Deep Learning Image Analysis of High-Throughput Toxicology Assay Images.

High-throughput chemical screening approaches often employ microscopy to capture photomicrographs from multi-well cell culture plates, generating thousands of images that require time-consuming human analysis. To automate this subjective and time-consuming manual process, we have developed a method that uses deep learning to automatically classify digital assay images. We have trained a convolutional neural network (CNN) to perform binary and multi-class classification. The binary classifier binned assay images into healthy (comparable to untreated controls) and altered (not comparable to untreated-control) classes with >98% accuracy; the multi-class classifier assigned "Healthy," "Intermediate" and "Altered" labels to assay images with >95% accuracy. Our dataset comprised high-resolution assay images from primary human hepatocytes and undifferentiated (proliferating) and differentiated 2D cultures of HepaRG cells. In this study we have focused on testing and fine-tuning various CNN architectures, including ResNet 34, 50 and 101. To visualize regions in the images that the CNN model used for classification, we employed Class Activation Maps (CAM). This allowed us to better understand the inner workings of the neural network and led to additional optimizations of the algorithm. The results indicate a strong correspondence between dosage and classifier-predicted scores, suggesting that these scores might be useful in further characterizing benchmark dose. Together, these results clearly demonstrate that deep-learning based automated image classification of cell morphology changes upon chemical-induced stress can yield highly accurate and reproducible assessments of cytotoxicity across a variety of cell types.

COVID-19 and beyond: Use of digital technology for pandemic response in Africa.

The use of technology has been ubiquitous in efforts to combat the ongoing COVID-19 pandemic. In this perspective, we review technologies and new approaches developed at the start of the pandemic; efforts earmarked by a flexible approach to problem solving, local tech entrepreneurship, and swift adoption of technology. We performed a systematic review of the use of technology during the initial wave of the COVID-19 pandemic in most African countries. We identified relevant articles by searching for mentions of technology, COVID-19, and specific country names. Articles were included if they specifically mentioned the use of technology or novel innovations in the response to the COVID-19 pandemic in an African country. The article search was conducted in August and included articles published between January and August 2020. We retrieved articles from journals, trusted news, government, and organization websites on Google, Google Scholar and PubMed. A total of 80 articles were retained and categorized under Disease Prevention (19 articles), Disease Surveillance xxx Antipoaching Tech Tracks COVID-19 Flare-Ups in South Africa - Scientific American. (2020, May 12), and Clinical Supplies and Management xxx Ethiopia's digital health response to COVID-19 - JSI. (2020, May 14). African nations used technology and innovative techniques to manage patients, monitor cases and disseminate information to counter the spread of COVID-19. The nature and outcomes of these efforts sometimes differed in Africa compared to other regions of the world due to its unique challenges and opportunities.

Mapping disparities in homicide trends across Brazil: 2000-2014.

BACKGROUND: Homicides are a major problem in Brazil. Drugs and arms trafficking, and land conflicts are three of the many factors driving homicide rates in Brazil. Understanding long-term spatiotemporal trends and social structural factors associated with homicides in Brazil would be useful for designing policies aimed at reducing homicide rates. METHODS: We obtained data from 2000 to 2014 from the Brazil Ministry of Health (MOH) Mortality Information System and sociodemographic data from the Brazil Institute of Geography and Statistics (IBGE). First, we quantified the rate of change in homicides at the municipality and state levels. Second, we used principal component regression and k-medoids clustering to examine differences in temporal trends across municipalities. Lastly, we used Bayesian hierarchical space-time models to describe spatio-temporal patterns and to assess the contribution of structural factors. RESULTS: There were significant variations in homicide rates across states and municipalities. We noted the largest decrease in homicide rates in the western and southeastern states of Sao Paulo, Rio de Janeiro and Espirito Santo, which coincided with an increase in homicide rates in the northeastern states of Ceará, Alagoas, Paraiba, Rio Grande Norte, Sergipe and Bahia during the fifteen-year period. The decrease in homicides in municipalities with populations of at least 250,000 coincided with an increase in municipalities with 25,000 people or less. Structural factors that predicted municipality-level homicide rates included crude domestic product, urbanization, border with neighboring countries and proportion of population aged fifteen to twenty-nine. CONCLUSIONS: Our findings support both a dissemination hypothesis and an interiorization hypothesis. These findings should be considered when designing interventions to curb homicide rates.

SWIFT-Active Screener: Accelerated document screening through active learning and integrated recall estimation.

BACKGROUND: In the screening phase of systematic review, researchers use detailed inclusion/exclusion criteria to decide whether each article in a set of candidate articles is relevant to the research question under consideration. A typical review may require screening thousands or tens of thousands of articles in and can utilize hundreds of person-hours of labor. METHODS: Here we introduce SWIFT-Active Screener, a web-based, collaborative systematic review software application, designed to reduce the overall screening burden required during this resource-intensive phase of the review process. To prioritize articles for review, SWIFT-Active Screener uses active learning, a type of machine learning that incorporates user feedback during screening. Meanwhile, a negative binomial model is employed to estimate the number of relevant articles remaining in the unscreened document list. Using a simulation involving 26 diverse systematic review datasets that were previously screened by reviewers, we evaluated both the document prioritization and recall estimation methods. RESULTS: On average, 95% of the relevant articles were identified after screening only 40% of the total reference list. In the 5 document sets with 5,000 or more references, 95% recall was achieved after screening only 34% of the available references, on average. Furthermore, the recall estimator we have proposed provides a useful, conservative estimate of the percentage of relevant documents identified during the screening process. CONCLUSION: SWIFT-Active Screener can result in significant time savings compared to traditional screening and the savings are increased for larger project sizes. Moreover, the integration of explicit recall estimation during screening solves an important challenge faced by all machine learning systems for document screening: when to stop screening a prioritized reference list. The software is currently available in the form of a multi-user, collaborative, online web application.

Detecting reports of unsafe foods in consumer product reviews.

OBJECTIVES: Access to safe and nutritious food is essential for good health. However, food can become unsafe due to contamination with pathogens, chemicals or toxins, or mislabeling of allergens. Illness resulting from the consumption of unsafe foods is a global health problem. Here, we develop a machine learning approach for detecting reports of unsafe food products in consumer product reviews from Amazon.com. MATERIALS AND METHODS: We linked Amazon.com food product reviews to Food and Drug Administration (FDA) food recalls from 2012 to 2014 using text matching approaches in a PostGres relational database. We applied machine learning methods and over- and under-sampling methods to the linked data to automate the detection of reports of unsafe food products. RESULTS: Our data consisted of 1 297 156 product reviews from Amazon.com. Only 5149 (0.4%) were linked to recalled food products. Bidirectional Encoder Representation from Transformations performed best in identifying unsafe food reviews, achieving an F1 score, precision and recall of 0.74, 0.78, and 0.71, respectively. We also identified synonyms for terms associated with FDA recalls in more than 20 000 reviews, most of which were associated with nonrecalled products. This might suggest that many more products should have been recalled or investigated. DISCUSSION AND CONCLUSION: Challenges to improving food safety include, urbanization which has led to a longer food chain, underreporting of illness and difficulty in linking contaminated food to illness. Our approach can improve food safety by enabling early identification of unsafe foods which can lead to timely recall thereby limiting the health and economic impact on the public.

Use of Deep Learning to Examine the Association of the Built Environment With Prevalence of Neighborhood Adult Obesity.

Importance: More than one-third of the adult population in the United States is obese. Obesity has been linked to factors such as genetics, diet, physical activity, and the environment. However, evidence indicating associations between the built environment and obesity has varied across studies and geographical contexts. Objective: To propose an approach for consistent measurement of the features of the built environment (ie, both natural and modified elements of the physical environment) and its association with obesity prevalence to allow for comparison across studies. Design: The cross-sectional study was conducted from February 14 through October 31, 2017. A convolutional neural network, a deep learning approach, was applied to approximately 150 000 high-resolution satellite images from Google Static Maps API (application programing interface) to extract features of the built environment in Los Angeles, California; Memphis, Tennessee; San Antonio, Texas; and Seattle (representing Seattle, Tacoma, and Bellevue), Washington. Data on adult obesity prevalence were obtained from the Centers for Disease Control and Prevention's 500 Cities project. Regression models were used to quantify the association between the features and obesity prevalence across census tracts. Main Outcomes and Measures: Model-estimated obesity prevalence (obesity defined as body mass index ≥30, calculated as weight in kilograms divided by height in meters squared) based on built environment information. Results: The study included 1695 census tracts in 6 cities. The age-adjusted obesity prevalence was 18.8% (95% CI, 18.6%-18.9%) for Bellevue, 22.4% (95% CI, 22.3%-22.5%) for Seattle, 30.8% (95% CI, 30.6%-31.0%) for Tacoma, 26.7% (95% CI, 26.7%-26.8%) for Los Angeles, 36.3% (95% CI, 36.2%-36.5%) for Memphis, and 32.9% (95% CI, 32.8%-32.9%) for San Antonio. Features of the built environment explained 64.8% (root mean square error [RMSE], 4.3) of the variation in obesity prevalence across all census tracts. Individually, the variation explained was 55.8% (RMSE, 3.2) for Seattle (213 census tracts), 56.1% (RMSE, 4.2) for Los Angeles (993 census tracts), 73.3% (RMSE, 4.5) for Memphis (178 census tracts), and 61.5% (RMSE, 3.5) for San Antonio (311 census tracts). Conclusions and Relevance: This study illustrates that convolutional neural networks can be used to automate the extraction of features of the built environment from satellite images for studying health indicators. Understanding the association between specific features of the built environment and obesity prevalence can lead to structural changes that could encourage physical activity and decreases in obesity prevalence.

Disparities in digital reporting of illness: A demographic and socioeconomic assessment.

Although digital reports of disease are currently used by public health officials for disease surveillance and decision making, little is known about environmental factors and compositional characteristics that may influence reporting patterns. The objective of this study is to quantify the association between climate, demographic and socio-economic factors on digital reporting of disease at the US county level. We reference approximately 1.5 million foodservice business reviews between 2004 and 2014, and use census data, machine learning methods and regression models to assess whether digital reporting of disease is associated with climate, socio-economic and demographic factors. The results show that reviews of foodservice businesses and digital reports of foodborne illness follow a clear seasonal pattern with higher reporting observed in the summer, when most foodborne outbreaks are reported and to a lesser extent in the winter months. Additionally, factors typically associated with affluence (such as, higher median income and fraction of the population with a bachelor's degrees) were positively correlated with foodborne illness reports. However, restaurants per capita and education were the most significant predictors of illness reporting at the US county level. These results suggest that well-known health disparities might also be reflected in the online environment. Although this is an observational study, it is an important step in understanding disparities in the online public health environment.

Social Media as a Sentinel for Disease Surveillance: What Does Sociodemographic Status Have to Do with It?

INTRODUCTION: Data from social media have been shown to have utility in augmenting traditional approaches to public health surveillance. Quantifying the representativeness of these data is needed for making accurate public health inferences. METHODS: We applied machine-learning methods to explore spatial and temporal dengue event reporting trends on Twitter relative to confirmed cases, and quantified associations with sociodemographic factors across three Brazilian states (São Paulo, Rio de Janeiro, and Minas Gerais) at the municipality level. RESULTS: Education and income were positive predictors of dengue reporting on Twitter. In contrast, municipalities with a higher percentage of older adults, and males were less likely to report suspected dengue disease on Twitter. Overall, municipalities with dengue disease tweets had higher mean per capita income and lower proportion of individuals with no primary school education. CONCLUSIONS: These observations highlight the need to understand population representation across locations, age, and racial/ethnic backgrounds in studies using social media data for public health research. Additional data is needed to assess and compare data representativeness across regions in Brazil.