Identifying time-resolved features of nocturnal sleep characteristics of narcolepsy using machine learning.

The differential diagnosis of narcolepsy type 1, a rare, chronic, central disorder of hypersomnolence, is challenging due to overlapping symptoms with other hypersomnolence disorders. While recent years have seen significant growth in our understanding of nocturnal polysomnography narcolepsy type 1 features, there remains a need for improving methods to differentiate narcolepsy type 1 nighttime sleep features from those of individuals without narcolepsy type 1. We aimed to develop a machine learning framework for identifying sleep features to discriminate narcolepsy type 1 from clinical controls, narcolepsy type 2 and idiopathic hypersomnia. The population included polysomnography data from 350 drug-free individuals (114 narcolepsy type 1, 90 narcolepsy type 2, 105 idiopathic hypersomnia, and 41 clinical controls) collected at the National Reference Centers for Narcolepsy in Montpelier, France. Several sets of nocturnal sleep features were explored, as well as the value of time-resolving sleep architecture by analysing sleep per quarter-night. Several patterns of nighttime sleep evolution emerged that differed between narcolepsy type 1, clinical controls, narcolepsy type 2 and idiopathic hypersomnia, with increased nighttime instability observed in patients with narcolepsy type 1. Using machine learning models, we identified rapid eye movement sleep onset as the best single polysomnography feature to distinguish narcolepsy type 1 from controls, narcolepsy type 2 and idiopathic hypersomnia. By combining multiple feature sets capturing different aspects of sleep across quarter-night periods, we were able to further improve between-group discrimination and could identify the most discriminative sleep features. Our results highlight salient polysomnography features and the relevance of assessing their time-dependent changes during sleep that could aid diagnosis and measure the impact of novel therapeutics in future clinical trials.

Dynamics of Cough Frequency in Adults Undergoing Treatment for Pulmonary Tuberculosis.

Background: Cough is the major determinant of tuberculosis transmission. Despite this, there is a paucity of information regarding characteristics of cough frequency throughout the day and in response to tuberculosis therapy. Here we evaluate the circadian cycle of cough, cough frequency risk factors, and the impact of appropriate treatment on cough and bacillary load. Methods: We prospectively evaluated human immunodeficiency virus-negative adults (n = 64) with a new diagnosis of culture-proven, drug-susceptible pulmonary tuberculosis immediately prior to treatment and repeatedly until treatment day 62. At each time point, participant cough was recorded (n = 670) and analyzed using the Cayetano Cough Monitor. Consecutive coughs at least 2 seconds apart were counted as separate cough episodes. Sputum samples (n = 426) were tested with microscopic-observation drug susceptibility broth culture, and in culture-positive samples (n = 252), the time to culture positivity was used to estimate bacillary load. Results: The highest cough frequency occurred from 1 pm to 2 pm, and the lowest from 1 am to 2 am (2.4 vs 1.1 cough episodes/hour, respectively). Cough frequency was higher among participants who had higher sputum bacillary load (P < .01). Pretreatment median cough episodes/hour was 2.3 (interquartile range [IQR], 1.2-4.1), which at 14 treatment days decreased to 0.48 (IQR, 0.0-1.4) and at the end of the study decreased to 0.18 (IQR, 0.0-0.59) (both reductions P < .001). By 14 treatment days, the probability of culture conversion was 29% (95% confidence interval, 19%-41%). Conclusions: Coughs were most frequent during daytime. Two weeks of appropriate treatment significantly reduced cough frequency and resulted in one-third of participants achieving culture conversion. Thus, treatment by 2 weeks considerably diminishes, but does not eliminate, the potential for airborne tuberculosis transmission.

Locally optimized non-local means denoising for low-dose X-ray backscatter imagery.

While recent years have seen considerable progress in image denoising, the leading techniques have been developed for digital photographs or other images that can have very different characteristics than those encountered in X-ray applications. In particular here we examine X-ray backscatter (XBS) images collected by airport security systems, where images are piecewise smooth and edge information is typically more correlated with objects while texture is dominated by statistical noise in the detected signal. In this paper, we show how multiple estimates for a denoised XBS image can be combined using a variational approach, giving a solution that enhances edge contrast by trading off gradient penalties against data fidelity terms. We demonstrate the approach by combining several estimates made using the non-local means (NLM) algorithm, a widely used patch-based denoising method. The resulting improvements hold the potential for improving automated analysis of low-SNR X-ray imagery and can be applied in other applications where edge information is of interest.

Novel biomarkers derived from the Maintenance of Wakefulness Test as predictors of sleepiness and response to treatment.

The Maintenance of Wakefulness Test (MWT) is a widely accepted objective test used to evaluate daytime somnolence and is commonly used in clinical studies evaluating novel therapeutics for excessive daytime sleepiness. In the latter, sleep onset latency (SOL) is typically the sole MWT endpoint. Here, we explored microsleeps, sleep probability measures derived from automated sleep scoring, and quantitative electroencephalography (qEEG) features as additional MWT biomarkers of daytime sleepiness, using data from a phase 1B trial of the selective orexin receptor 2 agonist danavorexton (TAK-925) in people with narcolepsy type 1 (NT1) or type 2 (NT2). Danavorexton treatment reduced the rate and duration of microsleeps during the MWT in NT1 (days 1 and 7; p ≤ 0.005) and microsleep rate in NT2 (days 1 and 7; p < 0.0001). Use of an EEG-sleep-staging-derived measure to determine the probability of wakefulness for each minute revealed a novel metric to track changes in daytime sleepiness, which were consistent with the θ/α ratio, a known biomarker of drowsiness. The slopes of line-fits to both the log-transformed sleepiness score or log-transformed θ/α ratio correlated well to (inverse) MWT SOL for NT1 (R = 0.93 and R = 0.83, respectively) and NT2 (R = 0.97 and R = 0.84, respectively), suggesting that individuals with narcolepsy have increased sleepiness immediately after lights-off. These analyses demonstrate that novel EEG-based biomarkers can augment SOL as predictors of sleepiness and its response to treatment and provide a novel framework for the analysis of wake EEG in hypersomnia disorders.

Using a smartwatch and smartphone to assess early Parkinson's disease in the WATCH-PD study over 12 months.

Digital measures may provide objective, sensitive, real-world measures of disease progression in Parkinson's disease (PD). However, multicenter longitudinal assessments of such measures are few. We recently demonstrated that baseline assessments of gait, tremor, finger tapping, and speech from a commercially available smartwatch, smartphone, and research-grade wearable sensors differed significantly between 82 individuals with early, untreated PD and 50 age-matched controls. Here, we evaluated the longitudinal change in these assessments over 12 months in a multicenter observational study using a generalized additive model, which permitted flexible modeling of at-home data. All measurements were included until participants started medications for PD. Over one year, individuals with early PD experienced significant declines in several measures of gait, an increase in the proportion of day with tremor, modest changes in speech, and few changes in psychomotor function. As measured by the smartwatch, the average (SD) arm swing in-clinic decreased from 25.9 (15.3) degrees at baseline to 19.9 degrees (13.7) at month 12 (P = 0.004). The proportion of awake time an individual with early PD had tremor increased from 19.3% (18.0%) to 25.6% (21.4%; P < 0.001). Activity, as measured by the number of steps taken per day, decreased from 3052 (1306) steps per day to 2331 (2010; P = 0.16), but this analysis was restricted to 10 participants due to the exclusion of those that had started PD medications and lost the data. The change of these digital measures over 12 months was generally larger than the corresponding change in individual items on the Movement Disorder Society-Unified Parkinson's Disease Rating Scale but not greater than the change in the overall scale. Successful implementation of digital measures in future clinical trials will require improvements in study conduct, especially data capture. Nonetheless, gait and tremor measures derived from a commercially available smartwatch and smartphone hold promise for assessing the efficacy of therapeutics in early PD.

Long-term measurement study of urban environmental low frequency noise.

BACKGROUND: Environmental low frequency noise (LFN < 125 Hz), ubiquitous in urban areas, is an understudied area of exposure science and an overlooked threat to population health. Environmental noise has historically been measured and regulated by A-weighted decibel (dBA) metrics, which more heavily weight frequencies between 2000 and 5000 Hz. Limited research has been conducted to measure and characterize the LFN components of urban environmental noise. OBJECTIVES: We characterized LFN noise at two urban sites in Greater Boston, Massachusetts (USA) using dBA and full spectrum noise measurements with aims to (1.) analyze spatio-temporal differences in the two datasets; (2.) compare and contrast LFN metrics with dBA noise metrics in the two sites; and (3.) assess meteorological covariate contributions to LFN in the dataset. METHODS: We measured A- and C-weighted, and flat, unweighted noise levels and 1/3-octave band continuously for 5 months using sound level meters sampling at f = 1 Hz and we recorded sound samples at 44.1 kHz. Our measurement sites were located in two urban, densely populated communities, burdened by close proximity to bus, rail, and aircraft routes. RESULTS: We found that (1.) LFN does not follow the same seasonal trends as A-weighted dBA loudness; there are spatial differences in LFN and its very low frequency noise components (VLFN) between two urban sites; (2.) VLFN and LFN are statistically significant drivers of LCeq (nearly independent of frequency) minus LAeq, (LCeq-LAeq) >10 dB, an accepted LFN metric; and (3.) LFN was minimally affected by high wind speeds at either Site. IMPACT STATEMENT: Environmental low-frequency noise (LFN < 125 Hz), ubiquitous in urban areas, is an understudied area of exposure science and an overlooked risk to population health. We measured environmental noise across the full spectrum of frequencies continuously for five months at two urban sites located in Environmental Justice communities. We found that LFN did not follow the same seasonal trends as A-weighted (dBA) loudness, and we observed spatial differences in LFN and very low frequency noise (VLFN < 20 Hz) at the two sites. Not characterizing LFN and basing noise regulations only on A-weightings, a poor predictor of LFN, may expose populations to LFN levels of concern.

Towards interpretable speech biomarkers: exploring MFCCs.

While speech biomarkers of disease have attracted increased interest in recent years, a challenge is that features derived from signal processing or machine learning approaches may lack clinical interpretability. As an example, Mel frequency cepstral coefficients (MFCCs) have been identified in several studies as a useful marker of disease, but are regarded as uninterpretable. Here we explore correlations between MFCC coefficients and more interpretable speech biomarkers. In particular we quantify the MFCC2 endpoint, which can be interpreted as a weighted ratio of low- to high-frequency energy, a concept which has been previously linked to disease-induced voice changes. By exploring MFCC2 in several datasets, we show how its sensitivity to disease can be increased by adjusting computation parameters.

Using a smartwatch and smartphone to assess early Parkinson's disease in the WATCH-PD study.

Digital health technologies can provide continuous monitoring and objective, real-world measures of Parkinson's disease (PD), but have primarily been evaluated in small, single-site studies. In this 12-month, multicenter observational study, we evaluated whether a smartwatch and smartphone application could measure features of early PD. 82 individuals with early, untreated PD and 50 age-matched controls wore research-grade sensors, a smartwatch, and a smartphone while performing standardized assessments in the clinic. At home, participants wore the smartwatch for seven days after each clinic visit and completed motor, speech and cognitive tasks on the smartphone every other week. Features derived from the devices, particularly arm swing, the proportion of time with tremor, and finger tapping, differed significantly between individuals with early PD and age-matched controls and had variable correlation with traditional assessments. Longitudinal assessments will inform the value of these digital measures for use in future clinical trials.

Voice Biomarkers of Recovery From Acute Respiratory Illness.

Voice analysis is an emerging technology which has the potential to provide low-cost, at-home monitoring of symptoms associated with a variety of health conditions. While voice has received significant attention for monitoring neurological disease, few studies have focused on voice changes related to flu-like symptoms. Herein, we investigate the relationship between changes in acoustic features of voice and self-reported symptoms during recovery from a flu-like illness in a cohort of 29 subjects. Acoustic features were automatically extracted from "sick" and "well" visit data collected in the laboratory setting, and feature down-selection was used to identify those that change significantly between visits. The selected acoustic features were extracted from at-home data and used to construct a combined distance metric that correlated with self-reported symptoms (0.63 rank correlation). Changes in self-reported symptoms corresponding to 10% of the ordinal scale used in the study were detected with an area under the curve of 0.72. The results show that acoustic features derived from voice recordings may provide an objective measure for diagnosing and monitoring symptoms of respiratory illnesses.

Photo-ID and telemetry highlight a global whale shark hotspot in Palawan, Philippines.

The Philippines is home to the second largest known population of whale sharks in the world. The species is listed as endangered due to continued population declines in the Indo-Pacific. Knowledge about the connectivity within Southeast Asia remains poor, and thus international management is difficult. Here, we employed pop-up archival tags, data mining and dedicated effort to understand an aggregation of whale sharks at Honda Bay, Palawan, Philippines, and its role in the species' conservation. Between Apr and Oct 2018, we conducted 159 surveys identifying 117 individual whale sharks through their unique spot patterns (96.5% male, mean 4.5 m). A further 66 individual whale sharks were identified from local operators, and data mined on social media platforms. The satellite telemetry data showed that the whale sharks moved broadly, with one individual moving to Sabah, Malaysia, before returning to the site <1 year later. Similarly, another tagged whale shark returned to the site at a similar periodicity after reaching the Malay-Filipino border. One individual whale shark first identified in East Kalimantan, Indonesia by a citizen scientist was resighted in Honda Bay ~3.5 years later. Honda Bay is a globally important site for the endangered whale shark with connectivity to two neighbouring countries, highlighting the need for international cooperation to manage the species.

Limited-View X-Ray Tomography Combining Attenuation and Compton Scatter Data: Approach and Experimental Results.

X-ray inspection systems are critical in medical, non-destructive testing, and security applications, with systems typically measuring attenuation along straight-line paths connecting sources and detectors. Computed tomography (CT) systems can provide higher-quality images than single- or dual-view systems, but the need to measure many projections leads to greater system cost and complexity. Typically, off-angle Compton scattered photons are treated as noise during tomographic inversion. We seek to maximize the image quality of limited-view systems by combining attenuation data with measurements of Compton-scattered photons, exploiting the fact that the broken-ray paths followed by scattered photons provide additional geometric sampling of the scene. We describe a single-scatter forward model for Compton-scatter data measured with energy-resolving detectors, and demonstrate a reconstruction algorithm for density that combines both attenuation and scatter measurements. The experimental results suggest that including Compton-scattered data in the reconstruction process can improve image quality for density reconstruction using limited-view systems.

Cough Frequency During Treatment Associated With Baseline Cavitary Volume and Proximity to the Airway in Pulmonary TB.

BACKGROUND: Cough frequency, and its duration, is a biomarker that can be used in low-resource settings without the need of laboratory culture and has been associated with transmission and treatment response. Radiologic characteristics associated with increased cough frequency may be important in understanding transmission. The relationship between cough frequency and cavitary lung disease has not been studied. METHODS: We analyzed data in 41 adults who were HIV negative and had culture-confirmed, drug-susceptible pulmonary TB throughout treatment. Cough recordings were based on the Cayetano Cough Monitor, and sputum samples were evaluated using microscopic observation drug susceptibility broth culture; among culture-positive samples, bacillary burden was assessed by means of time to positivity. CT scans were analyzed by a US-board-certified radiologist and a computer-automated algorithm. The algorithm evaluated cavity volume and cavitary proximity to the airway. CT scans were obtained within 1 month of treatment initiation. We compared small cavities (≤ 7 mL) and large cavities (> 7 mL) and cavities located closer to (≤ 10 mm) and farther from (> 10 mm) the airway to cough frequency and cough cessation until treatment day 60. RESULTS: Cough frequency during treatment was twofold higher in participants with large cavity volumes (rate ratio [RR], 1.98; P = .01) and cavities located closer to the airway (RR, 2.44; P = .001). Comparably, cough ceased three times faster in participants with smaller cavities (adjusted hazard ratio [HR], 2.89; P = .06) and those farther from the airway (adjusted HR, 3.61;, P = .02). Similar results were found for bacillary burden and culture conversion during treatment. CONCLUSIONS: Cough frequency during treatment is greater and lasts longer in patients with larger cavities, especially those closer to the airway.

Protocol for studying cough frequency in people with pulmonary tuberculosis.

INTRODUCTION: Cough is a key symptom of tuberculosis (TB) as well as the main cause of transmission. However, a recent literature review found that cough frequency (number of coughs per hour) in patients with TB has only been studied once, in 1969. The main aim of this study is to describe cough frequency patterns before and after the start of TB treatment and to determine baseline factors that affect cough frequency in these patients. Secondarily, we will evaluate the correlation between cough frequency and TB microbiological resolution. METHODS: This study will select participants with culture confirmed TB from 2 tertiary hospitals in Lima, Peru. We estimated that a sample size of 107 patients was sufficient to detect clinically significant changes in cough frequency. Participants will initially be evaluated through questionnaires, radiology, microscopic observation drug susceptibility broth TB-culture, auramine smear microscopy and cough recordings. This cohort will be followed for the initial 60 days of anti-TB treatment, and throughout the study several microbiological samples as well as 24 h recordings will be collected. We will describe the variability of cough episodes and determine its association with baseline laboratory parameters of pulmonary TB. In addition, we will analyse the reduction of cough frequency in predicting TB cure, adjusted for potential confounders. ETHICS AND DISSEMINATION: Ethical approval has been obtained from the ethics committees at each participating hospital in Lima, Peru, Asociación Benéfica PRISMA in Lima, Peru, the Universidad Peruana Cayetano Heredia in Lima, Peru and Johns Hopkins University in Baltimore, USA. We aim to publish and disseminate our findings in peer-reviewed journals. We also expect to create and maintain an online repository for TB cough sounds as well as the statistical analysis employed.

Standardizing Immunohistochemistry: A New Reference Control for Detecting Staining Problems.

A new standardized immunohistochemistry (IHC) control for breast cancer testing comprises formalin-fixed human epidermal growth factor receptor 2, estrogen receptor, or progesterone receptor peptide antigens covalently attached to 8-µm glass beads. The antigen-coated beads are suspended in a liquid matrix that hardens upon pipetting onto a glass microscope slide. The antigen-coated beads remain in place through deparaffinization, antigen retrieval, and immunostaining. The intensity of the beads' stain provides feedback regarding the efficacy of both antigen retrieval and immunostaining. As a first report, we tested the sensitivity and specificity of the new IHC controls ("IHControls"). To evaluate sensitivity, various staining problems were simulated. IHControls detected primary and secondary reagent degradation similarly to tissue controls. This first group of IHControls behaved similarly to tissue controls expressing high concentrations of the antigen. The IHControls were also able to detect aberrations in antigen retrieval, as simulated by sub-optimal times or temperatures. Specificity testing revealed that each antigen-coated bead was specific for its cognate IHC test antibody. The data support the conclusion that, like tissue controls, IHControls are capable of verifying the analytic components of an immunohistochemical stain. Unlike tissue controls, IHControls are prepared in large bulk lots, fostering day-to-day reproducibility that can be standardized across laboratories.

Analysis of the trajectory of osteoarthritis development in a mouse model by serial near-infrared fluorescence imaging of matrix metalloproteinase activities.

OBJECTIVE: A major hurdle in osteoarthritis (OA) research is the lack of sensitive detection and monitoring methods. It is hypothesized that proteases, such as matrix metalloproteinases (MMPs), are up-regulated in the early stages of OA development. This study was undertaken to investigate if a near-infrared (NIR) fluorescent probe activated by MMPs could visualize in vivo OA progression beginning in the early stages of the disease. METHODS: Using an MMP-activatable NIR fluorescent probe (MMPSense 680), we assessed the up-regulation of MMP activity in vitro by incubating human chondrocytes with the proinflammatory cytokine interleukin-1ß (IL-1ß). MMP activity was then evaluated in vivo serially in a mouse model of chronic, injury-induced OA. To track MMP activity over time, mice were imaged 1-8 weeks after OA-inducing surgery. Imaging results were correlated with histologic findings. RESULTS: In vitro studies confirmed that NIR fluorescence imaging identified enhanced MMP activity in IL-1ß-treated human chondrocytes. In vivo imaging showed significantly higher fluorescence intensity in OA knees compared to sham-operated (control) knees of the same mice. Additionally, the total emitted fluorescence intensity steadily increased over the entire course of OA progression that was examined. NIR fluorescence imaging results correlated with histologic findings, which showed an increase in articular cartilage structural damage over time. CONCLUSION: Imaging of MMP activity in a mouse model of OA provides sensitive and consistent visualization of OA progression, beginning in the early stages of OA. In addition to facilitating the preclinical study of OA modulators, this approach has the potential for future translation to humans.

Nonlocal means denoising of ECG signals.

Patch-based methods have attracted significant attention in recent years within the field of image processing for a variety of problems including denoising, inpainting, and super-resolution interpolation. Despite their prevalence for processing 2-D signals, they have received little attention in the 1-D signal processing literature. In this letter, we explore application of one such method, the nonlocal means (NLM) approach, to the denoising of biomedical signals. Using ECG as an example, we demonstrate that a straightforward NLM-based denoising scheme provides signal-to-noise ratio improvements very similar to state of the art wavelet-based methods, while giving ~3 × or greater reduction in metrics measuring distortion of the denoised waveform.

Validation of an automated cough detection algorithm for tracking recovery of pulmonary tuberculosis patients.

BACKGROUND: A laboratory-free test for assessing recovery from pulmonary tuberculosis (TB) would be extremely beneficial in regions of the world where laboratory facilities are lacking. Our hypothesis is that analysis of cough sound recordings may provide such a test. In the current paper, we present validation of a cough analysis tool. METHODOLOGY/PRINCIPAL FINDINGS: Cough data was collected from a cohort of TB patients in Lima, Peru and 25.5 hours of recordings were manually annotated by clinical staff. Analysis software was developed and validated by comparison to manual scoring. Because many patients cough in bursts, coughing was characterized in terms of cough epochs. Our software correctly detects 75.5% of cough episodes with a specificity of 99.6% (comparable to past results using the same definition) and a median false positive rate of 4 false positives/hour, due to the noisy, real-world nature of our dataset. We then manually review detected coughs to eliminate false positives, in effect using the algorithm as a pre-screening tool that reduces reviewing time to roughly 5% of the recording length. This cough analysis approach provides a foundation to support larger-scale studies of coughing rates over time for TB patients undergoing treatment.

Computerised lung sound analysis to improve the specificity of paediatric pneumonia diagnosis in resource-poor settings: protocol and methods for an observational study.

INTRODUCTION: WHO case management algorithm for paediatric pneumonia relies solely on symptoms of shortness of breath or cough and tachypnoea for treatment and has poor diagnostic specificity, tends to increase antibiotic resistance. Alternatives, including oxygen saturation measurement, chest ultrasound and chest auscultation, exist but with potential disadvantages. Electronic auscultation has potential for improved detection of paediatric pneumonia but has yet to be standardised. The authors aim to investigate the use of electronic auscultation to improve the specificity of the current WHO algorithm in developing countries. METHODS: This study is designed to test the hypothesis that pulmonary pathology can be differentiated from normal using computerised lung sound analysis (CLSA). The authors will record lung sounds from 600 children aged ≤5 years, 100 each with consolidative pneumonia, diffuse interstitial pneumonia, asthma, bronchiolitis, upper respiratory infections and normal lungs at a children's hospital in Lima, Peru. The authors will compare CLSA with the WHO algorithm and other detection approaches, including physical exam findings, chest ultrasound and microbiologic testing to construct an improved algorithm for pneumonia diagnosis. DISCUSSION: This study will develop standardised methods for electronic auscultation and chest ultrasound and compare their utility for detection of pneumonia to standard approaches. Utilising signal processing techniques, the authors aim to characterise lung sounds and through machine learning, develop a classification system to distinguish pathologic sounds. Data will allow a better understanding of the benefits and limitations of novel diagnostic techniques in paediatric pneumonia.

Computationally efficient bioelectric field modeling and effects of frequency-dependent tissue capacitance.

Standard bioelectric field models assume that the tissue is purely resistive and frequency independent, and that capacitance, induction, and propagation effects can be neglected. However, real tissue properties are frequency dependent, and tissue capacitance can be important for problems involving short stimulation pulses. A straightforward interpolation scheme is introduced here that can account for frequency-dependent effects, while reducing runtime over a direct computation by several orders of magnitude. The exact Helmholtz solution is compared to several approximate field solutions and is used to study neural stimulation. Results show that frequency-independent tissue capacitance always acts to attenuate the stimulation pulse, thereby increasing firing thresholds, while the dispersion effects introduced by frequency-dependent capacitance may decrease firing thresholds.

Cough detection algorithm for monitoring patient recovery from pulmonary tuberculosis.

In regions of the world where tuberculosis (TB) poses the greatest disease burden, the lack of access to skilled laboratories is a significant problem. A lab-free method for assessing patient recovery during treatment would be of great benefit, particularly for identifying patients who may have drug-resistant tuberculosis. We hypothesize that cough analysis may provide such a test. In this paper we describe algorithm development in support of a pilot study of TB patient coughing. We describe several approaches to event detection and classification, and show preliminary data which suggest that cough count decreases after the start of treatment in drug-responsive patients. Our eventual goal is development of a low-cost ambulatory cough analysis system that will help identify patients with drug-resistant tuberculosis.