Fine-Grained Intoxicated Gait Classification Using a Bilinear CNN.

Consuming excessive amounts of alcohol causes impaired mobility and judgment and driving accidents, resulting in more than 800 injuries and fatalities each day. Passive methods to detect intoxicated drivers beyond the safe driving limit can facilitate Just-In-Time alerts and reduce Driving Under the Influence (DUI) incidents. Popularly-owned smartphones are not only equipped with motion sensors (accelerometer and gyroscope) that can be employed for passively collecting gait (walk) data but also have the processing power to run computationally expensive machine learning models. In this paper, we advance the state-of-the-art by proposing a novel method that utilizes a Bi-linear Convolution Neural Network (BiCNN) for analyzing smartphone accelerometer and gyroscope data to determine whether a smartphone user is over the legal driving limit (0.08) from their gait. After segmenting the gait data into steps, we converted the smartphone motion sensor data to a Gramian Angular Field (GAF) image and then leveraged the BiCNN architecture for intoxication classification. Distinguishing GAF-encoded images of the gait of intoxicated vs. sober users is challenging as the differences between the classes (intoxicated vs. sober) are subtle, also known as a fine-grained image classification problem. The BiCNN neural network has previously produced state-of-the-art results on fine-grained image classification of natural images. To the best of our knowledge, our work is the first to innovatively utilize the BiCNN to classify GAF encoded images of smartphone gait data in order to detect intoxication. Prior work had explored using the BiCNN to classify natural images or explored other gait-related tasks but not intoxication Our complete intoxication classification pipeline consists of several important pre-processing steps carefully adapted to the BAC classification task, including step detection and segmentation, data normalization to account for inter-subject variability, data fusion, GAF image generation from time-series data, and a BiCNN classification model. In rigorous evaluation, our BiCNN model achieves an accuracy of 83.5%, outperforming the previous state-of-the-art and demonstrating the feasibility of our approach.

Domain Adaptation Methods for Lab-to-Field Human Context Recognition.

Human context recognition (HCR) using sensor data is a crucial task in Context-Aware (CA) applications in domains such as healthcare and security. Supervised machine learning HCR models are trained using smartphone HCR datasets that are scripted or gathered in-the-wild. Scripted datasets are most accurate because of their consistent visit patterns. Supervised machine learning HCR models perform well on scripted datasets but poorly on realistic data. In-the-wild datasets are more realistic, but cause HCR models to perform worse due to data imbalance, missing or incorrect labels, and a wide variety of phone placements and device types. Lab-to-field approaches learn a robust data representation from a scripted, high-fidelity dataset, which is then used for enhancing performance on a noisy, in-the-wild dataset with similar labels. This research introduces Triplet-based Domain Adaptation for Context REcognition (Triple-DARE), a lab-to-field neural network method that combines three unique loss functions to enhance intra-class compactness and inter-class separation within the embedding space of multi-labeled datasets: (1) domain alignment loss in order to learn domain-invariant embeddings; (2) classification loss to preserve task-discriminative features; and (3) joint fusion triplet loss. Rigorous evaluations showed that Triple-DARE achieved 6.3% and 4.5% higher F1-score and classification, respectively, than state-of-the-art HCR baselines and outperformed non-adaptive HCR models by 44.6% and 10.7%, respectively.

Search and Evaluation of Coevolving Problem and Solution Spaces in a Complex Healthcare Design Science Research Project.

This research employs design ethnography to study the design process of a design science research (DSR) project conducted over eight years. The DSR project focuses on chronic wounds and how Information Technology (IT) might support the management of those wounds. Since this is a new and complex problem not previously addressed by IT, it requires an exploration and discovery process. As such, we found that traditional DSR methodologies were not well-suited to guiding the design process. Instead we discovered that focusing on search, and in particular, the co-evolution of the problem and solution spaces, provides a much better focus for managing the DSR design process. The presentation of our findings from the ethnographic study includes a new representation for capturing the co-evolving problem/solution spaces, an illustration of the search process and co-evolving problem/solution spaces using the DSR project we studied, the need for changes in the purpose of DSR evaluation activities when using a search-focused design process, and how our proposed process extends and augments current DSR methodologies. Studying the DSR design process generates the knowledge that research project managers need for managing and guiding a DSR project, and contributes to our knowledge of the design process for research-oriented projects. Managerial Relevance Statement: From a managerial perspective, studying the design process provides the knowledge that research project managers need for managing and guiding DSR projects. In particular, research project managers can guide the search process by understanding when and why to explore different search spaces, to expand the solutions investigated, and to focus on promising solutions and to evaluate them. Overall, this research contributes to our knowledge of design and the design process, especially for highly research-oriented problems and solutions.

Chronic Wound Image Augmentation and Assessment Using Semi-Supervised Progressive Multi-Granularity EfficientNet.

Goal: Augment a small, imbalanced, wound dataset by using semi-supervised learning with a secondary dataset. Then utilize the augmented wound dataset for deep learning-based wound assessment. Methods: The clinically-validated Photographic Wound Assessment Tool (PWAT) scores eight wound attributes: Size, Depth, Necrotic Tissue Type, Necrotic Tissue Amount, Granulation Tissue type, Granulation Tissue Amount, Edges, Periulcer Skin Viability to comprehensively assess chronic wound images. A small corpus of 1639 wound images labeled with ground truth PWAT scores was used as reference. A Semi-Supervised learning and Progressive Multi-Granularity training mechanism were used to leverage a secondary corpus of 9870 unlabeled wound images. Wound scoring utilized the EfficientNet Convolutional Neural Network on the augmented wound corpus. Results: Our proposed Semi-Supervised PMG EfficientNet (SS-PMG-EfficientNet) approach estimated all 8 PWAT sub-scores with classification accuracies and F1 scores of about 90% on average, and outperformed a comprehensive list of baseline models and had a 7% improvement over the prior state-of-the-art (without data augmentation). We also demonstrate that synthetic wound image generation using Generative Adversarial Networks (GANs) did not improve wound assessment. Conclusions: Semi-supervised learning on unlabeled wound images in a secondary dataset achieved impressive performance for deep learning-based wound grading.

CovidRhythm: A Deep Learning Model for Passive Prediction of Covid-19 Using Biobehavioral Rhythms Derived From Wearable Physiological Data.

Goal: To investigate whether a deep learning model can detect Covid-19 from disruptions in the human body's physiological (heart rate) and rest-activity rhythms (rhythmic dysregulation) caused by the SARS-CoV-2 virus. Methods: We propose CovidRhythm, a novel Gated Recurrent Unit (GRU) Network with Multi-Head Self-Attention (MHSA) that combines sensor and rhythmic features extracted from heart rate and activity (steps) data gathered passively using consumer-grade smart wearable to predict Covid-19. A total of 39 features were extracted (standard deviation, mean, min/max/avg length of sedentary and active bouts) from wearable sensor data. Biobehavioral rhythms were modeled using nine parameters (mesor, amplitude, acrophase, and intra-daily variability). These features were then input to CovidRhythm for predicting Covid-19 in the incubation phase (one day before biological symptoms manifest). Results: A combination of sensor and biobehavioral rhythm features achieved the highest AUC-ROC of 0.79 [Sensitivity = 0.69, Specificity = 0.89, F[Formula: see text] = 0.76], outperforming prior approaches in discriminating Covid-positive patients from healthy controls using 24 hours of historical wearable physiological. Rhythmic features were the most predictive of Covid-19 infection when utilized either alone or in conjunction with sensor features. Sensor features predicted healthy subjects best. Circadian rest-activity rhythms that combine 24 h activity and sleep information were the most disrupted. Conclusions: CovidRhythm demonstrates that biobehavioral rhythms derived from consumer-grade wearable data can facilitate timely Covid-19 detection. To the best of our knowledge, our work is the first to detect Covid-19 using deep learning and biobehavioral rhythms features derived from consumer-grade wearable data.

User-centered preferences for a gait-informed alcohol intoxication app.

Background: mHealth technology can be used as a potential intervention for alcohol-related consequences. Applications designed to monitor alcohol use and relay information to the user may help to reduce risky behavior. Acceptability of such applications needs to be assessed. Methods: Survey data from 139 participants (29.8 years on average, 58% female) completing a single-session study for developing an application to detect blood alcohol concentration (BAC) from gait was analyzed to examine user preferences. Participants reported on their interest in an application for monitoring BAC from gait. Participants also reported on their preference for controlling features of the application. Acceptability and feasibility data were collected. Data were examined for the entire sample as well as differences in preference by age and gender were examined. Results: The majority of the sample indicated that they were interested in using an mHealth application to infer BAC from their gait. Users were interested in being able to control features of the application, such as monitoring BAC and reporting information to other individuals. Adults, as compared to emerging adults, preferred the ability to turn off the BAC-monitoring feature of the app. Females reported a preference for an app that does not allow the user to turn off notifications for BAC as well as safety features of the app. Conclusions: Results of the survey data indicate general interest in mHealth technology that monitors BAC from passive input. These results suggest that such an app may be accepted and used as an intervention for monitoring alcohol levels, which could mediate drinking and alcohol-related consequences.

Context-Aware Probabilistic Models for Predicting Future Sedentary Behaviors of Smartphone Users.

Sedentary behaviors are now prevalent as most modern jobs are done while seated. However, such sedentary behaviors have been found to increase the risk of several ailments including diabetes, cardiovascular disease, and all-cause mortality. Current interventions are mostly reactive and are triggered after the user has already been sedentary. Behavior change theory suggests that preventive sedentary interventions, which are triggered before a person becomes sedentary, are more likely to succeed. In this paper, we characterize user patterns of sedentary behaviors by analyzing smartphone-sensor data in a real-world dataset. Our work reveals location types (where), times of day/week (when), and smartphone contexts in which sedentary behaviors are most likely. Leveraging our findings, we then propose a set of context-aware probabilistic models that can predict sedentary behaviors in advance by analyzing smartphone sensor data. Our Context-Aware Predictive (CAP) models leverage smartphone-sensed contextual variables and the user's history of sedentary behaviors to predict their future sedentary behaviors. We rigorously analyze the performance of our models and discuss the implications of our work.

Continuous TBI Monitoring From Spontaneous Speech Using Parametrized Sinc Filters and a Cascading GRU.

Traumatic Brain Injury (TBI) is caused by a head injury that affects the brain, impairing cognitive and communication function and resulting in speech and language disorders. Over 80,000 individuals in the US suffer from long-term TBI disabilities and continuous monitoring after TBI is essential to facilitate rehabilitation and prevent regression. Prior work has demonstrated the feasibility of TBI monitoring from speech by leveraging advancements in Artificial Intelligence (AI) and speech processing technology. However, much of prior work explored TBI detection using scripted speech tasks such as diadochokinesis tests or reading a passage. Such scripted approaches require active user involvement that significantly burdens participants. Moreover, they are episodic, are not realistic, and do not provide a longitudinal picture of the user's TBI condition. This study proposes a continuous TBI monitoring from changes in acoustic features of spontaneous speech collected passively using the smartphone. Low-level acoustic features are extracted using parametrized Sinc filters (pSinc) that are then classified TBI (yes/no) using a cascading Gated Recurrent Unit (cGRU). The cGRU model utilizes a cell gate unit in the GRU to store and incorporate each individual's prediction history as prior knowledge into the model. In rigorous evaluation, our proposed method outperformed prior TBI classification methods on conversational speech recorded during patient-therapist discourses following TBI, achieving 83.87% balanced accuracy. Furthermore, unique words that are important in TBI prediction were identified using SHapley Additive exPlanations (SHAP). A correlation was also found between features acquired by the proposed method and coordination deficits following TBI.

Proximity Detection During Epidemics: Direct UWB TOA Versus Machine Learning Based RSSI.

In this paper, we compare the direct TOA-based UWB technology with the RSSI-based BLE technology using machine learning algorithms for proximity detection during epidemics in terms of complexity of implementation, availability in existing smart phones, and precision of the results. We establish the theoretical limits on the precision and confidence of proximity estimation for both technologies using the Cramer Rao Lower Bound (CRLB) and validate the theoretical foundations using empirical data gathered in diverse practical operating scenarios. We perform our empirical experiments at eight distances in three flat environments and one non-flat environment encompassing both Line of Sight (LOS) and Obstructed-LOS (OLOS) situations. We also analyze the effects of various postures (eight angles) of the person carrying the sensor, and four on-body locations of the sensor. To estimate the range with BLE RSSI, we use 14 features for training the Gradient Boosted Machines (GBM) learning algorithm and we compare the precision of results with those obtained from memoryless UWB TOA ranging algorithm. We show that the memoryless UWB TOA algorithm achieves 93.60% confidence, slightly outperforming the 92.85% confidence of the BLE RSSI with more complex GBM machine learning (ML) algorithm and the need for substantial training. The training process for the RSSI-based BLE social distance measurements involved 3000 measurements to create a training dataset for each scenario and post-processing of data to extract 14 features of RSSI, and the ML classification algorithm consumed 200 s of computational time. The memoryless UWB ranging algorithm achieves more robust results without any need for training in less than 0.5 s of computation time.

Diabetic Foot Ulcer Ischemia and Infection Classification Using EfficientNet Deep Learning Models.

Motivation: Infection (bacteria in the wound) and ischemia (insufficient blood supply) in Diabetic Foot Ulcers (DFUs) increase the risk of limb amputation. Goal: To develop an image-based DFU infection and ischemia detection system that uses deep learning. Methods: The DFU dataset was augmented using geometric and color image operations, after which binary infection and ischemia classification was done using the EfficientNet deep learning model and a comprehensive set of baselines. Results: The EfficientNets model achieved 99% accuracy in ischemia classification and 98% in infection classification, outperforming ResNet and Inception (87% accuracy) and Ensemble CNN, the prior state of the art (Classification accuracy of 90% for ischemia 73% for infection). EfficientNets also classified test images in a fraction (10% to 50%) of the time taken by baseline models. Conclusions: This work demonstrates that EfficientNets is a viable deep learning model for infection and ischemia classification.

Exploring Factors Associated With Mobile Phone Behaviors and Attitudes Toward Technology Among Adults With Alcohol Use Disorder and Implications for mHealth Interventions: Exploratory Study.

BACKGROUND: Alcohol use disorder (AUD) is associated with severe chronic medical conditions and premature mortality. Expanding the reach or access to effective evidence-based treatments to help persons with AUD is a public health objective. Mobile phone or smartphone technology has the potential to increase the dissemination of clinical and behavioral interventions (mobile health interventions) that increase the initiation and maintenance of sobriety among individuals with AUD. Studies about how this group uses their mobile phone and their attitudes toward technology may have meaningful implications for participant engagement with these interventions. OBJECTIVE: This exploratory study examined the potential relationships among demographic characteristics (race, gender, age, marital status, and income), substance use characteristics (frequency of alcohol and cannabis use), and clinical variables (anxiety and depression symptoms) with indicators of mobile phone use behaviors and attitudes toward technology. METHODS: A sample of 71 adults with AUD (mean age 42.9, SD 10.9 years) engaged in an alcohol partial hospitalization program completed 4 subscales from the Media Technology Usage and Attitudes assessment: Smartphone Usage measures various mobile phone behaviors and activities, Positive Attitudes and Negative Attitudes measure attitudes toward technology, and the Technological Anxiety/Dependence measure assesses level of anxiety when individuals are separated from their phone and dependence on this device. Participants also provided demographic information and completed the Epidemiologic Studies Depression Scale (CES-D) and the Generalized Anxiety Disorder (GAD-7) scale. Lastly, participants reported their frequency of alcohol use over the past 3 months using the Drug Use Frequency Scale. RESULTS: Results for the demographic factors showed a significant main effect for age, Smartphone Usage (P=.003; ηp2=0.14), and Positive Attitudes (P=.01; ηp2=0.07). Marital status (P=.03; ηp2=0.13) and income (P=.03; ηp2=0.14) were associated only with the Technological Anxiety and Dependence subscale. Moreover, a significant trend was found for alcohol use and the Technological Anxiety/Dependence subscale (P=.06; R2=0.02). Lastly, CES-D scores (P=.03; R2=0.08) and GAD symptoms (P=.004; R2=0.13) were significant predictors only of the Technological Anxiety/Dependence subscale. CONCLUSIONS: Findings indicate differences in mobile phone use patterns and attitudes toward technology across demographic, substance use, and clinical measures among patients with AUD. These results may help inform the development of future mHealth interventions among this population.

A Smartphone Physical Activity App for Patients in Alcohol Treatment: Single-Arm Feasibility Trial.

BACKGROUND: Alcohol use disorder (AUD) is a significant public health concern worldwide. Alcohol consumption is a leading cause of death in the United States and has a significant negative impact on individuals and society. Relapse following treatment is common, and adjunct intervention approaches to improve alcohol outcomes during early recovery continue to be critical. Interventions focused on increasing physical activity (PA) may improve AUD treatment outcomes. Given the ubiquity of smartphones and activity trackers, integrating this technology into a mobile app may be a feasible, acceptable, and scalable approach for increasing PA in individuals with AUD. OBJECTIVE: This study aims to test the Fit&Sober app developed for patients with AUD. The goals of the app were to facilitate self-monitoring of PA engagement and daily mood and alcohol cravings, increase awareness of immediate benefits of PA on mood and cravings, encourage setting and adjusting PA goals, provide resources and increase knowledge for increasing PA, and serve as a resource for alcohol relapse prevention strategies. METHODS: To preliminarily test the Fit&Sober app, we conducted an open pilot trial of patients with AUD in early recovery (N=22; 13/22, 59% women; mean age 43.6, SD 11.6 years). At the time of hospital admission, participants drank 72% of the days in the last 3 months, averaging 9 drinks per drinking day. The extent to which the Fit&Sober app was feasible and acceptable among patients with AUD during early recovery was examined. Changes in alcohol consumption, PA, anxiety, depression, alcohol craving, and quality of life were also examined after 12 weeks of app use. RESULTS: Participants reported high levels of satisfaction with the Fit&Sober app. App metadata suggested that participants were still using the app approximately 2.5 days per week by the end of the intervention. Pre-post analyses revealed small-to-moderate effects on increase in PA, from a mean of 5784 (SD 2511) steps per day at baseline to 7236 (SD 3130) steps per day at 12 weeks (Cohen d=0.35). Moderate-to-large effects were observed for increases in percentage of abstinent days (Cohen d=2.17) and quality of life (Cohen d=0.58) as well as decreases in anxiety (Cohen d=-0.71) and depression symptoms (Cohen d=-0.58). CONCLUSIONS: The Fit&Sober app is an acceptable and feasible approach for increasing PA in patients with AUD during early recovery. A future randomized controlled trial is necessary to determine the efficacy of the Fit&Sober app for long-term maintenance of PA, ancillary mental health, and alcohol outcomes. If the efficacy of the Fit&Sober app could be established, patients with AUD would have a valuable adjunct to traditional alcohol treatment that can be delivered in any setting and at any time, thereby improving the overall health and well-being of this population. TRIAL REGISTRATION: ClinicalTrials.gov NCT02958280; https://www.clinicaltrials.gov/ct2/show/NCT02958280.

Implementation of an antimicrobial stewardship programme in three regional hospitals in the south-east of Liberia: lessons learned.

Background: Antimicrobial stewardship (AMS) programmes can improve the use of antimicrobial agents. However, there is limited experience in the implementation of such programmes in low- and middle-income countries (LMICs). Objectives: To assess the effect of AMS measures in south-east Liberia on the quality of antimicrobial use in three regional hospitals. Methods: A bundle of three measures (local treatment guideline, training and regular AMS ward rounds) was implemented and quality indicators of antimicrobial use (i.e. correct compounds, dosage and duration) were assessed in a case series before and after AMS ward rounds. Primary endpoints were (i) adherence to the local treatment guideline; (ii) completeness of the microbiological diagnostics (according to the treatment guideline); and (iii) clinical outcome. The secondary endpoint was reduction in ceftriaxone use. Results: The majority of patients had skin and soft tissue infections (n = 108) followed by surgical site infections (n = 72), pneumonia (n = 64), urinary tract infection (n = 48) and meningitis (n = 18). After the AMS ward rounds, adherence to the local guideline improved for the selection of antimicrobial agents (from 34.5% to 61.0%, P < 0.0005), dosage (from 15.2% to 36.5%, P < 0.0005) and duration (from 13.2% to 31.0%, P < 0.0005). In total, 79.7% of patients (247/310) had samples sent for microbiological analysis. Overall, 92.3% of patients improved on Day 3 (286/310). The proportion of patients receiving ceftriaxone was significantly reduced after the AMS ward rounds from 51.3% to 14.2% (P < 0.0005). Conclusions: AMS measures can improve the quality of antimicrobial use in LMICs. However, long-term engagement is necessary to make AMS programmes in LMICs sustainable.

Machine Learning Estimation of COVID-19 Social Distance using Smartphone Sensor Data.

Airborne infectious diseases such as COVID-19 spread when healthy people are in close proximity to infected people. Technology-assisted methods to detect proximity in order to alert people are needed. In this work we systematically investigating Machine Learning (ML) methods to detect proximity by analyzing data gathered from smartphones' built-in Bluetooth, accelerometer and gyroscope sensors. We extracted 20 statistical features from raw sensor data, which were then classified (< 6ft or not) and regressed (distance estimate) using ML algorithms. We found that elliptical filtering of accelerometer and gyroscope sensors signal improved the performance of ML regression. The most predictive features were z-axis mean and fourth momentum for the accelerometer sensors, z-axis mean y-axis mean for the gyroscope sensor, and advertiser time and mean RSSI for Bluetooth radio. After rigorous evaluation of the performance of 19 ML classification and regression methods, we found that ensemble (boosted and bagged tree) methods and regression trees ML algorithms performed best when using data from a combination of Bluetooth radio, accelerometer and the gyroscope. We were able to classify proximity (< 6ft or not) with 100% accuracy using the accelerometer sensor and with 62%-97% accuracy with the Bluetooth radio.

Performance Evaluation of COVID-19 Proximity Detection Using Bluetooth LE Signal.

The risk of COVID-19 transmission increases when an uninfected person is less than 6 ft from an infected person for longer than 15 minutes. Infectious disease experts working on the COVID-19 pandemic call this high-risk situation being Too Close for Too Long (TCTL). Consequently, the problem of detecting the TCTL situation in order to maintain appropriate social distance has attracted considerable attention recently. One of the most prominent TCTL detection ideas being explored involves utilizing the Bluetooth Low-Energy (BLE) Received Signal Strength Indicator (RSSI) to determine whether the owners of two smartphones are observing the acceptable social distance of 6 ft. However, using RSSI measurements to detect the TCTL situation is extremely challenging due to the significant signal variance caused by multipath fading in indoor radio channel, carrying the smartphone in different pockets or positions, and differences in smartphone manufacturer and type of the device. In this study we utilize the Mitre Range Angle Structured (MRAS) Private Automated Contact Tracing (PACT) dataset to extensively evaluate the effectiveness of Machine Learning (ML) algorithms in comparison to classical estimation theory techniques to solve the TCTL problem. We provide a comparative performance evaluation of proximity classification accuracy and the corresponding confidence levels using classical estimation theory and a variety of ML algorithms. As the classical estimation method utilizes RSSI characteristics models, it is faster to compute, is more explainable, and drives an analytical solution for the precision bounds proximity estimation. The ML algorithms, Support Vector Machines (SVM), Random Forest, and Gradient Boosted Machines (GBM) utilized thirteen spatial, time-domain, frequency-domain, and statistical features extracted from the BLE RSSI data to generate the same results as classical estimation algorithms. We show that ML algorithms can achieve 3.60%~19.98% better precision, getting closer to achievable bounds for estimation.

Privacy-Preserving Deep Speaker Separation for Smartphone-Based Passive Speech Assessment.

Goal: Smartphones can be used to passively assess and monitor patients' speech impairments caused by ailments such as Parkinson's disease, Traumatic Brain Injury (TBI), Post-Traumatic Stress Disorder (PTSD) and neurodegenerative diseases such as Alzheimer's disease and dementia. However, passive audio recordings in natural settings often capture the speech of non-target speakers (cross-talk). Consequently, speaker separation, which identifies the target speakers' speech in audio recordings with two or more speakers' voices, is a crucial pre-processing step in such scenarios. Prior speech separation methods analyzed raw audio. However, in order to preserve speaker privacy, passively recorded smartphone audio and machine learning-based speech assessment are often performed on derived speech features such as Mel-Frequency Cepstral Coefficients (MFCCs). In this paper, we propose a novel Deep MFCC bAsed SpeaKer Separation (Deep-MASKS). Methods: Deep-MASKS uses an autoencoder to reconstruct MFCC components of an individual's speech from an i-vector, x-vector or d-vector representation of their speech learned during the enrollment period. Deep-MASKS utilizes a Deep Neural Network (DNN) for MFCC signal reconstructions, which yields a more accurate, higher-order function compared to prior work that utilized a mask. Unlike prior work that operates on utterances, Deep-MASKS operates on continuous audio recordings. Results: Deep-MASKS outperforms baselines, reducing the Mean Squared Error (MSE) of MFCC reconstruction by up to 44% and the number of additional bits required to represent clean speech entropy by 36%.

Comprehensive Assessment of Fine-Grained Wound Images Using a Patch-Based CNN With Context-Preserving Attention.

GOAL: Chronic wounds affect 6.5 million Americans. Wound assessment via algorithmic analysis of smartphone images has emerged as a viable option for remote assessment. METHODS: We comprehensively score wounds based on the clinically-validated Photographic Wound Assessment Tool (PWAT), which comprehensively assesses clinically important ranges of eight wound attributes: Size, Depth, Necrotic Tissue Type, Necrotic Tissue Amount, Granulation Tissue type, Granulation Tissue Amount, Edges, Periulcer Skin Viability. We proposed a DenseNet Convolutional Neural Network (CNN) framework with patch-based context-preserving attention to assess the 8 PWAT attributes of four wound types: diabetic ulcers, pressure ulcers, vascular ulcers and surgical wounds. RESULTS: In an evaluation on our dataset of 1639 wound images, our model estimated all 8 PWAT sub-scores with classification accuracies and F1 scores of over 80%. CONCLUSIONS: Our work is the first intelligent system that autonomously grades wounds comprehensively based on criteria in the PWAT rubric, alleviating the significant burden that manual wound grading imposes on wound care nurses.

Visual Analytics of Smartphone-Sensed Human Behavior and Health.

Smartphone health sensing tools, which analyze passively gathered human behavior data, can provide clinicians with a longitudinal view of their patients' ailments in natural settings. In this Visualization Viewpoints article, we postulate that interactive visual analytics (IVA) can assist data scientists during the development of such tools by facilitating the discovery and correction of wrong or missing user-provided ground-truth health annotations. IVA can also assist clinicians in making sense of their patients' behaviors by providing additional contextual and semantic information. We review the current state-of-the-art, outline unique challenges, and illustrate our viewpoints using our work as well as those of other researchers. Finally, we articulate open challenges in this exciting and emerging field of research.

Estimation of Blood Alcohol Concentration From Smartphone Gait Data Using Neural Networks.

Driving is a dynamic activity, which requires quick reflexes and decision making in order to respond to sudden changes in traffic conditions. Alcohol consumption impairs motor and cognitive skills, and causes many driving-related accidents annually. Passive methods of proactively detecting drivers who are too drunk to drive in order to notify them and prevent accidents, have recently been proposed. The effects of alcohol on a drinker's gait (walk) is a reliable indicator of their intoxication level. In this paper, we investigate detecting drinkers' intoxication levels from their gait by using neural networks to analyze sensor data gathered from their smartphone. Using data gathered from a large controlled alcohol study, we perform regression analysis using a Bi-directional Long Short Term Memory (Bi-LSTM) and Convolutional Neural Network (CNN) architectures to predict a person's Blood Alcohol Concentration (BAC) from their smartphone's accelerometer and gyroscope data. We innovatively proposed a comprehensive suite of pre-processing techniques and model-specific extensions to vanilla CNN and bi-LSTM models, which are well thought out and adapted specifically for BAC estimation. Our Bi-LSTM architecture achieves an RMSE of 0.0167 and the CNN architecture achieves an RMSE of 0.0168, outperforming state-of-the-art intoxication detection models using Bayesian Regularized Multilayer Perceptrons (MLP) (RMSE of 0.017) and the Random Forest (RF), with hand-crafted features. Moreover, our models learn features from raw sensor data, obviating the need for hand-crafted features, which is time consuming. Moreover, they achieve lower variance across folds and are hence more generalizable.

Applying Machine Learning Approaches to Suicide Prediction Using Healthcare Data: Overview and Future Directions.

Objective: Early identification of individuals who are at risk for suicide is crucial in supporting suicide prevention. Machine learning is emerging as a promising approach to support this objective. Machine learning is broadly defined as a set of mathematical models and computational algorithms designed to automatically learn complex patterns between predictors and outcomes from example data, without being explicitly programmed to do so. The model's performance continuously improves over time by learning from newly available data. Method: This concept paper explores how machine learning approaches applied to healthcare data obtained from electronic health records, including billing and claims data, can advance our ability to accurately predict future suicidal behavior. Results: We provide a general overview of machine learning concepts, summarize exemplar studies, describe continued challenges, and propose innovative research directions. Conclusion: Machine learning has potential for improving estimation of suicide risk, yet important challenges and opportunities remain. Further research can focus on incorporating evolving methods for addressing data imbalances, understanding factors that affect generalizability across samples and healthcare systems, expanding the richness of the data, leveraging newer machine learning approaches, and developing automatic learning systems.