activity estimates within the same scanner run. In other words, if the mean of the estimates for one type of trials is high (low) in a given scanner run, then the mean of the other type of trials is also high (low) for that same scanner run, and the run-level mean across all trials therefore shifts from run to run. Simulations show that these correlations occur whenever there is a need to deconvolve overlapping trial activities in the presence of noise. We show that subtracting each voxel’s run-level mean across all trials from all the estimates within that run (i.e., run-level mean centering of estimates), by cancelling out these mean shifts, leads to robust and significant improvements in MVPA classification accuracy. These improvements are seen in both simulated and real data across a wide variety of situations. However, we also point out that there could be cases when mean activations are expected to shift across runs and that run-level mean centering could be detrimental in some of these cases (e.g., different proportion of trial types between different runs). [ABSTRACT FROM AUTHOR]

Although a standard reinforcement learning model can capture many aspects of reward-seeking behaviors, it may not be practical for modeling human natural behaviors because of the richness of dynamic environments and limitations in cognitive resources. We propose a modular reinforcement learning model that addresses these factors. Based on this model, a modular inverse reinforcement learning algorithm is developed to estimate both the rewards and discount factors from human behavioral data, which allows predictions of human navigation behaviors in virtual reality with high accuracy across different subjects and with different tasks. Complex human navigation trajectories in novel environments can be reproduced by an artificial agent that is based on the modular model. This model provides a strategy for estimating the subjective value of actions and how they influence sensory-motor decisions in natural behavior. [ABSTRACT FROM AUTHOR]