Early adverse physiological event detection using commercial wearables: challenges and opportunities.

Data from commercial off-the-shelf (COTS) wearables leveraged with machine learning algorithms provide an unprecedented potential for the early detection of adverse physiological events. However, several challenges inhibit this potential, including (1) heterogeneity among and within participants that make scaling detection algorithms to a general population less precise, (2) confounders that lead to incorrect assumptions regarding a participant's healthy state, (3) noise in the data at the sensor level that limits the sensitivity of detection algorithms, and (4) imprecision in self-reported labels that misrepresent the true data values associated with a given physiological event. The goal of this study was two-fold: (1) to characterize the performance of such algorithms in the presence of these challenges and provide insights to researchers on limitations and opportunities, and (2) to subsequently devise algorithms to address each challenge and offer insights on future opportunities for advancement. Our proposed algorithms include techniques that build on determining suitable baselines for each participant to capture important physiological changes and label correction techniques as it pertains to participant-reported identifiers. Our work is validated on potentially one of the largest datasets available, obtained with 8000+ participants and 1.3+ million hours of wearable data captured from Oura smart rings. Leveraging this extensive dataset, we achieve pre-symptomatic detection of COVID-19 with a performance receiver operator characteristic (ROC) area under the curve (AUC) of 0.725 without correction techniques, 0.739 with baseline correction, 0.740 with baseline correction and label correction on the training set, and 0.777 with baseline correction and label correction on both the training and the test set. Using the same respective paradigms, we achieve ROC AUCs of 0.919, 0.938, 0.943 and 0.994 for the detection of self-reported fever, and 0.574, 0.611, 0.601, and 0.635 for detection of self-reported shortness of breath. These techniques offer improvements across almost all metrics and events, including PR AUC, sensitivity at 75% specificity, and precision at 75% recall. The ring allows continuous monitoring for detection of event onset, and we further demonstrate an improvement in the early detection of COVID-19 from an average of 3.5 days to an average of 4.1 days before a reported positive test result.

Prospective measurements of hearing threshold during military rifle training with in-ear, protected, noise exposure monitoring.

Although a causal relationship exists between military occupational noise exposure and hearing loss, researchers have struggled to identify and/or characterize specific operational noise exposures that produce measurable changes in hearing function shortly following an exposure. Growing evidence suggests that current standards for noise-exposure limits are not good predictors of true hearing damage. In this study, the aim was to capture the dose-response relationship during military rifle training exercises for noise exposure and hearing threshold. To capture exposure, a wearable system capable of measuring impulse noise simultaneously on-body and in-ear, behind hearing protection was used. To characterize hearing threshold changes, portable audiometry was employed within 2 h before and after exposure. The median 8-h time-weighted, protected, free-field equivalent in-ear exposure was 87.5 dBA at one site and 80.7 dBA at a second site. A significant dose-response correlation between in-ear noise exposure and postexposure hearing threshold changes across our population ( R = 0.40 , p = 0.0281) was observed. The results demonstrate an approach for establishing damage risk criteria (DRC) for in-ear, protected measurements based on hearing threshold changes. While an in-ear DRC does not currently exist, it may be critical for predicting the risk of injury for noise environments where protection is mandatory and fit status can vary.

Detection of COVID-19 using multimodal data from a wearable device: results from the first TemPredict Study.

Early detection of diseases such as COVID-19 could be a critical tool in reducing disease transmission by helping individuals recognize when they should self-isolate, seek testing, and obtain early medical intervention. Consumer wearable devices that continuously measure physiological metrics hold promise as tools for early illness detection. We gathered daily questionnaire data and physiological data using a consumer wearable (Oura Ring) from 63,153 participants, of whom 704 self-reported possible COVID-19 disease. We selected 73 of these 704 participants with reliable confirmation of COVID-19 by PCR testing and high-quality physiological data for algorithm training to identify onset of COVID-19 using machine learning classification. The algorithm identified COVID-19 an average of 2.75 days before participants sought diagnostic testing with a sensitivity of 82% and specificity of 63%. The receiving operating characteristic (ROC) area under the curve (AUC) was 0.819 (95% CI [0.809, 0.830]). Including continuous temperature yielded an AUC 4.9% higher than without this feature. For further validation, we obtained SARS CoV-2 antibody in a subset of participants and identified 10 additional participants who self-reported COVID-19 disease with antibody confirmation. The algorithm had an overall ROC AUC of 0.819 (95% CI [0.809, 0.830]), with a sensitivity of 90% and specificity of 80% in these additional participants. Finally, we observed substantial variation in accuracy based on age and biological sex. Findings highlight the importance of including temperature assessment, using continuous physiological features for alignment, and including diverse populations in algorithm development to optimize accuracy in COVID-19 detection from wearables.

Detecting Pathogen Exposure During the Non-symptomatic Incubation Period Using Physiological Data: Proof of Concept in Non-human Primates.

Background and Objectives: Early warning of bacterial and viral infection, prior to the development of overt clinical symptoms, allows not only for improved patient care and outcomes but also enables faster implementation of public health measures (patient isolation and contact tracing). Our primary objectives in this effort are 3-fold. First, we seek to determine the upper limits of early warning detection through physiological measurements. Second, we investigate whether the detected physiological response is specific to the pathogen. Third, we explore the feasibility of extending early warning detection with wearable devices. Research Methods: For the first objective, we developed a supervised random forest algorithm to detect pathogen exposure in the asymptomatic period prior to overt symptoms (fever). We used high-resolution physiological telemetry data (aortic blood pressure, intrathoracic pressure, electrocardiograms, and core temperature) from non-human primate animal models exposed to two viral pathogens: Ebola and Marburg (N = 20). Second, to determine reusability across different pathogens, we evaluated our algorithm against three independent physiological datasets from non-human primate models (N = 13) exposed to three different pathogens: Lassa and Nipah viruses and Y. pestis. For the third objective, we evaluated performance degradation when the algorithm was restricted to features derived from electrocardiogram (ECG) waveforms to emulate data from a non-invasive wearable device. Results: First, our cross-validated random forest classifier provides a mean early warning of 51 ± 12 h, with an area under the receiver-operating characteristic curve (AUC) of 0.93 ± 0.01. Second, our algorithm achieved comparable performance when applied to datasets from different pathogen exposures - a mean early warning of 51 ± 14 h and AUC of 0.95 ± 0.01. Last, with a degraded feature set derived solely from ECG, we observed minimal degradation - a mean early warning of 46 ± 14 h and AUC of 0.91 ± 0.001. Conclusion: Under controlled experimental conditions, physiological measurements can provide over 2 days of early warning with high AUC. Deviations in physiological signals following exposure to a pathogen are due to the underlying host's immunological response and are not specific to the pathogen. Pre-symptomatic detection is strong even when features are limited to ECG-derivatives, suggesting that this approach may translate to non-invasive wearable devices.

In-ear and on-body measurements of impulse-noise exposure.

Accurate quantification of noise exposure in military environments is challenging due to movement of listeners and noise sources, spectral and temporal noise characteristics, and varied use of hearing protection. This study evaluates a wearable recording device designed to measure on-body and in-ear noise exposure, specifically in an environment with significant impulse noise resulting from firearms. A commercial audio recorder was augmented to obtain simultaneous measurements inside the ear canal behind an integrated hearing protector, and near the outer ear. Validation measurements, conducted with an acoustic test fixture and shock tube, indicated high impulse peak insertion loss with a proper fit of the integrated hearing protector. The recording devices were worn by five subjects during a live-fire data collection at Marine Corps Base Quantico where Marines fired semi-automatic rifles. The field test demonstrated the successful measurement of high-level impulse waveforms with the on-body and in-ear recording system. Dual channels allowed for instantaneous fit estimates for the hearing protection component, and the device worked as intended in terms of hearing protection and noise dosimetry. Accurate measurements of noise exposure and hearing protector fit should improve the ability to model and assess the risks of noise-induced hearing loss.

Noise dosimetry for tactical environments.

Noise exposure and the subsequent hearing loss are well documented aspects of military life. Numerous studies have indicated high rates of noise-induced hearing injury (NIHI) in active-duty service men and women, and recent statistics from the U.S. Department of Veterans Affairs indicate a population of veterans with hearing loss that is growing at an increasing rate. In an effort to minimize hearing loss, the U.S. Department of Defense (DoD) updated its Hearing Conservation Program in 2010, and also has recently revised the DoD Design Criteria Standard Noise Limits (MIL-STD-1474E) which defines allowable noise levels in the design of all military acquisitions including weapons and vehicles. Even with such mandates, it remains a challenge to accurately quantify the noise exposure experienced by a Warfighter over the course of a mission or training exercise, or even in a standard work day. Noise dosimeters are intended for exactly this purpose, but variations in device placement (e.g., free-field, on-body, in/near-ear), hardware (e.g., microphone, analog-to-digital converter), measurement time (e.g., work day, 24-h), and dose metric calculations (e.g., time-weighted energy, peak levels, Auditory Risk Units), as well as noise types (e.g., continuous, intermittent, impulsive) can cause exposure measurements to be incomplete, inaccurate, or inappropriate for a given situation. This paper describes the design of a noise dosimeter capable of acquiring exposure data across tactical environments. Two generations of prototypes have been built at MIT Lincoln Laboratory with funding from the U.S. Army, Navy, and Marine Corps. Details related to hardware, signal processing, and testing efforts are provided, along with example tactical military noise data and lessons learned from early fieldings. Finally, we discuss the continued need to prioritize personalized dosimetry in order to improve models that predict or characterize the risk of auditory damage, to integrate dosimeters with hearing-protection devices, and to inform strategies and metrics for reducing NIHI.

Breast tumor characterization based on ultrawideband microwave backscatter.

Characterization of architectural tissue features such as the shape, margin, and size of a suspicious lesion is commonly performed in conjunction with medical imaging to provide clues about the nature of an abnormality. In this paper, we numerically investigate the feasibility of using multichannel microwave backscatter in the 1-11 GHz band to classify the salient features of a dielectric target. We consider targets with three shape characteristics: smooth, microlobulated, and spiculated; and four size categories ranging from 0.5 to 2 cm in diameter. The numerical target constructs are based on Gaussian random spheres allowing for moderate shape irregularities. We perform shape and size classification for a range of signal-to-noise ratios (SNRs) to demonstrate the potential for tumor characterization based on ultrawideband (UWB) microwave backscatter. We approach classification with two basis selection methods from the literature: local discriminant bases and principal component analysis. Using these methods, we construct linear classifiers where a subset of the bases expansion vectors are the input features and we evaluate the average rate of correct classification as a performance measure. We demonstrate that for 10 dB SNR, the target size is very reliably classified with over 97% accuracy averaged over 360 targets; target shape is classified with over 70% accuracy. The relationship between the SNR of the test data and classifier performance is also explored. The results of this study are very encouraging and suggest that both shape and size characteristics of a dielectric target can be classified directly from its UWB backscatter. Hence, characterization can easily be performed in conjunction with UWB radar-based breast cancer detection without requiring any special hardware or additional data collection.

Development of anatomically realistic numerical breast phantoms with accurate dielectric properties for modeling microwave interactions with the human breast.

Computational electromagnetics models of microwave interactions with the human breast serve as an invaluable tool for exploring the feasibility of new technologies and improving design concepts related to microwave breast cancer detection and treatment. In this paper, we report the development of a collection of anatomically realistic 3-D numerical breast phantoms of varying shape, size, and radiographic density which can readily be used in finite-difference time-domain computational electromagnetics models. The phantoms are derived from T1-weighted MRIs of prone patients. Each MRI is transformed into a uniform grid of dielectric properties using several steps. First, the structure of each phantom is identified by applying image processing techniques to the MRI. Next, the voxel intensities of the MRI are converted to frequency-dependent and tissue-dependent dielectric properties of normal breast tissues via a piecewise-linear map. The dielectric properties of normal breast tissue are taken from the recently completed large-scale experimental study of normal breast tissue dielectric properties conducted by the Universities of Wisconsin and Calgary. The comprehensive collection of numerical phantoms is made available to the scientific community through an online repository.

Ultrawideband microwave breast cancer detection: a detection-theoretic approach using the generalized likelihood ratio test.

Microwave imaging has been suggested as a promising modality for early-stage breast cancer detection. In this paper, we propose a statistical microwave imaging technique wherein a set of generalized likelihood ratio tests (GLRT) is applied to microwave backscatter data to determine the presence and location of strong scatterers such as malignant tumors in the breast. The GLRT is formulated assuming that the backscatter data is Gaussian distributed with known covariance matrix. We describe the method for estimating this covariance matrix offline and formulating a GLRT for several heterogeneous two-dimensional (2-D) numerical breast phantoms, several three-dimensional (3-D) experimental breast phantoms, and a 3-D numerical breast phantom with a realistic half-ellipsoid shape. Using the GLRT with the estimated covariance matrix and a threshold chosen to constrain the false discovery rate (FDR) of the image, we show the capability to detect and localize small (<0.6 cm) tumors in our numerical and experimental breast phantoms even when the dielectric contrast of the malignant-to-normal tissue is below 2:1.