An Extensive-Form Game Paradigm for Visual Field Testing via Deep Reinforcement Learning.

Glaucoma is the leading cause of irreversible but preventable blindness worldwide, and visual field testing is an important tool for its diagnosis and monitoring. Testing using standard visual field thresholding procedures is time-consuming, and prolonged test duration leads to patient fatigue and decreased test reliability. Different visual field testing algorithms have been developed to shorten testing time while maintaining accuracy. However, the performance of these algorithms depends heavily on prior knowledge and manually crafted rules that determine the intensity of each light stimulus as well as the termination criteria, which is suboptimal. We leverage deep reinforcement learning to find improved decision strategies for visual field testing. In our proposed algorithms, multiple intelligent agents are employed to interact with the patient in an extensive-form game fashion, with each agent controlling the test on one of the testing locations in the patient's visual field. Through training, each agent learns an optimized policy that determines the intensities of light stimuli and the termination criteria, which minimizes the error in sensitivity estimation and test duration at the same time. In simulation experiments, we compare the performance of our algorithms against baseline visual field testing algorithms and show that our algorithms achieve a better trade-off between estimation accuracy and test duration. By retaining testing accuracy with reduced test duration, our algorithms improve test reliability, clinic efficiency, and patient satisfaction, and translationally affect clinical outcomes.

RGC-Net: An Automatic Reconstruction and Quantification Algorithm for Retinal Ganglion Cells Based on Deep Learning.

Purpose: The purpose of this study was to develop a deep learning-based fully automated reconstruction and quantification algorithm which automatically delineates the neurites and somas of retinal ganglion cells (RGCs). Methods: We trained a deep learning-based multi-task image segmentation model, RGC-Net, that automatically segments the neurites and somas in RGC images. A total of 166 RGC scans with manual annotations from human experts were used to develop this model, whereas 132 scans were used for training, and the remaining 34 scans were reserved as testing data. Post-processing techniques removed speckles or dead cells in soma segmentation results to further improve the robustness of the model. Quantification analyses were also conducted to compare five different metrics obtained by our automated algorithm and manual annotations. Results: Quantitatively, our segmentation model achieves average foreground accuracy, background accuracy, overall accuracy, and dice similarity coefficient of 0.692, 0.999, 0.997, and 0.691 for the neurite segmentation task, and 0.865, 0.999, 0.997, and 0.850 for the soma segmentation task, respectively. Conclusions: The experimental results demonstrate that RGC-Net can accurately and reliably reconstruct neurites and somas in RGC images. We also demonstrate our algorithm is comparable to human manually curated annotations in quantification analyses. Translational Relevance: Our deep learning model provides a new tool that can trace and analyze the RGC neurites and somas efficiently and faster than manual analysis.

Objective Measurement of Social Gaze and Smile Behaviors in Children with Suspected Autism Spectrum Disorder During Administration of the Autism Diagnostic Observation Schedule, 2nd Edition.

Best practice for the assessment of autism spectrum disorder (ASD) symptom severity relies on clinician ratings of the Autism Diagnostic Observation Schedule, 2nd Edition (ADOS-2), but the association of these ratings with objective measures of children's social gaze and smiling is unknown. Sixty-six preschool-age children (49 boys, M = 39.97 months, SD = 10.58) with suspected ASD (61 confirmed ASD) were administered the ADOS-2 and provided social affect calibrated severity scores (SA CSS). Children's social gaze and smiling during the ADOS-2, captured with a camera contained in eyeglasses worn by the examiner and parent, were obtained via a computer vision processing pipeline. Children who gazed more at their parents (p = .04) and whose gaze at their parents involved more smiling (p = .02) received lower social affect severity scores, indicating fewer social affect symptoms, adjusted R2 = .15, p = .003.

Objective quantification of homophily in children with and without disabilities in naturalistic contexts.

Homophily, the tendency for individuals to preferentially interact with others similar to themselves is typically documented via self-report and, for children, adult report. Few studies have investigated homophily directly using objective measures of social movement. We quantified homophily in children with developmental disabilities (DD) and typical development (TD) using objective measures of position/orientation in preschool inclusion classrooms, designed to promote interaction between these groups of children. Objective measurements were collected using ultra-wideband radio-frequency tracking to determine social approach and social contact, measures of social movement and interaction. Observations of 77 preschoolers (47 with DD, and 30 TD) were conducted in eight inclusion classrooms on a total of 26 days. We compared DD and TD groups with respect to how children approached and shared time in social contact with peers using mixed-effects models. Children in concordant dyads (DD-DD and TD-TD) both moved toward each other at higher velocities and spent greater time in social contact than discordant dyads (DD-TD), evidencing homophily. DD-DD dyads spent less time in social contact than TD-TD dyads but were comparable to TD-TD dyads in their social approach velocities. Children's preference for similar peers appears to be a pervasive feature of their naturalistic interactions.

United States Federal Emergency Management Agency regional clustering by disaster exposure: a new paradigm for disaster response.

The Federal Emergency Management Agency (FEMA) divides the United States (US) into ten standard regions to promote local partnerships and priorities. These divisions, while longstanding, do not adequately address known hazard risk as reflected in past federal disaster declarations. From FEMA's inception in 1979 until 2020, the OpenFEMA dataset reports 4127 natural disaster incidents declared by 53 distinct state-level jurisdictions, listed by disaster location, type, and year. An unsupervised spectral clustering (SC) algorithm was applied to group these jurisdictions into regions based on affinity scores assigned to each pair of jurisdictions accounting for both geographic proximity and historical disaster exposures. Reassigning jurisdictions to ten regions using the proposed SC algorithm resulted in an adjusted Rand index (ARI) of 0.43 when compared with the existing FEMA regional structure, indicating little similarity between the current FEMA regions and the clustering results. Reassigning instead into six regions substantially improved cluster quality with a maximized silhouette score of 0.42, compared to a score of 0.34 for ten regions. In clustering US jurisdictions not only by geographic proximity but also by the myriad hazards faced in relation to one another, this study demonstrates a novel method for FEMA regional allocation and design that may ultimately improve FEMA disaster specialization and response.

Simulating COVID19 transmission from observed movement.

Current models of COVID-19 transmission predict infection from reported or assumed interactions. Here we leverage high-resolution observations of interaction to simulate infectious processes. Ultra-Wide Radio Frequency Identification (RFID) systems were employed to track the real-time physical movements and directional orientation of children and their teachers in 4 preschool classes over a total of 34 observations. An agent-based transmission model combined observed interaction patterns (individual distance and orientation) with CDC-published risk guidelines to estimate the transmission impact of an infected patient zero attending class on the proportion of overall infections, the average transmission rate, and the time lag to the appearance of symptomatic individuals. These metrics highlighted the prophylactic role of decreased classroom density and teacher vaccinations. Reduction of classroom density to half capacity was associated with an 18.2% drop in overall infection proportion while teacher vaccination receipt was associated with a 25.3% drop. Simulation results of classroom transmission dynamics may inform public policy in the face of COVID-19 and similar infectious threats.

Deep Learning-Based Retinal Nerve Fiber Layer Thickness Measurement of Murine Eyes.

Purpose: To design a robust and automated estimation method for measuring the retinal nerve fiber layer (RNFL) thickness using spectral domain optical coherence tomography (SD-OCT). Methods: We developed a deep learning-based image segmentation network for automated segmentation of the RNFL in SD-OCT B-scans of mouse eyes. In total, 5500 SD-OCT B-scans (5200 B-scans were used as training data with the remaining 300 B-scans used as testing data) were used to develop this segmentation network. Postprocessing operations were then applied on the segmentation results to fill any discontinuities or remove any speckles in the RNFL. Subsequently, a three-dimensional retina thickness map was generated by z-stacking 100 segmentation processed thickness B-scan images together. Finally, the average absolute difference between algorithm predicted RNFL thickness compared to the ground truth manual human segmentation was calculated. Results: The proposed method achieves an average dice similarity coefficient of 0.929 in the SD-OCT segmentation task and an average absolute difference of 0.0009 mm in thickness estimation task on the basis of the testing dataset. We also evaluated our segmentation algorithm on another biological dataset with SD-OCT volumes for RNFL thickness after the optic nerve crush injury. Results were shown to be comparable between the predicted and manually measured retina thickness values. Conclusions: Experimental results demonstrate that our automated segmentation algorithm reliably predicts the RNFL thickness in SD-OCT volumes of mouse eyes compared to laborious and more subjective manual SD-OCT RNFL segmentation. Translational Relevance: Automated segmentation using a deep learning-based algorithm for murine eye OCT effectively and rapidly produced nerve fiber layer thicknesses comparable to manual segmentation.

Enhancing Multimedia Imbalanced Concept Detection Using VIMP in Random Forests.

Recent developments in social media and cloud storage lead to an exponential growth in the amount of multimedia data, which increases the complexity of managing, storing, indexing, and retrieving information from such big data. Many current content-based concept detection approaches lag from successfully bridging the semantic gap. To solve this problem, a multi-stage random forest framework is proposed to generate predictor variables based on multivariate regressions using variable importance (VIMP). By fine tuning the forests and significantly reducing the predictor variables, the concept detection scores are evaluated when the concept of interest is rare and imbalanced, i.e., having little collaboration with other high level concepts. Using classical multivariate statistics, estimating the value of one coordinate using other coordinates standardizes the covariates and it depends upon the variance of the correlations instead of the mean. Thus, conditional dependence on the data being normally distributed is eliminated. Experimental results demonstrate that the proposed framework outperforms those approaches in the comparison in terms of the Mean Average Precision (MAP) values.

Driver Missense Mutation Identification Using Feature Selection and Model Fusion.

Driver mutations propel oncogenesis and occur much less frequently than passenger mutations. The need for automatic and accurate identification of driver mutations has increased dramatically with the exponential growth of mutation data. Current computational solutions to identify driver mutations rely on sequence homology. Here we construct a machine learning-based framework that does not rely on sequence homology or domain knowledge to predict driver missense mutations. A windowing approach to represent the local environment of the sequence around the mutation point as a mutation sample is applied, followed by extraction of three sequence-level features from each sample. After selecting the most significant features, the support vector machine and multimodal fusion strategies are employed to give final predictions. The proposed framework achieves relatively high performance and outperforms current state-of-the-art algorithms. The ease of deploying the proposed framework and the relatively accurate performance make this solution applicable to large-scale mutation data analyses.

Knowledge-assisted sequential pattern analysis with heuristic parameter tuning for labor contraction prediction.

The optimal dosing regimen of remifentanil for relieving labor pain should achieve maximal efficacy during contractions and little effect between contractions. Toward such a need, we propose a knowledge-assisted sequential pattern analysis with heuristic parameter tuning to predict the changes in intrauterine pressure,which indicates the occurrence of labor contractions. This enables giving the drug shortly before each contraction starts. Asequential association rule mining based patient selection strategy is designed to dynamically select data for training regression models. A novel heuristic parameter tuning method is proposed to decide the appropriate value ranges and searching strategies for both the regularization factor and the Gaussian kernel parameter of leastsquares support vector machine with radial basis function (RBF) kernel, which is used as the regression model for time series prediction. The parameter tuning method utilizes information extracted from the training dataset, and it is adaptive to the characteristics of time series. The promising experimental results show that the proposed framework is able to achieve the lowest prediction errors as compared to some existing methods.

Wavelet analysis in current cancer genome research: a survey.

With the rapid development of next generation sequencing technology, the amount of biological sequence data of the cancer genome increases exponentially, which calls for efficient and effective algorithms that may identify patterns hidden underneath the raw data that may distinguish cancer Achilles' heels. From a signal processing point of view, biological units of information, including DNA and protein sequences, have been viewed as one-dimensional signals. Therefore, researchers have been applying signal processing techniques to mine the potentially significant patterns within these sequences. More specifically, in recent years, wavelet transforms have become an important mathematical analysis tool, with a wide and ever increasing range of applications. The versatility of wavelet analytic techniques has forged new interdisciplinary bounds by offering common solutions to apparently diverse problems and providing a new unifying perspective on problems of cancer genome research. In this paper, we provide a survey of how wavelet analysis has been applied to cancer bioinformatics questions. Specifically, we discuss several approaches of representing the biological sequence data numerically and methods of using wavelet analysis on the numerical sequences.

Prediction of uterine contractions using knowledge-assisted sequential pattern analysis.

The usage of the systemic opioid remifentanil in relieving the labor pain has attracted much attention recently. An optimal dosing regimen for administration of remifentanil during labor relies on anticipating the timing of uterine contractions. These predictions should be made early enough to maximize analgesia efficacy during contractions and minimize the impact of the medication between contractions. We have designed a knowledge-assisted sequential pattern analysis framework to 1) predict the intrauterine pressure in real time; 2) anticipate the next contraction; and 3) develop a sequential association rule mining approach to identify the patterns of the contractions from historical patient tracings (HT).

Rule mining and classification in a situation assessment application: a belief-theoretic approach for handling data imperfections.

Management of data imprecision and uncertainty has become increasingly important, especially in situation awareness and assessment applications where reliability of the decision-making process is critical (e.g., in military battlefields). These applications require the following: 1) an effective methodology for modeling data imperfections and 2) procedures for enabling knowledge discovery and quantifying and propagating partial or incomplete knowledge throughout the decision-making process. In this paper, using a Dempster-Shafer belief-theoretic relational database (DS-DB) that can conveniently represent a wider class of data imperfections, an association rule mining (ARM)-based classification algorithm possessing the desirable functionality is proposed. For this purpose, various ARM-related notions are revisited so that they could be applied in the presence of data imperfections. A data structure called belief itemset tree is used to efficiently extract frequent itemsets and generate association rules from the proposed DS-DB. This set of rules is used as the basis on which an unknown data record, whose attributes are represented via belief functions, is classified. These algorithms are validated on a simplified situation assessment scenario where sensor observations may have caused data imperfections in both attribute values and class labels.

Stochastic clustering for organizing distributed information sources.

The number of information sources and the volumes of data in these information sources have greatly increased, which may be attributed to the ever-increasing complexity of real-world applications. The enormous amount of information available in the information sources in a distributed information-providing environment has created a need to provide users with tools to effectively and efficiently navigate and retrieve information. Queries in such an environment often access information from multiple information sources. This may be attributed to navigational characteristics. Clusters provide a structure for organizing the large number of information sources for efficient browsing, searching, and retrieval. This paper presents a stochastically-based clustering mechanism, called the Markov model mediator (MMM), to group the information sources into a set of useful clusters. Each information source cluster groups those information sources that show similarities in their data access behavior. Information sources within the same cluster are expected to be able to provide most of the required information among themselves for user queries that are closely related with respect to a particular application. This can significantly improve system response time, query performance, and result in an overall improvement in decision support. Empirical studies on real databases are performed and the results demonstrate that our proposed mechanism leads to a better set of clusters in comparison with other clustering methods. This serves to illustrate the effectiveness of our proposed MMM mechanism.