Smartwatch Accelerometer Step Counting That Rejects False Positives During Non-Walking Wrist Movement.

IntroductionPhysical activity is an essential aspect of a healthy lifestyle for both physical and mental health states. As step count is one of the most utilized measures for quantifying physical activity it is important that activity-monitoring devices be both sensitive and specific in recording actual steps taken and disregard non-stepping body movements. The objective of this study was to assess the specificity of five activity monitors during a variety of prescribed non-stepping activities.MethodsParticipants wore five activity monitors simultaneously for a variety of prescribed activities including deskwork, taking an elevator, taking a bus journey, automobile driving, washing and drying dishes; functional reaching task; indoor cycling; outdoor cycling; and indoor rowing. Each task was carried out for either a specific duration of time or over a specific distance. Activity monitors tested were the ActivPAL micro™, NL-2000™ pedometer, Withings Smart Activity Monitor Tracker (Pulse O2)™, Fitbit One™ and Jawbone UP™. Participants were video-recorded while carrying out the prescribed activities and the false positive step count registered on each activity monitor was obtained and compared to the video.ResultsAll activity monitors registered a significant number of false positive steps per minute during one or more of the prescribed activities. The Withings™ activity performed best, registering a significant number of false positive steps per minute during the outdoor cycling activity only (P = 0.025). The Jawbone™ registered a significant number of false positive steps during the functional reaching task and while washing and drying dishes, which involved arm and hand movement (P < 0.01 for both). The ActivPAL™ registered a significant number of false positive steps during the cycling exercises (P < 0.001 for both).ConclusionAs a number of false positive steps were registered on the activity monitors during the non-stepping activities, the authors conclude that non-stepping physical activities can result in the false detection of steps. This can negatively affect the quantification of physical activity with regard to step count as an output. The Withings™ activity monitor performed best with regard to specificity during the activities of daily living tested.

[Purpose]The present study compared energy metabolism between walking and running at equivalent speeds during two incremental exercise tests.[Methods]Thirty four university students (18 males, 16 females) were recruited. Each participant completed two trials, consisting of walking (Walk) and running (Run) trials on different days, with 2-3 days apart. Exercise on a treadmill was started from initial stage of 3 min (3.0 k/m in Walk trial, 5.0 km/h in Run trial), and the speed for walking and running was progressively every minute by 0.5 km/h. The changes in metabolic variables, heart rate (HR), and rating of perceived exertion (RPE) during exercise were compared between the trials.[Results]Energy expenditure (EE) increased with speed in each trial. However, the Walk trial had a significantly higher EE than the Run trial at speeds exceeding 92 ± 2 % of the maximal walking speed (MWS, p < 0.01). Similarly, carbohydrate (CHO) oxidation was significantly higher in the Walk trial than in the Run trial at above 92 ± 2 %MWS in males (p < 0.001) and above 93 ± 1 %MWS in females (p < 0.05).[Conclusion]These findings suggest that EE and CHO oxidation during walking increase non-linearly with speed, and walking at a fast speed causes greater metabolic responses than running at the equivalent speed in young participants.

The aim of this study was to determine the reliability, validity, and optimal placement of pedometers in children with cerebral palsy (CP) who ambulate without aids. Seventeen participants aged 7 to 17 years with CP (eight males, nine females; mean age 12y 4mo; SD 3y 2mo), who could ambulate without aids, wore four New Lifestyles pedometers (NL-1000) on an elasticized waist belt. Fourteen participants had hemiplegia, two diplegia, and one triplegia; all were classified in Gross Motor Function Classification System (GMFCS) level I (n=8) or II (n=9). Participants completed 3-minute walking and running trials around an indoor course and were videotaped to verify the actual number of steps taken during each trial. Inter-pedometer reliability was determined by comparing pedometer readings using intraclass correlation coefficients (ICCs). Validity was determined by comparing pedometer step counts with video step counts using ICC, t-tests, and Bland-Altman plots. Optimal pedometer placement was determined using Wilcoxon signed-rank tests to compare the percentage error for pedometers positioned on the dominant and non-dominant hips. Excellent reliability (ICC 0.88-0.99) and validity (ICC 0.78-0.95) were demonstrated with no significant difference between the video step counts and pedometer step counts. There was no significant difference between the step counts recorded by pedometers on the dominant and non-dominant hips. This study showed that NL-1000 pedometers have a high degree of reliability and validity in ambulant children with CP in controlled conditions.

The Physical Activity Guidelines Advisory Committee Scientific Report identified important research gaps to inform future guidance for adolescents, including limited evidence on the importance of sedentary behaviors (screen time) and their interactions with physical activity for adolescent health outcomes, including overweight and obesity. To identify the independent associations of physical activity and screen time categories, and the interactions between physical activity and screen time categories, with body mass index (BMI) and overweight and obesity in adolescents. This cross-sectional study used data from the Adolescent Brain Cognitive Development (ABCD) Study collected from September 10, 2018, to September 29, 2020. Data were analyzed from July 8 to December 20, 2022. A total of 5797 adolescents aged 10 to 14 years from 21 racially and ethnically diverse study sites across the US were included in the analysis. Categories of total step count per day (with 1000 to 6000 steps per day indicating low, >6000 to 12 000 steps per day indicating medium, and >12 000 steps per day indicating high), as measured by a wearable digital device (Fitbit), and categories of self-reported screen time hours per day (with 0 to 4 hours per day indicating low, >4 to 8 hours per day indicating medium, and >8 hours per day indicating high). Participant BMI was calculated as weight in kilograms divided by height in meters squared and converted into sex- and age-specific percentiles in accordance with the Centers for Disease Control and Prevention growth curves and definitions. Individuals were classified as having overweight or obesity if their BMI was in the 85th percentile or higher for sex and age. Among 5797 adolescents included in the analytic sample, 50.4% were male, 61.0% were White, 35.0% had overweight or obesity, and the mean (SD) age was 12.0 (0.6) years. Mean (SD) reported screen time use was 6.5 (5.4) hours per day, and mean (SD) overall step count was 9246.6 (3111.3) steps per day. In models including both screen time and step count, medium (risk ratio [RR], 1.24; 95% CI, 1.12-1.37) and high (RR, 1.29; 95% CI, 1.16-1.44) screen time categories were associated with higher overweight or obesity risk compared with the low screen time category. Medium (RR, 1.19; 95% CI, 1.06-1.35) and low (RR, 1.30; 95% CI, 1.11-1.51) step count categories were associated with higher overweight or obesity risk compared with the high step count category. Evidence of effect modification between screen time and step count was observed for BMI percentile. For instance, among adolescents with low screen use, medium step count was associated with a 1.55 higher BMI percentile, and low step count was associated with a 7.48 higher BMI percentile. However, among those with high screen use, step count categories did not significantly change the association with higher BMI percentile (low step count: 8.79 higher BMI percentile; medium step count: 8.76 higher BMI percentile; high step count: 8.26 higher BMI percentile). In this cross-sectional study, a combination of low screen time and high step count was associated with lower BMI percentile in adolescents. These results suggest that high step count may not offset higher overweight or obesity risk for adolescents with high screen time, and low screen time may not offset higher overweight or obesity risk for adolescents with low step count. These findings addressed several research gaps identified by the Physical Activity Guidelines Advisory Committee Scientific Report and may be used to inform future screen time and physical activity guidance for adolescents.

Accelerometer-based activity monitors are commonly used by researchers and clinicians to assess physical activity. Recently, the Kenz Lifecorder EX (KL) and ActiGraph GT1M (AG) accelerometers have been made commercially available, but there is limited research on the validity of these devices. Therefore, we sought to validate step count, activity energy expenditure (EE), and total EE output from the KL and AG during treadmill walking and running. Ten male and 10 female participants performed 10 min treadmill walking and running trials, at speeds of 54, 80, 107, 134, 161, and 188 m.min-1. Step counts were hand tallied by 2 observers, and indirect calorimetry was used to validate the accelerometers' estimates of EE. AG total EE was calculated using the Freedson equation. Analysis of variance (ANOVA) and Pearson's correlations were used to analyze the data. At the slowest walking speed, the AG and KL counted 64% +/- 15% and 92% +/- 6% of the observed steps, respectively. At all other treadmill speeds, both activity monitors undercounted, compared with observed steps, by < or =3%. The KL underestimated activity EE at faster running speeds (p < 0.01), overestimated total EE at some walking speeds, and underestimated total EE at some running speeds (p < 0.01). The Freedson equation inaccurately measured total EE at most walking and running speeds. The KL and the AG are moderately priced accelerometers that provide researchers and clinicians with accurate estimates of step counts and activity EE at most walking and running speeds.

Step counting is an effective method to assess the activity level of grazing sheep. However, existing step-counting algorithms have limited adaptability to sheep walking patterns and fail to eliminate false step counts caused by abnormal behaviors. Therefore, this study proposed a step-counting algorithm based on behavior classification designed explicitly for grazing sheep. The algorithm utilized regional peak detection and peak-to-valley difference detection to identify running and leg-shaking behaviors in sheep. It distinguished leg shaking from brisk walking behaviors through variance feature analysis. Based on the recognition results, different step-counting strategies were employed. When running behavior was detected, the algorithm divided the sampling window by the baseline step frequency and multiplied it by a scaling factor to accurately calculate the number of steps for running. No step counting was performed for leg-shaking behavior. For other behaviors, such as slow and brisk walking, a window peak detection algorithm was used for step counting. Experimental results demonstrate a significant improvement in the accuracy of the proposed algorithm compared to the peak detection-based method. In addition, the experimental results demonstrated that the average calculation error of the proposed algorithm in this study was 6.244%, while the average error of the peak detection-based step-counting algorithm was 17.556%. This indicates a significant improvement in the accuracy of the proposed algorithm compared to the peak detection method.

Accurate structural variant detection from short-read sequencing data remains challenged by false positives, particularly for heterozygous deletions where reduced allelic support and coverage-based detection methods are ambiguous. Existing SV genotyping and filtering approaches suffer from significant recall reductions, dependencies on additional pre-computed resources, or restriction to depth-based signals that overlook read level evidence. Here we present SVhet, a novel computational framework that leverages the heterozygosity patterns detected from different read evidences to identify false heterozygous deletions. Comprehensive benchmarking using 31 Human Genome Structural Variation Consortium Phase 3 samples demonstrated SVhet's ability to further reduce false positives while maintaining baseline recall. Hybrid approach of duphold and SVhet achieved up to 60% reduction in false positive counts while preserving recall. We also showed SVhet to be computationally efficient that can complete a whole genome structural variant callset under 5min using 4 CPU cores. SVhet is available under a permissive MIT license via https://github.com/snakesch/SVhet. SVhet provides an accurate and efficient solution for evaluating heterozygous deletions derived from short read sequencing data. SVhet can be used as a standalone tool or in conjunction with other filtering tools such as duphold. Importantly, it does not require additional variant sets, and can operate with minimal compute. Altogether, SVhet adds to the current effort to achieve accurate structural variant detection using short reads.

Twenty-two patients with biopsy proved carcinoma of the breast received radiolabeled MAb B723 (5–1 mCi of 125I per 2.0–0.25 mg MAb, Iodo-GenTM method) intravenously 6–30 days prior to definitive surgery (mastectomy or lumpectomy/axillary dissection). Using the Neoprobe 1000TM gamma detecting probe, gamma counts of breast and axillary tissues were obtained preoperatively, intraoperatively, and on ex vivo specimens. In breast tissue, the RIGSTM system identified tumor that was histopathologically confirmed in 11 of 14 patients. There were two false positives each having a histopathologic diagnosis of apocrine metaplasia and hyperplasia. Unsuspected tumor was histopathologically documented in 3 of 6 breast biopsies performed based on the preoperative presence of high external gamma counts in the countralateral breast or in a quadrant other than that of the original primary All six patients had negative mammograms and physical exams. The 3 false positives had diagnoses of aprocrine metaplasia and hyperplasia. In axillary tissue, probe counts identified metastatic disease in 3 of 8 patients and verified absence of disease in 10 of 14 patients. False positive counts were obtained in 4 having histopathologic diagnoses of sinus histocytosis or reactive nodes. RIGS appears to be able to identify residual, subclinical, and multicentric carcinoma of the breast and delineate the pattern of antigenic drainage of tumor into lymph nodes.

In this digital world, numerous credit card-based transactions take place all over the world. Concomitantly, gaps in process flows and technology result in many fraudulent transactions. Owing to the spurt in the number of reported fraudulent transactions, customers and credit card service providers incur significant financial and reputation losses respectively. Therefore, building a powerful fraud detection system is paramount. It is noteworthy that fraud detection datasets, by nature, are highly unbalanced. Consequently, almost all of the supervised classifiers, when built on the unbalanced datasets, yield high false negative rates. But, the extant oversampling methods while reducing the false negatives, increase the false positives. In this paper, we propose a novel data oversampling method using Generative Adversarial Network (GAN). We use GAN and its variant to generate synthetic data of fraudulent transactions. To evaluate the effectiveness of the proposed method, we employ machine learning classifiers on the data balanced by GAN. Our proposed GAN-based oversampling method simultaneously achieved high precision, F1-score and dramatic reduction in the count of false positives compared to the state-of-the-art synthetic data generation based oversampling methods such as Synthetic Minority Oversampling Technique (SMOTE), Adaptive Synthetic Sampling (ADASYN) and random oversampling. Moreover, an ablation study involving the oversampling based on the ensemble of SMOTE and GAN/WGAN generated datasets indicated that it is outperformed by the proposed methods in terms of F1 score and false positive count.

Manufacturers of step counting devices apply filters to their step counting algorithms to prevent accumulation of steps when none are taken (i.e. false positives). However because filters prevent steps from being recorded during short, intermittent walking bouts, it is possible that these filters may be a source of error. Since few manufacturers disclose the type of filter they use, we decided to investigate this topic. PURPOSE: To determine whether the devices used in this study have a filter, and to describe the effects of the filter on short, intermittent walking bouts with varied walk and pause durations. METHODS: In Parts A and B, 20 participants performed intermittent walking bouts for 2 min, at a cadence of 100 steps/min. In Part A participants were instructed to walk a certain number of steps (i.e. 4, 6, 8, 10, and 12) followed by a 10-sec pause and repeat this until the trial ended. In Part B participants were instructed to walk four steps followed by various pause intervals (i.e. 8, 6, 4, 2, and 1 sec) and repeat this. A researcher counted steps using a hand-tally device (criterion). “Percent of actual steps taken” was used for statistical analysis. A one-way repeated measures ANOVA was completed for both parts. In the case of significant overall effects (p < 0.05), the results were further examined using planned contrasts to see which conditions differed from the criterion. RESULTS: In Parts A and B the multivariate results for ActiGraph GT3X (AG) (without low frequency extension) worn at the wrist, StepWatch 3, and Yamax Digi-Walker SW-200 were not significantly different from the criterion, indicating absence of a step count filter. Walking bouts shorter than 4 steps (AG at the hip), 6 steps (Withings), 8 steps (Omron and Garmin Vivofit 2), and 12 steps (Polar A360), resulted in a significant decrease in the number of steps counted, indicating presence of a filter. The minimum pause needed to break up a walking bout was 1 sec (Fitbit Charge, Fitbit Zip, and Withings), and < 1 sec (Omron HJ-322U). For both the Polar and Garmin, the longer the pause, the less likely they were to record steps. CONCLUSIONS: Devices with step count filters will contribute to error in daily step counts because steps taken during short, intermittent walking bouts (e.g., meal preparation, and housework) are not registered.

This study evaluated the effect of flight altitude and canopy separation of container-grown Fire Chief™ arborvitae (Thuja occidentalis L.) on counting accuracy. Images were taken at 6, 12, and 22 m above the ground using unmanned aircraft systems. Plants were spaced to achieve three canopy separation treatments: 5 cm between canopy edges, canopy edges touching, and 5 cm of canopy edge overlap. Plants were placed on two different ground covers: black fabric and gravel. A counting algorithm was trained using Feature Analyst®. Total counting error, false positives, and unidentified plants were reported for images analyzed. In general, total counting error was smaller when plants were fully separated. The effect of ground cover on counting accuracy varied with the counting algorithm. Total counting error for plants placed on gravel (−8) was larger than for those on a black fabric (−2), however, false positive counts were similar for black fabric (6) and gravel (6). Nevertheless, output images of plants placed on gravel did not show a negative effect due to the ground cover but was impacted by differences in image spatial resolution.

We conduct experiments that show the Area Under the Precision Recall Curve (AUPRC) metric provides a more meaningful insight into the impact of Random Undersampling than Area Under the Receiver Operating Characteristic Curve (AUC). Evaluating experiments with multiple metrics is a robust method for overcoming challenges in Machine Learning, such as class imbalance. Random Undersampling is a technique to deal with class imbalance. We find Random Undersampling may provide an improvement to AUC scores. However, at the same time, Random Undersampling may be detrimental to AUPRC scores. AUPRC is a metric that involves precision, whereas AUC does not. In the classification of imbalanced Big Data, an increase in false positive counts has a more noticeable drop in precision scores. Therefore, in application domains where false positives are undesirable, optimizing models for AUPRC is a wise choice. Our contribution is to compare the performance of models in terms of AUPRC and AUC to show the impact of Random Undersampling on the classification of imbalanced Big Data. We compare the performance via experiments in the classification of highly imbalanced Big Data. Models are built with data in its original class ratio, and with data undersampled into 5 distinct class ratios. We report the results of 600 experiments where we apply Random Undersampling to a dataset with about 175 million instances. To the best of our knowledge we are the first to utilize Medicare Part D data which became available in 2021.

BACKGROUND: Given the efficacy of pedometer-based programs to improve physical activity and health status in various clinical sub-populations, implementation of similar programs for persons with vision loss should be considered. Due to alterations in the gait patterns of persons with vision loss, factors such as environmental familiarity and mobility aids may influence results of pedometer validation trials. PURPOSE: The purpose of this study was to establish validity evidence for the Centrios talking pedometer relative to walking environment and mobility aid use in adults with vision loss. METHODS: Fifteen adults with legal blindness (10 females; age = 38 ± 14 years; BMI = 26.5 ± 4.2 kg·m2) completed two walking sessions over an unfamiliar, quarter-mile walking course while wearing a pedometer at the right and left hip. Walking speed, pedometer-determined step counts, and actual step counts were recorded during the first session, which represented walking in an unfamiliar environment. Following the completion of additional walking trials over the same route, outcome measures were reassessed during a second walking session over the course, which reflected walking in a "familiar environment." To obtain validity evidence for the talking pedometer, absolute percent error (APE) was calculated between actual and pedometer-determined steps. Paired t-tests were used to assess differences in APE relative to mounting position (right hip vs. left hip; mobility aid side vs. non-mobility aid side) across unfamiliar and familiar walking trials. RESULTS: The Centrios pedometer accurately reported step counts during unfamiliar (average speed = 3.2 mph) and familiar (average speed = 3.7 mph) walking trials when mounted at the hip opposite the user's mobility aid (APE = 3.1% and 2.4%, respectively), but was less accurate on the mobility aid side during unfamiliar and familiar walking (APE = 9.2% and 6.3%, respectively). Paired t-tests revealed no significant differences (p >.05) in APE relative to mounting position and familiarity with the environment. CONCLUSIONS: During walking in familiar and unfamiliar settings, the Centrios talking pedometer provides an accurate step count in adults with visual impairment when mounted at the hip opposite of the user's mobility aid.

Validity of FitBit, Jawbone UP, Nike+ and other wearable devices for level and stair walking

MHealth devices demonstrate validity and reliability in detecting steps in chronic stroke survivors who rely on assistive devices