Current artificial intelligence (AI) applications for the diagnosis of coronavirus disease 2019 (COVID-19) often lack a biological foundation in the decision-making process. In this study, we have employed AI for COVID-19 diagnosis using mass spectrometry (MS) data and leveraged explainable AI (X-AI) to explain the decision-making process on a local (per-sample) and global (all samples) basis. We first assessed eight machine learning models with five feature engineering techniques using a five-fold stratified cross-validation. The best accuracy was achieved by Random Forest (RF) classifier using the ratio of areas under the curve (AUC) from the MS data as features. These features were chosen on the basis of tentatively representing both human and viral proteins in human gargle samples. We evaluated the RF classifier on a 70%−30% train-test split strategy of 152 human gargle samples, yielding an accuracy of 94.12% on the test dataset. Employing X-AI, we further interpreted the RF model using shapely additive explanations (SHAP) and feature importance techniques, including permutation and impurity-based feature importances. With these interpretation models offering a local and global explanation for the machine learning model decisions, we devised a straightforward, three-stage X-AI framework that can enable medical practitioners to understand the mechanisms of a black-box AI model. To the medical practitioner, this instills trust in the AI model by providing the rationales for its decisions.

Towards providing personalized care, digital mental-wellness apps today ask questions to learn about subjects. However, not all subjects using these apps will have mood problems; thus, they do not need follow-up questions. In this study, we investigate an alternate mechanism to handle such non-sensitive posts (i.e., those not indicating mood problems) in college settings. To do so, we generate and use training data provided by a cohort of peer college students so that responses to non-sensitive posts are contextual, emotionally aware, and empathetic while also being terminal (not asking follow-up questions). Using data from a real mental-wellness app used by students, we identify that AI models trained with our peer-provided dataset generate desirable responses to non-sensitive posts, while models trained with state-of-the-art (Facebook’s) Empathetic Dataset yields responses that ask many follow-up questions, hence giving a perception of being intrusive. We believe that mental wellness apps today must not assume that any subject using these apps has mood problems. Perceptions of intrusiveness (i.e., apps asking many questions) must be a factor in design. We also believe that peer students can provide rich and reliable training datasets for college mental wellness apps, a topic that is not yet explored.

Worldwide 219 million people have been infected and 4.5 million have lost their lives in ongoing Covid-19 pandemic. Until vaccines became widely available, precautions and safety measures like wearing masks, physical distancing, avoiding face touching were some of the primary means to curb the spread of virus. Face touching is a compulsive human behavior that can not be prevented without constantly making a conscious effort, even then it is inevitable. To address this problem, we have designed a smartwatch-based solution, CovidAlert, that leverages Random Forest algorithm trained on accelerometer and gyroscope data from the smartwatch to detect hand transition to face and sends a quick haptic alert to the users. CovidAlert is highly energy efficient as it employs STA/LTA algorithm as a gatekeeper to curtail the usage of Random Forest model on the watch when user is inactive. The overall accuracy of system is 88.4% with low false negatives and false positives. We also demonstrated the system viability by implementing it on a commercial Fossil Gen 5 smartwatch.

Flooding is one of the most dangerous weather events today. Between $2015-2019$, on average, flooding has caused more than $130$ deaths every year in the USA alone. The devastating nature of flood necessitates the continuous monitoring of water level in the rivers and streams to detect the incoming flood. In this work, we have designed and implemented an efficient vision-based ensemble solution to continuously detect the water level in the creek. Our solution adapts template matching algorithm to find the region of interest by leveraging edge maps, and combines two parallel approach to identify the water level. While first approach fits a linear regression model in edge map to identify the water line, second approach uses a split sliding window to compute the sum of squared difference in pixel intensities to find the water surface. We evaluated the proposed system on $4306$ images collected between $3$rd October and $18$th December in 2019 with the frequency of $1$ image in every $10$ minutes. The system exhibited low error rate as it achieved $4.8$, $3.1\%$ and $0.92$ scores for MAE, MAPE and $R^2$ evaluation metrics, respectively. We believe the proposed solution is very practical as it is pervasive, accurate, doesn't require installation of any additional infrastructure in the water body and can be easily adapted to other locations.

In this paper, we report experimental results on assessing the impact of COVID-19 on college students by processing free-form texts generated by them. By free-form texts, we mean textual entries posted by college students (enrolled in a four year US college) via an app specifically designed to assess and improve their mental health. Using a dataset comprising of more than 9000 textual entries from 1451 students collected over four months (split between pre and post COVID-19), and established NLP techniques, a) we assess how topics of most interest to student change between pre and post COVID-19, and b) we assess the sentiments that students exhibit in each topic between pre and post COVID-19. Our analysis reveals that topics like Education became noticeably less important to students post COVID-19, while Health became much more trending. We also found that across all topics, negative sentiment among students post COVID-19 was much higher compared to pre-COVID-19. We expect our study to have an impact on policy-makers in higher education across several spectra, including college administrators, teachers, parents, and mental health counselors.

More than a year after the COVID-19 pandemic was declared, the need still exists for accurate, rapid, inexpensive and non-invasive diagnostic methods that yield high specificity and sensitivity towards the current and newly emerging SARS-CoV-2 strains. Compared to the nasopharyngeal swabs, several studies have established saliva as a more amenable specimen type for early detection of SARS-CoV-2. Considering the limitations and high demand for COVID-19 testing, we employed MALDI-ToF mass spectrometry in the analysis of 60 gargle samples from human donors and compared the resultant spectra against COVID-19 status. Several standards, including isolated human serum immunoglobulins, and controls, such as pre-COVID-19 saliva and heat inactivated SARS-CoV-2 virus, were simultaneously analyzed to provide a relative view of the saliva and viral proteome as they would appear in this workflow. Five potential biomarker peaks were established that demonstrated high concordance with COVID-19 positive individuals. Overall, the agreement of these results with RT-qPCR testing on nasopharyngeal swabs was ≥90% for the studied cohort, which consisted of young and largely asymptomatic student athletes. From a clinical standpoint, the results from this pilot study suggest that MALDI-ToF could be used to develop a relatively rapid and inexpensive COVID-19 assay.

The severe acute respiratory syndrome coronavirus type-2 (SARS-CoV-2) caused a global pandemic and immensely affected the global economy. Accurate, cost-effective, and quick tests have proven substantial in identifying infected people and mitigating the spread. Recently, multiple alternative platforms for testing coronavirus disease 2019 (COVID-19) have been published that show high agreement with current gold standard real-time polymerase chain reaction (RT-PCR) results. These new methods do away with nasopharyngeal (NP) swabs, eliminate the need for complicated reagents, and reduce the burden on RT-PCR test reagent supply. In the present work, we have designed an artificial intelligence-based (AI) testing method to provide confidence in the results. Current AI applications for COVID-19 studies often lack a biological foundation in the decision-making process, and our AI approach is one of the earliest to leverage explainable AI (X-AI) algorithms for COVID-19 diagnosis using mass spectrometry. Here, we have employed X-AI to explain the decision-making process on a local (per-sample) and global (all samples) basis underscored by biologically relevant features. We evaluated our technique with data extracted from human gargle samples and achieved a testing accuracy of 94.12 techniques would strengthen the relationship between AI and clinical diagnostics by providing biomedical researchers and healthcare workers with trustworthy and, most importantly, explainable test results

In recent years, there have been a surge in ubiquitous technologies such as smartwatches and fitness trackers that can track the human physical activities effortlessly. These devices have enabled common citizens to track their physical fitness and encourage them to lead a healthy lifestyle. Among various exercises, walking and running are the most common ones people do in everyday life, either through commute, exercise, or doing household chores. If done at the right intensity, walking and running are sufficient enough to help individual reach the fitness and weight-loss goals. Therefore, it is important to measure walking/ running speed to estimate the burned calories along with preventing them from the risk of soreness, injury, and burnout. Existing wearable technologies use GPS sensor to measure the speed which is highly energy inefficient and does not work well indoors. In this paper, we design, implement and evaluate a convolutional neural network based algorithm that leverages accelerometer and gyroscope sensory data from the wrist-worn device to detect the speed with high precision. Data from $15$ participants were collected while they were walking/running at different speeds on a treadmill. Our speed detection algorithm achieved $4.2\%$ and $9.8\%$ MAPE (Mean Absolute Error Percentage) value using $70-15-15$ train-test-evaluation split and leave-one-out cross-validation evaluation strategy respectively.

We design a framework based on Mask Region-based Convolutional Neural Network to automatically detect and separately extract anatomical components of mosquitoes-thorax, wings, abdomen and legs from images. Our training dataset consisted of 1500 smartphone images of nine mosquito species trapped in Florida. In the proposed technique, the first step is to detect anatomical components within a mosquito image. Then, we localize and classify the extracted anatomical components, while simultaneously adding a branch in the neural network architecture to segment pixels containing only the anatomical components. Evaluation results are favorable. To evaluate generality, we test our architecture trained only with mosquito images on bumblebee images. We again reveal favorable results, particularly in extracting wings. Our techniques in this paper have practical applications in public health, taxonomy and citizen-science efforts.

Among all animals, mosquitoes are responsible for the most deaths worldwide. Interestingly, not all types of mosquitoes spread diseases, but rather, a select few alone are competent enough to do so. In the case of any disease outbreak, an important first step is surveillance of vectors (i.e., those mosquitoes capable of spreading diseases). To do this today, public health workers lay several mosquito traps in the area of interest. Hundreds of mosquitoes will get trapped. Naturally, among these hundreds, taxonomists have to identify only the vectors to gauge their density. This process today is manual, requires complex expertise/ training, and is based on visual inspection of each trapped specimen under a microscope. It is long, stressful and self-limiting. This paper presents an innovative solution to this problem. Our technique assumes the presence of an embedded camera (similar to those in smart-phones) that can take pictures of trapped mosquitoes. Our techniques proposed here will then process these images to automatically classify the genus and species type. Our CNN model based on Inception-ResNet V2 and Transfer Learning yielded an overall accuracy of 80% in classifying mosquitoes when trained on 25, 867 images of 250 trapped mosquito vector specimens captured via many smart-phone cameras. In particular, the accuracy of our model in classifying Aedes aegypti and Anopheles stephensi mosquitoes (both of which are especially deadly vectors) is amongst the highest. We also present important lessons learned and practical impact of our techniques in this paper.

In this paper, we investigate the impact of age diversity on accuracy for activity recognition among women with wrist-worn wearables. Using a sample of 10 elder women and 10 younger women, and by monitoring five activities related to cardiac care (Running, Brisk Walking, Walking, Standing and Sitting), we show that while personalized models are best, activities classification based on age specific models are definitely superior in terms of accuracy compared to classification using mixed age models. We do so by a) extracting 11 features from inertial sensing data; b) reducing dimensionality using Linear Discriminant Analysis methods; c) quantifying variance among features using Principal Component Analysis; d) clustering activities; and finally e) comparing classification accuracies of all activities for personalized, age-specific and mixed-age models. We believe that our study is unique, and potentially important for superior healthcare for women, a demographic that is largely underserved today across the world.

In this paper, we explore the feasibility of leveraging the accelerometer and gyroscope sensors in modern smartphones to detect instances of distracted driving activities (e.g., calling, texting, and reading while driving). To do so, we conducted an experiment with 16 subjects on a realistic driving simulator. As discussed later, the simulator is equipped with a realistic steering wheel, acceleration/braking pedals, and a wide screen to visualize background vehicular traffic. It is also programmed to simulate multiple environmental conditions like daytime, nighttime, fog, and rain/snow. The subjects were instructed to drive the simulator while performing a randomized sequence of activities that included texting, calling, and reading from a phone while they were driving, during which the accelerometer and gyroscope in the phone were logging sensory data. By extracting features from this sensory data, we then implemented a machine learning technique based on random forests to detect distracted driving. Our technique achieves very good precision, recall, and F-measure across all environmental conditions we tested. We believe that our contributions in this paper can have a significant impact on enhancing road safety.

Mosquito borne diseases continue to pose grave dangers to global health. An important step in combating their spread in any area is to identify the type of species prevalent there. To do so, trained personnel lay local mosquito traps, and after collecting trapped specimens, they visually inspect each specimen to identify the species type and log their counts. This process takes hours and is cognitively very demanding. In this paper, we design a smart-phone based system that allows anyone to take images of a still mosquito that is either alive or dead (but still retaining its physical form) and automatically classifies the species type. Our system integrates image processing, feature selection, unsupervised clustering, and an SVM based machine learning algorithm for classification. Results with a total of 101 mosquito specimens spread across nine different vector carrying species (that were captured from a real outdoor trap in Tampa, Florida) demonstrate high accuracy in species identification. When implemented as a smart-phone application, the latency and energy consumption were minimal. With our system, the currently manual process of species identification and recording can be sped up, while also minimizing the ensuing cognitive workload of personnel. Secondly, ordinary citizens can use our system in their own homes for self-awareness and information sharing with public health agencies.

In a 2012 survey, in the United States alone, there were more than 35 000 reported suicides with approximately 1800 of being psychiatric inpatients. Recent Centers for Disease Control and Prevention (CDC) reports indicate an upward trend in these numbers. In psychiatric facilities, staff perform intermittent or continuous observation of patients manually in order to prevent such tragedies, but studies show that they are insufficient, and also consume staff time and resources. In this paper, we present the Watch-Dog system, to address the problem of detecting self-harming activities when attempted by in-patients in clinical settings. Watch-Dog comprises of three key components-Data sensed by tiny accelerometer sensors worn on wrists of subjects; an efficient algorithm to classify whether a user is active versus dormant (i.e., performing a physical activity versus not performing any activity); and a novel decision selection algorithm based on random forests and continuity indices for fine grained activity classification. With data acquired from 11 subjects performing a series of activities (both self-harming and otherwise), Watch-Dog achieves a classification accuracy of , , and for same-user 10-fold cross-validation, cross-user 10-fold cross-validation, and cross-user leave-one-out evaluation, respectively. We believe that the problem addressed in this paper is practical, important, and timely. We also believe that our proposed system is practically deployable, and related discussions are provided in this paper.

Chronic obstructive pulmonary disease (COPD) and congestive heart failure (CHF) are leading chronic health concerns among the aging population today. They are both typically characterized by episodes of cough that share similarities. In this paper, we design TussisWatch, a smart-phone-based system to record and process cough episodes for early identification of COPD or CHF. In our technique, for each cough episode, we do the following: 1) filter noise; 2) use domain expertise to partition each cough episode into multiple segments, indicative of disease or otherwise; 3) identify a limited number of audio features for each cough segment; 4) remove inherent biases as a result of sample size differences; and 5) design a two-level classification scheme, based on the idea of Random Forests, to process a recorded cough segment. Our classifier, at the first-level, identifies whether or not a given cough segment indicates a disease. If yes, the second-level classifier identifies the cough segment as symptomatic of COPD or CHF. Testing with a cohort of 9 COPD, 9 CHF, and 18 CONTROLS subjects spread across both the genders, races, and ages, our system achieves good performance in terms of Sensitivity, Specificity, Accuracy, and Area under ROC curve. The proposed system has the potential to aid early access to healthcare, and may be also used to educate patients on self-care at home.

In this paper, we attempt to detect instances of distracted driving using a wrist-worn wearable embedded with an accelerometer and a gyroscope. In our experimental study, 16 adult participants were asked to drive a driving simulator that is equipped with realistic driving conditions like brakes, accelerator, steering wheel and a large screen for road scene visualization. The simulator is also programmed for drivers to experience different environmental scenarios like day time, night time, fog and rain/ snow. While driving, participants engaged in a randomized sequence of calling, texting and reading from a phone while simultaneously driving. Throughout the experiment, each subject wore a wearable watch on the wrist which recorded the resulting acceleration and rotation of the wrist via an embedded accelerometer and gyroscope. Subsequently, we extracted a selected number of features from the sensory data, and designed machine learning techniques for detecting instances of distracted driving. Our performance evaluations reveal very good Precision, Recall, and F1-Scores. We believe that our paper introduces a new and potentially important application of wristworn wearables to enhance road safety.

Our society, unfortunately, is increasingly becoming exposed to explosion events that could have serious consequences. While explosion events like intentionally triggered bombs cause obvious harm to life and property, other explosions intended for benign purposes in quarries and construction zones may also cause unintended harm as a result of emanating seismic vibrations. As of today, detecting explosions, ranging them, and estimating their intensities are all accomplished only by seismometers that sense the associated ground vibrations and pressure changes as a result of their triggering. Unfortunately, seismometers are bulky, expensive and unsuitable for the ubiquitous use. In this paper, our broad motivation is to demonstrate the feasibility of leveraging the pervasive sensing and processing capabilities of modern smartphones to analyze explosion events. Within this context, we specifically address the problem of ranging and estimating the intensity of an explosion by leveraging the accelerometer and pressure sensors in the smartphone. To do so, we emplaced a number of smartphones in the vicinity of real explosion blasts conducted at a university mining laboratory, where the material blasted was Dynamite with Ammonium Nitrate Fuel Oil (ANFO). We then collected the corresponding accelerometer and pressure readings sensed by the phone. We extracted a number of novel features, and designed a machine learning based algorithmic framework for ranging and estimating the intensity of the explosion event. After an extensive validation, we find that the average-case error in ranging (i.e., estimating the distance to the source of the explosion event) and estimating the intensity of explosive material (in terms of its charge weight) are 8.24% and 7.37%, respectively. We also present perspectives on encoding our algorithm as a smartphone app, identify several critical challenges that will be encountered in real-time data processing of smartphone accelerometers and pressure sensors in the context of pervasive sensing of explosions, and also identify other practical issues like the diversity of smartphones. To the best of our knowledge, our work is pioneering in demonstrating the feasibility of using smartphones to analyze explosion events. We believe there are significant societal benefits emanating from our work.

Current state-of-the-art systems in the literature using wearables are not capable of distinguishing a large number of fine-grained and/or complex human activities, which may appear similar but with vital differences in context, such as lying on floor versus lying on bed versus lying on sofa. This paper fills the gap by proposing a novel system, called HuMAn, that recognizes and classifies complex at-home activities of humans with wearable sensing. Specifically, HuMAn makes such classifications feasible by leveraging selective multi-modal sensor suites from wearable devices, and enhances the richness of sensed information for activity classification by carefully leveraging placement of the wearable devices across multiple positions on the human body. The HuMAn system consists of the following components: (a) a practical feature set extraction method from selected multi-modal sensor suites; and (b) a novel two-level structured classification algorithm that improves accuracy by leveraging sensors in multiple body positions; and (c) improved refinement in classification of complex activities with minimal external infrastructure support (e.g., only a few Bluetooth beacons used for location context). The proposed system is evaluated with 10 users in real home environments. Experimental results demonstrate that the HuMAn system can detect 21 complex at-home activities with high degree of accuracy. For same-user evaluation strategy, the average activity classification accuracy is as high as 95 percent over all of the 21 activities. For the case of 10-fold cross-validation evaluation strategy, the average classification accuracy is 92 percent, and for the case of leave-one-out cross-validation strategy, the average classification accuracy is 75 percent.

Mosquito borne diseases have been amongst the most important healthcare concerns since time. An important component in combating the spread of infections in any geographic region of interest has been to identify the type of species that are prevalent in that region. As of today, dedicated personnel are assigned in most (if not all nations) to trap samples and identify them. Unfortunately, the process of identifying the actual species of mosquito is currently a manual process requiring highly trained personnel to visually inspect each specimen one by one under a microscope to make the identification. In this paper, we propose a system to automate this process. Specifically, we demonstrate results of an experiment we conducted where learning algorithms were designed to process images of captured mosquito samples taken via a smart-phone camera in order to identify the actual species. Using a total sample size of 60 images that included 7 species collected by the Hillsborough County Mosquito and Aquatic Weed Control Unit (in the city of Tampa) our proposed technique using Random Forests achieved an overall accuracy of 83:3% in correctly identifying the species of mosquito with good precision and recall. While our proposed technique will greatly benefit the state-of-the-art in species identification, we also believe that common citizens can also use our proposed system to improve existing mosquito control programs across the globe.

Plants are vital to the health of our biosphere, and effectively sustaining their growth is fundamental to the existence of life on this planet. A critical aspect, which decides the sustainability of plant growth is the quality of soil. All other things being fixed, the quality of soil greatly impacts the plant stress, which in turn impacts overall health. Although plant stress manifests in many ways, one of the clearest indicators are colors of the leaves. In this paper, we conducted an experimental study in a greenhouse for detecting plant stress caused by nutrient deficiencies in soil using smart-phone cameras, coupled with image processing and machine learning algorithms. The greenhouse experiment was conducted by growing two plant species; willows (Salix Pentandra) and poplars (Populus deltoides x nigra, DN34), in two treatments. These treatments included: unamended tailings (collected from a lead mine tailings pond and characterized by nutrient deficiency), and biosolids amended tailings. Biosolids are very rich in nutrients and were added to the tailings in one of the two treatments to supply plants with nutrients. Subsequently, we captured various images of plant leaves grown in both soils. Each image taken was pre-processed via filtration to remove associated noise, and was segmented into pixels to facilitate scalability of analysis. Subsequently, we designed random forests based algorithms to detect the stress of leaves as indicated by their coloring. In a dataset consisting of 34 leaves, our technique yields classifications with a high degree of prediction, recall and F1 score. Our work in this paper, while restricted to two types of plants and soils, can be generalized. We see applications in the emerging area of urban farming in terms of empowering citizens with tools and technologies for enhancing quality of farming practices.