Nabil Alshurafa

<sec> <title>BACKGROUND</title> Smoking and overeating are repetitive hand-to-mouth behaviors that contribute to highly prevalent yet preventable diseases. Most existing wearable systems have not been validated in free-living conditions to detect these behaviors in real-time. </sec> <sec> <title>OBJECTIVE</title> Leveraging shared behavioral patterns of eating and smoking, we developed HabitSense, a wearable system that integrates thermal sensors, a privacy conscious camera, and on-device algorithms, to detect smoking and eating events in real time and trigger a paired smartwatch to collect contextual data using ecological momentary assessment (EMA). We evaluated the detection accuracy of HabitSense in a free-living user study. </sec> <sec> <title>METHODS</title> Seventeen participants (9 in the smoking cohort and 8 in the eating cohort) were instructed to wear HabitSense, a custom necklace paired with a smartwatch, during waking hours for 7 consecutive days. Two separate machine-learned algorithms processed data from the thermal sensor array and camera on-device. When HabitSense predicted a smoking or eating event, the smartwatch prompted a micro- Ecological Momentary Assessment (micro-EMA) asking the participant to confirm or refute the prediction (“Are you smoking?” yes/no; “Are you eating?” yes/no). Additionally, an integrated camera recorded video to enable visual confirmation of each predicted smoking and eating event. </sec> <sec> <title>RESULTS</title> In total, 780.6 hours of sensor data were collected, capturing 217 smoking episodes and 87 eating episodes. The necklace generated 229 smoking-event predictions, of which 209 (91%) were true positives and 20 (9%) were false positives. 8 undetected smoking episodes were identified through manual review of the video footage (3% of total episodes). Participants responded to 212 EMA smoking-event prompts (92.6%); of these responses, 206 (97.2%) were correct (i.e., participants responded “yes” during actual smoking events and vice-versa). The necklace also generated 84 eating-event predictions, of which 67 (79.8%) were true positives and 17 (20.2%) were false positives. 20 undetected meals were identified in video footage (23% of total meals). </sec> <sec> <title>CONCLUSIONS</title> The findings suggest that the proposed system is feasible for automated and objective monitoring of contextual triggers associated with smoking relapse. HabitSense demonstrated high accuracy in smoking detection and strong response rates to smoking-triggered EMAs, supporting its potential for real-time behavioral assessment in free-living settings. For eating detection, the variability and complexity of food-related behaviors indicate that more advanced machine-learning approaches may be required, particularly for deployment on highly resource-constrained wearable devices. Future work will expand EMA queries to capture contextual factors surrounding smoking and eating episodes, leverage these data to develop just-in-time smartwatch-based interventions. Ultimately, this work aims to enable a personalized, adaptive intervention system that accounts for individual differences in behavior, a dimension often insufficiently addressed in current smoking cessation strategies. </sec>

OBJECTIVE: The objective of our 14-day technology-supported free-living study was to assess how psychological, environmental, and social factors affect overeating among participants with obesity. METHODS: ), who collectively logged 2004 meals, wore and used study devices for meal verification, and completed daily food recalls administered by dietitians. Participants reported on stress, affect, hunger, and meal contexts through Ecological Momentary Assessments (EMA). To explore the factors influencing caloric intake per meal, we employed a two-level mixed-effects location scale model, capturing both between-subject (BS) and within-subject (WS) factors based on the EMA data. This is a secondary analysis of the SenseWhy study, focusing on the association between stress and intake. RESULTS: Our analysis identified six BS factors (e.g., stress, perception of overeating, restaurant food, later meals, pleasure-seeking meal) and ten WS factors (e.g., biological hunger, perceived overeating, uncontrolled eating, social eating, restaurant food, snacks) to be significantly associated with caloric intake. Notably, participants who were more stressed, on average, consumed more calories (0.74; p = 0.002) with high consistency (-0.7; p = 0.048) between individuals. When stressed and not at home, participants consumed less calories (-0.62; p = 0.0043). CONCLUSION: Conventional strategies for managing stress-related overeating fall short. Effectively addressing overeating requires an understanding of both psychological and contextual factors.

Wearable cameras are increasingly used as an observational and interventional tool for human behaviors by providing detailed visual data of hand-related activities. This data can be leveraged to facilitate memory recall for logging of behavior or timely interventions aimed at improving health. However, continuous processing of RGB images from these cameras consumes significant power impacting battery lifetime, generates a large volume of unnecessary video data for post-processing, raises privacy concerns, and requires substantial computational resources for real-time analysis. We introduce THOR, a real-time adaptive spatio-temporal RGB frame sampling method that leverages thermal sensing to capture hand-object patches and classify them in real time. We use low-resolution thermal camera data to identify moments when a person switches from one hand-related activity to another and adjust the RGB frame sampling rate by increasing it during activity transitions and reducing it during periods of sustained activity (when the system has enough information to identify the activity). Additionally, we use the thermal cues from the hand to localize the region of interest (i.e. , the hand-object interaction) in each RGB frame, allowing the system to crop and process only the necessary part of the image for activity recognition. We develop a wearable device to validate our method through an in-the-wild study with 14 participants and over 30 activities, and further evaluate it on Ego4D (923 participants across 9 countries, totaling 3,670 hours of video). Our results show that using only 3% of the original RGB video data, our method captures all the activity segments, and achieves a hand-related activity recognition F1-score (95%) comparable to using the entire RGB video (94%). Our work provides a more practical path for the longitudinal use of wearable cameras to monitor hand-related activities and health-risk behaviors in real time.

Overeating is a key contributor to obesity, yet identifying and characterizing its underlying causes remains challenging. While prior research has leveraged Ecological Momentary Assessment (EMA) to capture psychological and contextual factors in real-time, few studies have integrated EMA with passive sensing to uncover fine-grained, individualized consumption behaviors. In this work, we present a multimodal framework combining psychological and contextual data from a custom-built EMA app with validated camera-derived meal microstructure features from a neck-worn activity-oriented wearable camera. Across 41 participants, the camera captured 6,343 hours of footage over 312 days, yielding annotated bites, chews, meal start/end times, and dietitian-confirmed caloric intake. Using supervised contrastive learning, we generated meal-level representations, projected them using UMAP, and applied k-means clustering to identify behavioral phenotypes. We then conducted a z-score analysis to highlight features most distinctive to each cluster. Among the eight discovered groups, three consistently showed high purity for overeating meals (average purity <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$=0.99$</tex>), revealing nuanced, data-driven overeating phenotypes that may inform targeted intervention strategies.

Continuous biomarker monitoring is revolutionizing chronic disease management, with glucose monitoring for diabetes as the primary example. Given the success of this approach, a transition to continuous protein monitoring (CPM, a real-time, implantable or wearable technology) could similarly advance precision medicine. In this work, we review state-of-the-art CPM platforms and their prospective clinical impact across both chronic disorders-metabolic, cardiovascular, autoimmune, and neurodegenerative-and acute crises, such as sepsis and transplant dysfunction. We also highlight remaining barriers to widespread adoption, including sensor stability, robust machine learning models for live interpretation, and responsible data handling for patient privacy. With continued engineering and clinical validation, emerging biosensor technologies could transform disease management, facilitating earlier interventions and individualizing treatment strategies, ultimately improving patient outcomes.

<sec> <title>BACKGROUND</title> Digital behaviors such as sleep, social interaction, and productivity reflect how individuals structure daily life. Among university students, online activity patterns mirror academic schedules, social rhythms, and lifestyle habits, with disruptions linked to sleep, stress, and well-being. Existing approaches—including wearables, apps, and surveys—yield useful insights but depend on self-report or active participation, limiting adherence in real-world use. Passive sensing of network traffic provides a scalable and less burdensome alternative, enabling unobtrusive capture of smartphone usage patterns while preserving privacy. </sec> <sec> <title>OBJECTIVE</title> This study evaluated whether encrypted smartphone network traffic, collected via a standard virtual private network (VPN), can be used to capture patterns of digital behavior. We assessed feasibility (sustained data capture) and acceptability (usability, burden, and privacy perceptions), and examined whether traffic-derived features reveal aspects of digital behavior—including timing, intensity, and regularity—relevant to health and daily functioning. </sec> <sec> <title>METHODS</title> We conducted a two-week prospective observational study at New York University. Thirty-eight students enrolled; 29 provided valid network data, 27 remained active for more than five days, and 25 completed the exit interview. Participants installed the WireGuard VPN client on personal smartphones, which enabled passive capture of encrypted network traffic. Feasibility was assessed across two domains: user retention and data coverage. Acceptability was evaluated using the System Usability Scale (SUS), NASA Task Load Index (NASA-TLX), and semi-structured exit interviews. Beyond evaluating feasibility and acceptability, we conducted exploratory analyses that visualized traffic-derived features in relation to digital activity patterns. </sec> <sec> <title>RESULTS</title> Of the 29 participants who contributed valid data, 27 (93%) remained active for more than five days. Mean data coverage was 74.1% (median 77.1%). Participants contributed an average of 311.6 hours of monitored traffic (~13 days, SD 3.5), with totals ranging from 121 to 496 hours. Usability ratings were high (mean SUS score = 78) and perceived workload low (NASA-TLX scores minimal). Participants described the system as easy to install, unobtrusive, and generally trustworthy, though some reported temporarily disabling the VPN during activities they considered private. No inferential statistical tests were conducted; analyses were descriptive. Exploratory analyses indicated that traffic-derived features reflected daily digital activity rhythms and revealed distinctive lifestyle patterns, including gaming and irregular late-night food delivery use. </sec> <sec> <title>CONCLUSIONS</title> VPN-based monitoring of encrypted smartphone traffic was feasible and acceptable, enabling sustained passive data collection with minimal burden. The findings demonstrate the potential of this approach as a scalable and device-agnostic method for digital phenotyping—capable of capturing fine-grained behavioral rhythms while preserving privacy. With broader validation and deployment, the technique could expand the toolkit for studying health, well-being, and cognitive function in everyday life. </sec> <sec> <title>CLINICALTRIAL</title> Not applicable. This study was not registered as a clinical trial because it did not involve randomization. </sec>

PURPOSE: While interventions to mitigate and prevent perinatal maternal distress exist, none are personalized based on participants' daily experiences and intervention response. This study compared maternal distress outcomes (depressive symptoms, anxiety symptoms, perceived stress) between perinatal individuals receiving a personalized mobile health-enhanced cognitive-behavioral intervention and individuals receiving usual prenatal care. METHODS: Pregnant individuals ≤ 22 weeks' gestation recruited from six prenatal care clinics were randomized to the intervention or usual prenatal care. Intervention participants received a 12-session adaptation of the Mothers and Babies intervention (MB-P), personalized by just-in-time stress reduction and mindfulness content based on elevated physiologic or self-reported stress. Primary outcomes were depressive and anxiety symptoms, and perceived stress. Secondary outcomes were behavioral activation, decentering of thoughts, social support, and mood regulation. Outcomes were measured at baseline, one-week post-intervention, one month postpartum, and three months postpartum. An intent-to-treat approach using mixed-effects models guided analysis. RESULTS: Forty-nine individuals were randomized to MB-P and fifty-one to usual prenatal care. Participants were 70% White, 33.7 years old on average, and 16.2 weeks gestation. At three months postpartum, intervention participants had lower depressive symptomatology (d = 0.43) and less perceived stress (d = 0.46) than controls. Intervention participants exhibited greater behavioral activation three months postpartum (d = 0.41), greater decentering post-intervention (d = 0.37), and greater mood regulation post-intervention (d = 0.56) and three months postpartum (d = 0.55). CONCLUSION: MB-P improved maternal depression and anxiety and mechanisms of behavioral activation, decentering, and mood regulation when compared to usual prenatal care. Future research should examine MB-P impact compared to standard MB without just-in-time content. TRIAL REGISTRATION: Clinical Trials.gov, NCT05052281.

Overeating contributes to obesity and poses a significant public health threat. The SenseWhy study (2018-2022) monitored 65 individuals with obesity in free-living settings, collecting 2302 meal-level observations (48 per participant), using an activity-oriented wearable camera, a mobile app, and dietitian-administered 24-hour dietary recalls. Micromovements (e.g., bites, chews) were manually labeled from 6343 hours of footage spanning 657 days. Psychological and contextual information was gathered before and after meals through Ecological Momentary Assessments (EMAs). We predicted overeating episodes based on EMA-derived features and passive sensing data (mean AUROC = 0.86; mean AUPRC = 0.84). Using semi-supervised learning on EMA-derived features alone, we identified five distinct overeating phenotypes: "Take-out Feasting," "Evening Restaurant Reveling," "Evening Craving," "Uncontrolled Pleasure Eating," and "Stress-driven Evening Nibbling." These results highlight the complex interplay between behavioral, psychological, and contextual factors associated with overeating, providing a foundation for personalized interventions.

Wearable cameras are increasingly used as an observational and interventional tool for human behaviors by providing detailed visual data of hand-related activities. This data can be leveraged to facilitate memory recall for logging of behavior or timely interventions aimed at improving health. However, continuous processing of RGB images from these cameras consumes significant power impacting battery lifetime, generates a large volume of unnecessary video data for post-processing, raises privacy concerns, and requires substantial computational resources for real-time analysis. We introduce THOR, a real-time adaptive spatio-temporal RGB frame sampling method that leverages thermal sensing to capture hand-object patches and classify them in real-time. We use low-resolution thermal camera data to identify moments when a person switches from one hand-related activity to another, and adjust the RGB frame sampling rate by increasing it during activity transitions and reducing it during periods of sustained activity. Additionally, we use the thermal cues from the hand to localize the region of interest (i.e., the hand-object interaction) in each RGB frame, allowing the system to crop and process only the necessary part of the image for activity recognition. We develop a wearable device to validate our method through an in-the-wild study with 14 participants and over 30 activities, and further evaluate it on Ego4D (923 participants across 9 countries, totaling 3,670 hours of video). Our results show that using only 3% of the original RGB video data, our method captures all the activity segments, and achieves hand-related activity recognition F1-score (95%) comparable to using the entire RGB video (94%). Our work provides a more practical path for the longitudinal use of wearable cameras to monitor hand-related activities and health-risk behaviors in real time.

Objectives: Food contamination is receiving heightened attention, but few comprehensive studies describe heavy metal content in foods or dietary exposure risks to heavy metals among vulnerable groups.We analyzed arsenic (As), cadmium (Cd), lead (Pb), and minerals in a market basket of foods from Montevideo, Uruguay and estimated daily dietary exposure in school children.Methods: From two 24-hr recalls for 862 low-average income children aged 7 we identified commonly consumed foods and beverages (~90% of the diet by frequency).We purchased fresh, frozen or ultra processed foods in several categories: milk, dairy & eggs (12 items), fruits (10), vegetables ( 14), breads & rolls (4), pastries & crackers (6), desserts & sweets (12), meats & cold cuts (19), fats & oils (2), cereals & legumes ( 9), frozen or instant foods (13), sauces & condiments (6).Multiple brands of the same food were pooled.Three samples were taken from each pooled item; digested with nitric and perchloric acid; analyzed via ICP-AES.Metal & mineral values were multiplied by reported food consumption to estimate dietary intake levels.Results: Preliminary findings suggest that As (< LOQ-1.6 g/ g), Cd (< LOQ-0.3 g/g), and Pb (< LOQ-355 g/g) in foods were generally low.Estimated dietary As, Cd, and Pb exposure ranged 0-194, 0.04-6.92,and 0.05-24.1 g/day, respectively.For lead >50% of children had dietary intake above current FDA guideline of 2.2 g/day.These diets provided 0.03-7.0,0.02-8.6,and 0.44-735.4mg/day of Fe, Zn and Ca, respectively.Conclusions: Food contamination with Pb and As is a potential concern in this population.Still, the same foods provide 3-4 orders of magnitude higher levels of minerals essential for growth and development compared to heavy metals.

Health-risk behaviors such as overeating and smoking have a profound impact on public health, making their monitoring and mitigation critical. Wearable RGB-Thermal cameras are being employed to monitor these behaviors by capturing hand-to-mouth (HTM) gestures, which are central to them. However, detection models relying on single modalities—either RGB or thermal—often struggle to accurately distinguish these confounding gestures due to inherent sensor limitations, such as sensitivity to lighting conditions or thermal occlusions. We present a family of fusion models that integrate RGB and thermal video data using early-, decision- , and a novel mid-fusion architecture, RGB-Thermal Fusion Video Network (RTFVNet), designed to enhance the recognition of HTM gestures associated with eating and smoking. Our evaluation shows that while decision fusion achieves the highest F1-score of 88% (0.44 TFLOPs), RTFVNet offers an optimal balance between performance (85%) and complexity (0.37 TFLOPs) for gesture classification of eating, smoking, and non-gesture activities.

BACKGROUND: Digital interventions can leverage user data to predict their health behavior, which can improve users' ability to make behavioral changes. Presenting predictions (e.g. how much a user might improve on an outcome) can be nuanced considering their uncertainty. Incorporating predictions raises design-related questions, such as how to present prediction data in a concise and actionable manner. PURPOSE: We conducted co-design sessions with end-users of a digital binge-eating intervention to learn how users would engage with prediction data and inform how to present these data visually. We additionally sought to understand how prediction intervals would help users understand uncertainty in these predictions and how users would perceive their actual progress relative to their prediction. METHODS: We conducted interviews with 22 adults with recurrent binge eating and obesity. We showed prototypes of hypothetical prediction displays for 5 evidence-based behavior change strategies, with the predicted success of each strategy for reducing binge eating in the week ahead (e.g. selecting to work on self-image this week might lead to 4 fewer binges while mood might lead to 1 fewer). We used thematic analysis to analyze data and generate themes. RESULTS: Users welcomed using prediction data, but wanted to maintain their autonomy and minimize negative feelings if they do not achieve their predictions. Although preferences varied, users generally preferred designs that were simple and helped them quickly compare prediction data across strategies. CONCLUSIONS: Predictions should be presented in efficient, organized layouts and with encouragement. Future studies should empirically validate findings in practice. CLINICAL TRIAL INFORMATION: The Clinical Trials Registration #: NCT06349460.

Stepped-care obesity treatments aim to improve efficiency by early identification of non-responders and adjusting interventions but lack validated models. We trained a random forest classifier to improve the predictive utility of a clinical decision rule (>0.5 lb weight loss/week) that identifies non-responders in the first 2 weeks of a stepped-care weight loss trial (SMART). From 2009 to 2021, 1058 individuals with obesity participated in three studies: SMART, Opt-IN, and ENGAGED. The model was trained on 80% of the SMART data (224 participants), and its in-distribution generalizability was tested on the remaining 20% (remaining 57 participants). The out-of-distribution generalizability was tested on the ENGAGED and Opt-IN studies (472 participants). The model predicted weight loss at month 6 with an 84.5% AUROC and an 86.3% AUPRC. SHAP identified predictive features: weight loss at week 2, ranges/means and ranges of weight loss, slope, and age. The SMART-trained model showed generalizable performance with no substantial difference across studies.

Wearable camera and thermal sensing systems are increasingly used for real-time eating detection and timely notifications to remind users to log their meals. However, confounding gestures such as irrelevant hand movements can cause false device confirmations of eating in real-time. Delaying the device confirmation of an eating episode, until the system is certain, can improve accuracy of eating detection, but prevents the capture of shorter bouts of eating. Balancing the trade-off between errors and detection delay is key to developing effective methods that provide immediate user feedback. This paper presents a real-time, hand-object-based method for automated detection of eating and drinking gestures and identifies the minimum number of gestures needed to reliably detect an eating episode. Unlike prior work, our method considers both hand motion and the object-in-hand and uses a low-power thermal sensor to reduce false positives. We evaluated our method on 36 participants, 28 of whom wore a wearable camera for up to 14 days in free-living environments. The results show that eating episodes can be accurately detected using 10 gestures or within the first 1.5 minutes of the eating episode, achieving an F1-score of 89.0%. Our findings provide evaluation guidelines for designing real-time intervention systems to address problematic eating behaviors.

Skin cancer, particularly melanoma, is a major health concern due to rising incidence rates, largely driven by ultraviolet (UV) radiation overexposure, making it essential to monitor and manage sun exposure effectively. While existing wearable UV sensors track exposure, they often rely on external power sources, limiting their battery lifetime. This study presents a self-sustaining wearable UV sensor that integrates solar energy harvesting, enabling continuous monitoring without need for frequent recharging. The device uses low-power components to measure UVA and UVB radiation with high accuracy. It is powered by a solar panel made from Ethylene Tetrafluoroethylene (ETFE), which provides continuous energy to recharge a LiPo battery. It transmits data via BLE for real-time feedback and can be used for personalized sun protection recommendations. A usability study with 10 participants demonstrated the sensor's effectiveness in raising UV awareness and encouraging sun protection habits.

Wearable cameras provide an objective method to visually confirm and automate the detection of health-risk behaviors such as smoking and overeating, which is critical for developing and testing adaptive treatment interventions. Despite the potential of wearable camera systems, adoption is hindered by inadequate clinician input in the design, user privacy concerns, and user burden. To address these barriers, we introduced HabitSense, an open-source, multi-modal neck-worn platform developed with input from focus groups with clinicians (N=36) and user feedback from in-wild studies involving 105 participants over 35 days. Optimized for monitoring health-risk behaviors, the platform utilizes RGB, thermal, and inertial measurement unit sensors to detect eating and smoking events in real time. In a 7-day study involving 15 participants, HabitSense recorded 768 hours of footage, capturing 420.91 minutes of hand-to-mouth gestures associated with eating and smoking data crucial for training machine learning models, achieving a 92% F1-score in gesture recognition. To address privacy concerns, the platform records only during likely health-risk behavior events using SECURE, a smart activation algorithm. Additionally, HabitSense employs on-device obfuscation algorithms that selectively obfuscate the background during recording, maintaining individual privacy while leaving gestures related to health-risk behaviors unobfuscated. Our implementation of SECURE has resulted in a 48% reduction in storage needs and a 30% increase in battery life. This paper highlights the critical roles of clinician feedback, extensive field testing, and privacy-enhancing algorithms in developing an unobtrusive, lightweight, and reproducible wearable system that is both feasible and acceptable for monitoring health-risk behaviors in real-world settings.

Human studies often rely on wearable lifelogging cameras that capture videos of individuals and their surroundings to aid in visual confirmation or recollection of daily activities like eating, drinking, and smoking. However, this may include private or sensitive information that may cause some users to refrain from using such monitoring devices. Also, short battery lifetime and large form factors reduce applicability for long-term capture of human activity. Solving this triad of interconnected problems is challenging due to wearable embedded systems’ energy, memory, and computing constraints. Inspired by this critical use case and the unique design problem, we developed NIR-sighted, an architecture for wearable video cameras that navigates this design space via three key ideas: (i) reduce storage and enhance privacy by discarding masked pixels and frames, (ii) enable programmers to generate effective masks with low computational overhead, and (iii) enable the use of small MCUs by moving masking and compression off-chip. Combined together in an end-to-end system, NIR-sighted’s masking capabilities and off-chip compression hardware shrinks systems, stores less data, and enables programmer-defined obfuscation to yield privacy enhancement. The user’s privacy is enhanced significantly as nowhere in the pipeline is any part of the image stored before it is obfuscated. We design a wearable camera called NIR-sightedCam based on this architecture; it is compact and can record IR and grayscale video at 16 and 20+ fps, respectively, for 26 hours nonstop (59 hours with IR disabled) at a fraction of comparable platforms power draw. NIR-sightedCam includes a low-power Field Programmable Gate Array that implements our mJPEG compress/obfuscate hardware, Blindspot. We additionally show the potential for privacy-enhancing function and clinical utility via an in-lab eating study, validated by a nutritionist.

Social presence has been known to impact eating behavior among people with obesity; however, the dual study of eating behavior and social presence in real-world settings is challenging due to the inability to reliably confirm the co-occurrence of these important factors. High-resolution video cameras can detect timing while providing visual confirmation of behavior; however, their potential to capture all-day behavior is limited by short battery lifetime and lack of autonomy in detection. Low-resolution infrared (IR) sensors have shown promise in automating human behavior detection; however, it is unknown if IR sensors contribute to behavior detection when combined with RGB cameras. To address these challenges, we designed and deployed a low-power, and low-resolution RGB video camera, in conjunction with a low-resolution IR sensor, to test a learned model's ability to detect eating and social presence. We evaluated our system in the wild with 10 participants with obesity; our models displayed slight improvement when detecting eating (5%) and significant improvement when detecting social presence (44%) compared with using a video-only approach. We analyzed device failure scenarios and their implications for future wearable camera design and machine learning pipelines. Lastly, we provide guidance for future studies using low-cost RGB and IR sensors to validate human behavior with context.

Automated detection of intake gestures with wearable sensors has been a critical area of research for advancing our understanding and ability to intervene in people's eating behavior. Numerous algorithms have been developed and evaluated in terms of accuracy. However, ensuring the system is not only accurate in making predictions but also efficient in doing so is critical for real-world deployment. Despite the growing research on accurate detection of intake gestures using wearables, many of these algorithms are often energy inefficient, impeding on-device deployment for continuous and real-time monitoring of diet. This article presents a template-based optimized multicenter classifier that enables accurate intake gesture detection while maintaining low-inference time and energy consumption using a wrist-worn accelerometer and gyroscope. We designed an Intake Gesture Counter smartphone application (CountING) and validated the practicality of our algorithm against seven state-of-the-art approaches on three public datasets (In-lab FIC, Clemson, and OREBA). Compared with other methods, we achieved optimal accuracy (81.60% F1 score) and very low inference time (15.97 msec per 2.20-sec data sample) on the Clemson dataset, and among the top performing algorithms, we achieve comparable accuracy (83.0% F1 score compared with 85.6% in the top performing algorithm) but superior inference time (13.8x faster, 33.14 msec per 2.20-sec data sample) on the In-lab FIC dataset and comparable accuracy (83.40% F1 score compared with 88.10% in the top-performing algorithm) but superior inference time (33.9x faster, 16.71 msec inference time per 2.20-sec data sample) on the OREBA dataset. On average, our approach achieved a 25-hour battery lifetime (44% to 52% improvement over state-of-the-art approaches) when tested on a commercial smartwatch for continuous real-time detection. Our approach demonstrates an effective and efficient method, enabling real-time intake gesture detection using wrist-worn devices in longitudinal studies.

BACKGROUND: Predicting the likelihood of success of weight loss interventions using machine learning (ML) models may enhance intervention effectiveness by enabling timely and dynamic modification of intervention components for nonresponders to treatment. However, a lack of understanding and trust in these ML models impacts adoption among weight management experts. Recent advances in the field of explainable artificial intelligence enable the interpretation of ML models, yet it is unknown whether they enhance model understanding, trust, and adoption among weight management experts. OBJECTIVE: This study aimed to build and evaluate an ML model that can predict 6-month weight loss success (ie, ≥7% weight loss) from 5 engagement and diet-related features collected over the initial 2 weeks of an intervention, to assess whether providing ML-based explanations increases weight management experts' agreement with ML model predictions, and to inform factors that influence the understanding and trust of ML models to advance explainability in early prediction of weight loss among weight management experts. METHODS: We trained an ML model using the random forest (RF) algorithm and data from a 6-month weight loss intervention (N=419). We leveraged findings from existing explainability metrics to develop Prime Implicant Maintenance of Outcome (PRIMO), an interactive tool to understand predictions made by the RF model. We asked 14 weight management experts to predict hypothetical participants' weight loss success before and after using PRIMO. We compared PRIMO with 2 other explainability methods, one based on feature ranking and the other based on conditional probability. We used generalized linear mixed-effects models to evaluate participants' agreement with ML predictions and conducted likelihood ratio tests to examine the relationship between explainability methods and outcomes for nested models. We conducted guided interviews and thematic analysis to study the impact of our tool on experts' understanding and trust in the model. RESULTS: =7.9; P=.02) compared with the other 2 methods with odds ratios of 2.52 (95% CI 0.91-7.69) and 3.95 (95% CI 1.50-11.76). From our study, we inferred that our software not only influenced experts' understanding and trust but also impacted decision-making. Several themes were identified through interviews: preference for multiple explanation types, need to visualize uncertainty in explanations provided by PRIMO, and need for model performance metrics on similar participant test instances. CONCLUSIONS: Our results show the potential for weight management experts to agree with the ML-based early prediction of success in weight loss treatment programs, enabling timely and dynamic modification of intervention components to enhance intervention effectiveness. Our findings provide methods for advancing the understandability and trust of ML models among weight management experts.

Health science researchers studying human behavior rely on wearable cameras to visually confirm behaviors in real-world settings. However, privacy concerns significantly impede their adoption. Lens orientation and activity-oriented cameras have potential in balancing the need to visually validate the wearers' activities while reducing privacy concerns. To increase adoption and further alleviate privacy concerns while maintaining utility, generative stylizing approaches, like cartooning using generative adversarial networks (GANs), have recently shown promise. We investigate different cartoon-based obfuscation of activity-oriented footage through two studies. The first deploys crowdsourcing methods (n=60), while the second is experiential, where participants (n=49) don the device for an entire day and report concerns on their footage. Our findings support that cartoonization of activity-oriented data significantly reduces privacy concerns, particularly among bystanders in high privacy-concerning scenarios, while maintaining context verification (90% of participants). Through thematic analysis, we provide further insight for the community on best practices for cartoonization of activity-oriented videos.

Wearable devices have long held the potential to provide real-time objective measures of behavior. However, due to challenges in real-world deployment, these systems are rarely tested rigorously in free-living settings. To reduce this challenge for future researchers, in this paper, we describe our experience developing several generations of a multi-sensor, neck-worn eating-detection system that has been tested with 130 participants across multiple studies in both laboratory and free-living settings. We describe the challenges faced in the development and deployment of the system by (1) presenting example deployment details captured either by the sensing system or the ground truth collector and (2) using structured interviews and surveys with developers and stakeholders of the system, collecting qualitative data on their experience. We performed thematic analysis and provided detailed lessons learned explaining factors that impact the experience of building and deploying such a wearable in a free-living setting, reducing challenges for future researchers. We believe that our experience will help future researchers develop successful mobile health (mHealth) systems that translate into reliable free-living deployments.

Objectives: Overeating interventions and research often focus on single determinants and use subjective or nonpersonalized measures. We aim to (1) identify automatically detectable features that predict overeating and (2) build clusters of eating episodes that identify theoretically meaningful and clinically known problematic overeating behaviors (e.g., stress eating), as well as new phenotypes based on social and psychological features. Method: Up to 60 adults with obesity in the Chicagoland area will be recruited for a 14-day free-living observational study. Participants will complete ecological momentary assessments and wear 3 sensors designed to capture features of overeating episodes (e.g., chews) that can be visually confirmed. Participants will also complete daily dietitian-administered 24-hour recalls of all food and beverages consumed. Analysis: Overeating is defined as caloric consumption exceeding 1 standard deviation of an individual's mean consumption per eating episode. To identify features that predict overeating, we will apply 2 complementary machine learning methods: correlation-based feature selection and wrapper-based feature selection. We will then generate clusters of overeating types and assess how they align with clinically meaningful overeating phenotypes. Conclusions: over a multiweek period with visual confirmation of eating behaviors. An additional strength of this study is the assessment of predictors of problematic eating during periods when individuals are not on a structured diet and/or engaged in a weight loss intervention. Our assessment of overeating episodes in real-world settings is likely to yield new insights regarding determinants of overeating that may translate into novel interventions.