A labeled spectral dataset with cassava disease occurrences using virus titre determination protocol.

In this work, we present a novel dataset composed of spectral data and images of cassava crops with and without diseases. Together with the description of the dataset, we describe the protocol to collect such data in a controlled environment and in an open field where pests are not controlled. Crop disease diagnosis has been done in the past through the analysis of plant images taken with a smartphone camera. However, in some cases, disease symptoms are not visible. Furthermore, for some cassava diseases, once symptoms have manifested on the aerial part of the plant, the root which is the edible part of the plant has been totally destroyed. The goal of collecting this multimodality of the crop disease is early intervention, following the hypothesis that diseased crops without visible symptoms can be detected using spectral information. We collected visible and near-infrared spectra captured from leaves infected with two common cassava diseases namely; Cassava Brown Streak Disease and Cassava Mosaic Disease, as well as from healthy plants. Together, we also captured leaf imagery data that corresponds to the spectral information. In our experiments, biochemical data is collected and taken as the ground truth. Finally, agricultural experts provided a disease score per plant leaf from 1 to 5, 1 representing healthy and 5 severely diseased. The process of disease monitoring and data collection took 19 and 15 consecutive weeks for screenhouse and open field, respectively, until disease symptoms were visibly seen by the human eye.

Application of Artificial Intelligence to the Monitoring of Medication Adherence for Tuberculosis Treatment in Africa: Algorithm Development and Validation.

Background: Artificial intelligence (AI) applications based on advanced deep learning methods in image recognition tasks can increase efficiency in the monitoring of medication adherence through automation. AI has sparsely been evaluated for the monitoring of medication adherence in clinical settings. However, AI has the potential to transform the way health care is delivered even in limited-resource settings such as Africa. Objective: We aimed to pilot the development of a deep learning model for simple binary classification and confirmation of proper medication adherence to enhance efficiency in the use of video monitoring of patients in tuberculosis treatment. Methods: We used a secondary data set of 861 video images of medication intake that were collected from consenting adult patients with tuberculosis in an institutional review board-approved study evaluating video-observed therapy in Uganda. The video images were processed through a series of steps to prepare them for use in a training model. First, we annotated videos using a specific protocol to eliminate those with poor quality. After the initial annotation step, 497 videos had sufficient quality for training the models. Among them, 405 were positive samples, whereas 92 were negative samples. With some preprocessing techniques, we obtained 160 frames with a size of 224 × 224 in each video. We used a deep learning framework that leveraged 4 convolutional neural networks models to extract visual features from the video frames and automatically perform binary classification of adherence or nonadherence. We evaluated the diagnostic properties of the different models using sensitivity, specificity, F1-score, and precision. The area under the curve (AUC) was used to assess the discriminative performance and the speed per video review as a metric for model efficiency. We conducted a 5-fold internal cross-validation to determine the diagnostic and discriminative performance of the models. We did not conduct external validation due to a lack of publicly available data sets with specific medication intake video frames. Results: Diagnostic properties and discriminative performance from internal cross-validation were moderate to high in the binary classification tasks with 4 selected automated deep learning models. The sensitivity ranged from 92.8 to 95.8%, specificity from 43.5 to 55.4%, F1-score from 0.91 to 0.92, precision from 88% to 90.1%, and AUC from 0.78 to 0.85. The 3D ResNet model had the highest precision, AUC, and speed. Conclusions: All 4 deep learning models showed comparable diagnostic properties and discriminative performance. The findings serve as a reasonable proof of concept to support the potential application of AI in the binary classification of video frames to predict medication adherence.

A land use regression model using machine learning and locally developed low cost particulate matter sensors in Uganda.

The application of land use regression (LUR) modeling for estimating air pollution exposure has been used only rarely in sub-Saharan Africa (SSA). This is generally due to a lack of air quality monitoring networks in the region. Low cost air quality sensors developed locally in sub-Saharan Africa presents a sustainable operating mechanism that may help generate the air monitoring data needed for exposure estimation of air pollution with LUR models. The primary objective of our study is to investigate whether a network of locally developed low-cost air quality sensors can be used in LUR modeling for accurately predicting monthly ambient fine particulate matter (PM2.5) air pollution in urban areas of central and eastern Uganda. Secondarily, we aimed to explore whether the application of machine learning (ML) can improve LUR predictions compared to ordinary least squares (OLS) regression. We used data for the entire year of 2020 from a network of 23 PM2.5 low-cost sensors located in urban municipalities of eastern and central Uganda. Between January 1, 2020 and December 31, 2020, these sensors collected highly time-resolved measurement data of PM2.5 air concentrations. We used monthly-averaged PM2.5 concentration data for LUR prediction modeling of monthly PM2.5 concentrations. We used eight different ML base-learner algorithms as well as ensemble modeling. We applied 5-fold cross validation (80% training/20% test random splits) to evaluate the models with resampling and Root mean squared error (RMSE). The relative explanatory power and accuracy of the ML algorithms were evaluated by comparing coefficient of determination (R2) and RMSE, using OLS as the reference approach. The overall average PM2.5 concentration during the study period was 52.22 µg/m3 (IQR: 38.11, 62.84 µg/m3)-well above World Health Organization PM2.5 ambient air guidelines. From the base-learner and ensemble models, RMSE and R2 values ranged between 7.65 µg/m3 - 16.85 µg/m3 and 0.24-0.84, respectively. Extreme gradient boosting (xgbTree) performed best out of the base learner algorithms (R2 = 0.84; RMSE = 7.65 µg/m3). Model performance from ensemble modeling with Lasso and Elastic-Net Regularized Generalized Linear Models (glmnet) did not outperform xgbTree, but prediction performance was comparable to that of xgbTree. The most important temporal and spatial predictors of monthly PM2.5 levels were monthly precipitation, percent of the population using solid fuels for cooking, distance to Lake Victoria, and greenspace (NDVI) within a 500-m buffer of air monitors. In conclusion, data from locally developed low-cost PM sensors provide evidence that they can be used for spatio-temporal prediction modeling of air pollution exposures in Uganda. Moreover, the non-parametric ML and ensemble approaches to LUR modeling clearly outperformed OLS regression algorithm for the prediction of monthly PM2.5 concentrations. Deploying low-cost air quality sensors in concert with implementation of data quality control measures, can help address the critical need for expanding and improving air quality monitoring in resource-constrained settings of sub-Saharan Africa. These low-cost sensors, in conjunction with non-parametric ML algorithms, may provide a rapid path forward for PM2.5 exposure assessment and to spur air pollution epidemiology research in the region.

Identifying patterns in urban housing density in developing countries using convolutional networks and satellite imagery.

The use of Deep Neural Networks for remote sensing scene image analysis is growing fast. Despite this, data sets on developing countries are conspicuously absent in the public domain for benchmarking machine learning algorithms, rendering existing data sets unrepresentative. Secondly, current literature uses low-level semantic scene image class definitions, which may not have many relevant applications in certain domains. To examine these problems, we applied Convolutional Neural Networks (CNN) to high-level scene image classification for identifying patterns in urban housing density in a developing country setting. An end-to-end model training workflow is proposed for this purpose. A method for quantifying spatial extent of urban housing classes which gives insight into settlement patterns is also proposed. The method consists of computing the ratio between area covered by a given housing class and total area occupied by all classes. In the current work this method is implemented based on grid count, whereby the number of predicted grids for one housing class is divided by the total grid count for all classes. Results from the proposed method were validated against building density data computed on OpenStreetMap data. Our results for scene image classification are comparable to current state-of-the-art, despite focusing only on most difficult classes in those works. We also contribute a new satellite scene image data set that captures some general characteristics of urban housing in developing countries. The data set has similar but also some distinct attributes to existing data sets.

A dataset of necrotized cassava root cross-section images.

Cassava brown streak disease is a major disease affecting cassava. Along with foliar chlorosis and stem lesions, a very common symptom of cassava brown streak disease is the development of a dry, brown corky rot within the starch bearing tuberous roots, also known as necrosis. This paper presents a dataset of curated image data of necrosis bearing roots across different cassava varieties. The dataset contains images of cassava root cross-sections based on trial harvests from Uganda and Tanzania. The images were taken using a smartphone camera. The resulting dataset consists of 10,052 images making this the largest publicly available dataset for crop root necrosis. The data is comprehensive and contains different variations of necrosis expression including root cross-section types, number of necrosis lesions, presentation of the necrosis lesions. The dataset is important and can be used to train machine learning models which quantify the percentage of cassava root damage caused by necrosis.

Design Choices for Automated Disease Surveillance in the Social Web.

The social web has emerged as a dominant information architecture accelerating technology innovation on an unprecedented scale. The utility of these developments to public health use cases like disease surveillance, information dissemination, outbreak prediction and so forth has been widely investigated and variously demonstrated in work spanning several published experimental studies and deployed systems. In this paper we provide an overview of automated disease surveillance efforts based on the social web characterized by their different high level design choices regarding functional aspects like user participation and language parsing approaches. We briefly discuss the technical rationale and practical implications of these different choices in addition to the key limitations associated with these systems within the context of operable disease surveillance. We hope this can offer some technical guidance to multi-disciplinary teams on how best to implement, interpret and evaluate disease surveillance programs based on the social web.

Correction: Evaluating Subcriticality during the Ebola Epidemic in West Africa.

[This corrects the article DOI: 10.1371/journal.pone.0140651.].

Evaluating Subcriticality during the Ebola Epidemic in West Africa.

The 2014-2015 Ebola outbreak is the largest and most widespread to date. In order to estimate ongoing transmission in the affected countries, we estimated the weekly average number of secondary cases caused by one individual infected with Ebola throughout the infectious period for each affected West African country using a stochastic hidden Markov model fitted to case data from the World Health Organization. If the average number of infections caused by one Ebola infection is less than 1.0, the epidemic is subcritical and cannot sustain itself. The epidemics in Liberia and Sierra Leone have approached subcriticality at some point during the epidemic; the epidemic in Guinea is ongoing with no evidence that it is subcritical. Response efforts to control the epidemic should continue in order to eliminate Ebola cases in West Africa.