Calibration and validation of accelerometer-based activity monitors: A systematic review of machine-learning approaches

Quantifying Asymmetric Gait Pattern Changes Using a Hidden Markov Model Similarity Measure (HMM-SM) on Inertial Sensor Signals

Wearable gait analysis systems using inertial sensors offer the potential for easy-to-use gait assessment in lab and free-living environments. This can enable objective long-term monitoring and decision making for individuals with gait disabilities. This study explores a novel approach that applies a hidden Markov model-based similarity measure (HMM-SM) to assess changes in gait patterns based on the gyroscope and accelerometer signals from just one or two inertial sensors. Eleven able-bodied individuals were equipped with a system which perturbed gait patterns by manipulating stance-time symmetry. Inertial sensor data were collected from various locations on the lower body to train hidden Markov models. The HMM-SM was evaluated to determine whether it corresponded to changes in gait as individuals deviated from their baseline, and whether it could provide a reliable measure of gait similarity. The HMM-SM showed consistent changes in accordance with stance-time symmetry in the following sensor configurations: pelvis, combined upper leg signals, and combined lower leg signals. Additionally, the HMM-SM demonstrated good reliability for the combined upper leg signals (ICC = 0.803) and lower leg signals (ICC = 0.795). These findings provide preliminary evidence that the HMM-SM could be useful in assessing changes in overall gait patterns. This could enable the development of compact, wearable systems for unsupervised gait assessment, without the requirement to pre-identify and measure a set of gait parameters.

Validation of two novel human activity recognition models for typically developing children and children with Cerebral Palsy

Human Activity Recognition models have potential to contribute to valuable and detailed knowledge of habitual physical activity for typically developing children and children with Cerebral Palsy. The main objective of the present study was to develop and validate two Human Activity Recognition models. One trained on data from typically developing children (n = 63), the second also including data from children with Cerebral Palsy (n = 16), engaging in standardised activities and free play. Our data was collected using accelerometers and ground truth was established with video annotations. Additionally, we aimed to investigate the influence of window settings on model performance. Utilizing the Extreme gradient boost (XGBoost) classifier, twelve sub-models were created, with 1-,3- and 5-seconds windows, with and without overlap. Both Human Activity Recognition models demonstrated excellent predictive capabilities (>92%) for standardised activities for both typically developing and Cerebral Palsy. From all window sizes, the 1-second window performed best for all test groups. Accuracy was slightly lower (>75%) for the Cerebral Palsy test group performing free play activities. The impact of window size and overlap varied depending on activity. In summary both Human Activity Recognition models effectively predict standardised activities, surpassing prior models for typically developing and children with Cerebral Palsy. Notably, the model trained on combined typically developing children and Cerebral Palsy data performed exemplary across all test groups. Researchers should select window settings aligned with their specific research objectives.

Accelerometer techniques for capturing human movement validated against direct observation: a scoping review

Objective.Accelerometers are devices commonly used to measure human physical activity and sedentary time. Accelerometer capabilities and analytical techniques have evolved rapidly, making it difficult for researchers to keep track of advances and best practices for data processing and analysis. The objective of this scoping review is to determine the existing methods for analyzing accelerometer data for capturing human movement which have been validated against the criterion measure of direct observation.Approach.This scoping review searched 14 academic and 5 grey databases. Two independent raters screened by title and abstract, then full text. Data were extracted using Microsoft Excel and checked by an independent reviewer.Mainresults.The search yielded 1039 papers and the final analysis included 115 papers. A total of 71 unique accelerometer models were used across a total of 4217 participants. While all studies underwent validation from direct observation, most direct observation occurred live (55%) or using recordings (42%). Analysis techniques included machine learning (ML) approaches (22%), the use of existing cut-points (18%), receiver operating characteristic curves to determine cut-points (14%), and other strategies including regressions and non-ML algorithms (8%).Significance.ML techniques are becoming more prevalent and are often used for activity identification. Cut-point methods are still frequently used. Activity intensity is the most assessed activity outcome; however, both the analyses and outcomes assessed vary by wear location. This scoping review provides a comprehensive overview of accelerometer analysis and validation techniques using direct observation and is a useful tool for researchers using accelerometers.

Joint Associations of Diet and Device-Measured Physical Activity with Mortality and Incident CVD and Cancer: A Prospective Analysis of the UK Biobank Study

BACKGROUND

We examined the joint associations of diet and device-measured intensity-specific physical activity (PA) with all-cause mortality (ACM), cardiovascular disease (CVD), and cancer incidence.

METHODS

We used data from 79,988 participants from the UK Biobank, a population-based prospective cohort study. Light PA (LPA), moderate-to-vigorous PA (MVPA), vigorous PA (VPA), and total PA (TPA) were measured using a wrist-worn accelerometer. Diet quality score (DQS) was based on 10 foods and ranged from 0 (unhealthiest) to 100 (healthiest) points. We derived joint PA and diet variables. Outcomes were ACM, CVD, and cancer incidence including PA, diet and adiposity-related (PDAR) cancer.

RESULTS

During a median follow-up of 8 years, 2,863 deaths occurred, 11,053 participants developed CVD, 7,005 developed cancer, and 3,400 developed PDAR cancer. Compared with the least favorable referent group (bottom PA tertile/low DQS), participants with middle and high (total and intensity specific) PA, except for LPA, had lower ACM risk and incident CVD risk, regardless of DQS. For example, among middle and high VPA and high DQS groups, CVD HR were 0.79 (95% CI, 0.74-0.86) and 0.75 (95% CI, 0.69-0.82), respectively. The pattern of cancer results was less pronounced but in agreement with the ACM and CVD incidence findings (e.g., HR, 0.90, 95% CI, 0.81-0.99; 0.88, 0.79-0.98; and 0.82, 0.74-0.92 among high VPA for low, moderate, and high DQS groups, respectively).

CONCLUSIONS

Device-measured PA reveals novel joint associations with diet on health outcomes.

IMPACT

Our results emphasize the crucial role of PA in addition to a healthy diet for reducing chronic diseases and mortality risk.

Associations of accelerometer-estimated free-living daily activity impact intensities with 10-year probability of osteoporotic fractures in adults

PURPOSE

Accelerometers are used to objectively measure physical activity; however, the relationship between accelerometer-based activity parameters and bone health is not well understood. This study examines the association between accelerometer-estimated daily activity impact intensities and future risk estimates of major osteoporotic fractures in a large population-based cohort.

METHODS

Participants were 3165 adults 46 years of age from the Northern Finland Birth Cohort 1966 who agreed to wear a hip-worn accelerometer during all waking hours for 14 consecutive days. Raw accelerometer data were converted to resultant acceleration. Impact magnitude peaks were extracted and divided into 32 intensity bands, and the osteogenic index (OI) was calculated to assess the osteogenic effectiveness of various activities. Additionally, the impact peaks were categorized into three separate impact intensity categories (low, medium, and high). The 10-year probabilities of hip and all major osteoporotic fractures were estimated with FRAX-tool using clinical and questionnaire data in combination with body mass index collected at the age of 46 years. The associations of daily activity impact intensities with 10-year fracture probabilities were examined using three statistical approaches: multiple linear regression, partial correlation, and partial least squares (PLS) regression.

RESULTS

On average, participants' various levels of impact were 8331 (SD = 3478) low; 2032 (1248) medium; and 1295 (1468) high impacts per day. All three statistical approaches found a significant positive association between the daily number of low-intensity impacts and 10-year probability of hip and all major osteoporotic fractures. In contrast, increased number of moderate to very high daily activity impacts was associated with a lower probability of future osteoporotic fractures. A higher OI was also associated with a lower probability of future major osteoporotic fractures.

CONCLUSION

Low-intensity impacts might not be sufficient for reducing fracture risk in middle-aged adults, while high-intensity impacts could be beneficial for preventing major osteoporotic fractures.

Development and validation of a smartwatch algorithm for differentiating physical activity intensity in health monitoring

To develop and validate a machine learning based algorithm to estimate physical activity (PA) intensity using the smartwatch with the capacity to record PA and determine outdoor state. Two groups of participants, including 24 adults (13 males) and 18 children (9 boys), completed a sequential activity trial. During each trial, participants wore a smartwatch, and energy expenditure was measured using indirect calorimetry as gold standard. The support vector machine algorithm and the least squares regression model were applied for the metabolic equivalent (MET) estimation using raw data derived from the smartwatch. Exercise intensity was categorized based on MET values into sedentary activity (SED), light activity (LPA), moderate activity (MPA), and vigorous activity (VPA). The classification accuracy was evaluated using area under the ROC curve (AUC). The METs estimation accuracy were assessed via the mean absolute error (MAE), the correlation coefficient, Bland-Altman plots, and intraclass correlation (ICC). A total of 24 adults aged 21-34 years and 18 children aged 9-13 years participated in the study, yielding 1790 and 1246 data points for adults and children respectively for model building and validation. For adults, the AUC for classifying SED, MVPA, and VPA were 0.96, 0.88, and 0.86, respectively. The MAE between true METs and estimated METs was 0.75 METs. The correlation coefficient and ICC were 0.87 (p < 0.001) and 0.89, respectively. For children, comparable levels of accuracy were demonstrated, with the AUC for SED, MVPA, and VPA being 0.98, 0.89, and 0.85, respectively. The MAE between true METs and estimated METs was 0.80 METs. The correlation coefficient and ICC were 0.79 (p < 0.001) and 0.84, respectively. The developed model successfully estimated PA intensity with high accuracy in both adults and children. The application of this model enables independent investigation of PA intensity, facilitating research in health monitoring and potentially in areas such as myopia prevention and control.

Development of a multi-wear-site, deep learning-based physical activity intensity classification algorithm using raw acceleration data

BACKGROUND

Accelerometers are widely adopted in research and consumer devices as a tool to measure physical activity. However, existing algorithms used to estimate activity intensity are wear-site-specific. Non-compliance to wear instructions may lead to misspecifications. In this study, we developed deep neural network models to classify device placement and activity intensity based on raw acceleration data. Performances of these models were evaluated by making comparisons to the ground truth and results derived from existing count-based algorithms.

METHODS

54 participants (26 adults 26.9±8.7 years; 28 children, 12.1±2.3 years) completed a series of activity tasks in a laboratory with accelerometers attached to each of their hip, wrist, and chest. Their metabolic rates at rest and during activity periods were measured using the portable COSMED K5; data were then converted to metabolic equivalents, and used as the ground truth for activity intensity. Deep neutral networks using the Long Short-Term Memory approach were trained and evaluated based on raw acceleration data collected from accelerometers. Models to classify wear-site and activity intensity, respectively, were evaluated.

RESULTS

The trained models correctly classified wear-sites and activity intensities over 90% of the time, which outperformed count-based algorithms (wear-site correctly specified: 83% to 85%; wear-site misspecified: 64% to 75%). When additional parameters of age, height and weight of participants were specified, the accuracy of some prediction models surpassed 95%.

CONCLUSIONS

Results of the study suggest that accelerometer placement could be determined prospectively, and non-wear-site-specific algorithms had satisfactory accuracies. The performances, in terms of intensity classification, of these models also exceeded typical count-based algorithms. Without being restricted to one specific wear-site, research protocols for accelerometers wear could allow more autonomy to participants, which may in turn improve their acceptance and compliance to wear protocols, and in turn more accurate results.

Emerging methods for measuring physical activity using accelerometry in children and adolescents with neuromotor disorders: a narrative review

BACKGROUND

Children and adolescents with neuromotor disorders need regular physical activity to maintain optimal health and functional independence throughout their development. To this end, reliable measures of physical activity are integral to both assessing habitual physical activity and testing the efficacy of the many interventions designed to increase physical activity in these children. Wearable accelerometers have been used for children with neuromotor disorders for decades; however, studies most often use disorder-specific cut points to categorize physical activity intensity, which lack generalizability to a free-living environment. No reviews of accelerometer data processing methods have discussed the novel use of machine learning techniques for monitoring physical activity in children with neuromotor disorders.

METHODS

In this narrative review, we discuss traditional measures of physical activity (including questionnaires and objective accelerometry measures), the limitations of standard analysis for accelerometry in this unique population, and the potential benefits of applying machine learning approaches. We also provide recommendations for using machine learning approaches to monitor physical activity.

CONCLUSIONS

While wearable accelerometers provided a much-needed method to quantify physical activity, standard cut point analyses have limitations in children with neuromotor disorders. Machine learning models are a more robust method of analyzing accelerometer data in pediatric neuromotor disorders and using these methods over disorder-specific cut points is likely to improve accuracy of classifying both type and intensity of physical activity. Notably, there remains a critical need for further development of classifiers for children with more severe motor impairments, preschool aged children, and children in hospital settings.

Effectiveness of digital home rehabilitation and supervision for stroke survivors: A systematic review and meta-analysis

Objective

Stroke survivors often experience residual impairments and motor decline post-discharge. While digital home rehabilitation combined with supervision could be a promising approach for reducing human resources, increasing motor ability, and supporting rehabilitation persistence there is a lack of reviews synthesizing the effects. Thus, this systematic review and meta-analysis aimed to synthesize the effect of digital home rehabilitation and supervision in improving motor ability of upper limb, static balance, stroke-related quality of life, and self-reported arm function among stroke survivors.

Methods

Six electronic databases, grey literature, ongoing studies, and reference lists were searched for relevant studies. Two investigators independently reviewed titles, abstracts, screened full texts for eligibility and performed data extraction. Meta-analysis of 13 independent studies were grouped into four separate meta-analyses. The Grading of Recommendations, Assessments, Development and Evaluations (GRADE) tool was used for evaluating the overall quality of the evidence.

Results

Meta-analyses showed no statistically significant difference between intervention (digital home rehabilitation) and control groups (home training/clinic-based) of all outcomes including motor ability of upper limb, static balance, stroke-related quality of life, and self-reported arm function. In the sub-group analysis digital home rehabilitation was associated with better quality of arm use (standardized mean difference = 0.68, 95% confidence interval: [0.27, 1.09], p = 0.001).

Conclusions

This result indicated that digital home rehabilitation has similar effects and could potentially replace home training or clinic-based services. This review highlights better-targeted digital motor interventions to examine the effects of interventions further. The quality of evidence was moderate to high in motor and self-reported arm outcomes, and low for balance and quality of life.

Applying machine learning to consumer wearable data for the early detection of complications after pediatric appendectomy

When children are discharged from the hospital after surgery, their caregivers often rely on subjective assessments (e.g., appetite, fatigue) to monitor postoperative recovery as objective assessment tools are scarce at home. Such imprecise and one-dimensional evaluations can result in unwarranted emergency department visits or delayed care. To address this gap in postoperative monitoring, we evaluated the ability of a consumer-grade wearable device, Fitbit, which records multimodal data about daily physical activity, heart rate, and sleep, in detecting abnormal recovery early in children recovering after appendectomy. One hundred and sixty-two children, ages 3-17 years old, who underwent an appendectomy (86 complicated and 76 simple cases of appendicitis) wore a Fitbit device on their wrist for 21 days postoperatively. Abnormal recovery events (i.e., abnormal symptoms or confirmed postoperative complications) that arose during this period were gathered from medical records and patient reports. Fitbit-derived measures, as well as demographic and clinical characteristics, were used to train machine learning models to retrospectively detect abnormal recovery in the two days leading up to the event for patients with complicated and simple appendicitis. A balanced random forest classifier accurately detected 83% of these abnormal recovery days in complicated appendicitis and 70% of abnormal recovery days in simple appendicitis prior to the true report of a symptom/complication. These results support the development of machine learning algorithms to predict onset of abnormal symptoms and complications in children undergoing surgery, and the use of consumer wearables as monitoring tools for early detection of postoperative events.

Predicting energy intake with an accelerometer-based intake-balance method

Nutritional interventions often rely on subjective assessments of energy intake (EI), but these are susceptible to measurement error. To introduce an accelerometer-based intake-balance method for assessing EI using data from a time-restricted eating (TRE) trial. Nineteen participants with overweight/obesity (25-63 years old; 16 females) completed a 12-week intervention (NCT03129581) in a control group (unrestricted feeding; n 8) or TRE group (n 11). At the start and end of the intervention, body composition was assessed by dual-energy X-ray absorptiometry (DXA) and daily energy expenditure (EE) was assessed for 2 weeks via wrist-worn accelerometer. EI was back-calculated as the sum of net energy storage (from DXA) and EE (from accelerometer). Accelerometer-derived EI estimates were compared against estimates from the body weight planner of the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK). Mean EI for the control group declined by 138 and 435 kJ/day for the accelerometer and NIDDK methods, respectively (both P ≥ 0·38), v. 1255 and 1469 kJ/day, respectively, for the TRE group (both P < 0·01). At follow-up, the accelerometer and NIDDK methods showed excellent group-level agreement (mean bias of -297 kJ/day across arms; standard error of estimate 1054 kJ/day) but high variability at the individual level (limits of agreement from -2414 to +1824 kJ/day). The accelerometer-based intake-balance method showed plausible sensitivity to change, and EI estimates were biologically and behaviourally plausible. The method may be a viable alternative to self-report EI measures. Future studies should assess criterion validity using doubly labelled water.

Assessment of 24-hour physical behaviour in adults via wearables: a systematic review of validation studies under laboratory conditions

BACKGROUND

Wearable technology is used by consumers and researchers worldwide for continuous activity monitoring in daily life. Results of high-quality laboratory-based validation studies enable us to make a guided decision on which study to rely on and which device to use. However, reviews in adults that focus on the quality of existing laboratory studies are missing.

METHODS

We conducted a systematic review of wearable validation studies with adults. Eligibility criteria were: (i) study under laboratory conditions with humans (age ≥ 18 years); (ii) validated device outcome must belong to one dimension of the 24-hour physical behavior construct (i.e., intensity, posture/activity type, and biological state); (iii) study protocol must include a criterion measure; (iv) study had to be published in a peer-reviewed English language journal. Studies were identified via a systematic search in five electronic databases as well as back- and forward citation searches. The risk of bias was assessed based on the QUADAS-2 tool with eight signaling questions.

RESULTS

Out of 13,285 unique search results, 545 published articles between 1994 and 2022 were included. Most studies (73.8% (N = 420)) validated an intensity measure outcome such as energy expenditure; only 14% (N = 80) and 12.2% (N = 70) of studies validated biological state or posture/activity type outcomes, respectively. Most protocols validated wearables in healthy adults between 18 and 65 years. Most wearables were only validated once. Further, we identified six wearables (i.e., ActiGraph GT3X+, ActiGraph GT9X, Apple Watch 2, Axivity AX3, Fitbit Charge 2, Fitbit, and GENEActiv) that had been used to validate outcomes from all three dimensions, but none of them were consistently ranked with moderate to high validity. Risk of bias assessment resulted in 4.4% (N = 24) of all studies being classified as "low risk", while 16.5% (N = 90) were classified as "some concerns" and 79.1% (N = 431) as "high risk".

CONCLUSION

Laboratory validation studies of wearables assessing physical behaviour in adults are characterized by low methodological quality, large variability in design, and a focus on intensity. Future research should more strongly aim at all components of the 24-hour physical behaviour construct, and strive for standardized protocols embedded in a validation framework.

Sensitivity and specificity of measuring children's free-living cycling with a thigh-worn Fibion® accelerometer

Objective

Cycling is an important part of children's active travel, but its measurement using accelerometry is a challenge. The aim of the present study was to evaluate physical activity duration and intensity, and sensitivity and specificity of free-living cycling measured with a thigh-worn accelerometer.

Methods

Participants were 160 children (44 boys) aged 11.5 ± 0.9 years who wore a triaxial Fibion® accelerometer on right thigh for 8 days, 24 h per day, and reported start time and duration of all cycling, walking and car trips to a travel log. Linear mixed effects models were used to predict and compare Fibion-measured activity and moderate-to-vigorous activity duration, cycling duration and metabolic equivalents (METs) between the travel types. Sensitivity and specificity of cycling bouts during cycling trips as compared to walking and car trips was also evaluated.

Results

Children reported a total of 1,049 cycling trips (mean 7.08 ± 4.58 trips per child), 379 walking trips (3.08 ± 2.81) and 716 car trips (4.79 ± 3.96). There was no difference in activity and moderate-to-vigorous activity duration (p > .105), a lower cycling duration (-1.83 min, p < .001), and a higher MET-level (0.95, p < .001) during walking trips as compared to cycling trips. Both activity (-4.54 min, p < .001), moderate-to-vigorous activity (-3.60 min, p < .001), cycling duration (-1.74 min, p < .001) and MET-level (-0.99, p < .001) were lower during car trips as compared to cycling trips. Fibion showed the sensitivity of 72.2% and specificity of 81.9% for measuring cycling activity type during the reported cycling trips as compared to walking and car trips when the minimum required duration for cycling was less than 29 s.

Conclusions

The thigh-worn Fibion® accelerometer measured a greater duration of cycling, a lower MET-level, and a similar duration of total activity and moderate-to-vigorous activity during free-living cycling trips as compared to walking trips, suggesting it can be used to measure free-living cycling activity and moderate-to-vigorous activity duration in 10-12-year-old children.

Validation of an Algorithm for Measurement of Sedentary Behaviour in Community-Dwelling Older Adults

Accurate measurement of sedentary behaviour in older adults is informative and relevant. Yet, activities such as sitting are not accurately distinguished from non-sedentary activities (e.g., upright activities), especially in real-world conditions. This study examines the accuracy of a novel algorithm to identify sitting, lying, and upright activities in community-dwelling older people in real-world conditions. Eighteen older adults wore a single triaxial accelerometer with an onboard triaxial gyroscope on their lower back and performed a range of scripted and non-scripted activities in their homes/retirement villages whilst being videoed. A novel algorithm was developed to identify sitting, lying, and upright activities. The algorithm's sensitivity, specificity, positive predictive value, and negative predictive value for identifying scripted sitting activities ranged from 76.9% to 94.8%. For scripted lying activities: 70.4% to 95.7%. For scripted upright activities: 75.9% to 93.1%. For non-scripted sitting activities: 92.3% to 99.5%. No non-scripted lying activities were captured. For non-scripted upright activities: 94.3% to 99.5%. The algorithm could, at worst, overestimate or underestimate sedentary behaviour bouts by ±40 s, which is within a 5% error for sedentary behaviour bouts. These results indicate good to excellent agreement for the novel algorithm, providing a valid measure of sedentary behaviour in community-dwelling older adults.

Relationship between physical activity and central sensitization in chronic low back pain: Insights from machine learning

BACKGROUND AND OBJECTIVES

Chronic low back pain (CLBP) is a leading cause of disability. The management guidelines for the management of CLBP often recommend optimizing physical activity (PA). Among a subsample of patients with CLBP, central sensitization (CS) is present. However, knowledge about the association between PA intensity patterns, CLBP, and CS is limited. The objective PA computed by conventional approaches (e.g. cut-points) may not be sensitive enough to explore this association. This study aimed to investigate PA intensity patterns in patients with CLBP and low or high CS (CLBP-, CLBP+, respectively) by using advanced unsupervised machine learning approach, Hidden semi-Markov model (HSMM).

METHODS

Forty-two patients were included (23 CLBP-, 19 CLBP+). CS-related symptoms (e.g. fatigue, sensitivity to light, psychological features) were assessed by a CS Inventory. Patients wore a standard 3D-accelerometer for one week and PA was recorded. The conventional cut-points approach was used to compute the time accumulation and distribution of PA intensity levels in a day. For the two groups, two HSMMs were developed to measure the temporal organization of and transition between hidden states (PA intensity levels), based on the accelerometer vector magnitude.

RESULTS

Based on the conventional cut-points approach, no significant differences were found between CLBP- and CLBP+ groups (p = 0.87). In contrast, HSMMs revealed significant differences between the two groups. For the 5 identified hidden states (rest, sedentary, light PA, light locomotion, and moderate-vigorous PA), the CLBP- group had a higher transition probability from rest, light PA, and moderate-vigorous PA states to the sedentary state (p < 0.001). In addition, the CBLP- group had a significantly shorter bout duration of the sedentary state (p < 0.001). The CLBP+ group exhibited longer durations of active (p < 0.001) and inactive states (p = 0.037) and had higher transition probabilities between active states (p < 0.001).

CONCLUSIONS

HSMM discloses the temporal organization and transitions of PA intensity levels based on accelerometer data, yielding valuable and detailed clinical information. The results imply that patients with CLBP- and CLBP+ have different PA intensity patterns. CLBP+ patients may adopt the distress-endurance response pattern with a prolonged bout duration of activity engagement.

AccNet24: A deep learning framework for classifying 24-hour activity behaviours from wrist-worn accelerometer data under free-living environments

OBJECTIVE

Although machine learning techniques have been repeatedly used for activity prediction from wearable devices, accurate classification of 24-hour activity behaviour categories from accelerometry data remains a challenge. We developed and validated a deep learning-based framework for classifying 24-hour activity behaviours from wrist-worn accelerometers.

METHODS

Using an openly available dataset with free-living wrist-based raw accelerometry data from 151 participants (aged 18-91 years), we developed a deep learning framework named AccNet24 to classify 24-hour activity behaviours. First, the acceleration signal (x, y, and z-axes) was segmented into 30-second nonoverlapping windows, and signal-to-image conversion was performed for each segment. Deep features were automatically extracted from the signal images using transfer learning and transformed into a lower-dimensional feature space. These transformed features were then employed to classify the activity behaviours as sleep, sedentary behaviour, and light-intensity (LPA) and moderate-to-vigorous physical activity (MVPA) using a bidirectional long short-term memory (BiLSTM) recurrent neural network. AccNet24 was trained and validated with data from 101 and 25 randomly selected participants and tested with the remaining unseen 25 participants. We also extracted 112 hand-crafted time and frequency domain features from 30-second windows and used them as inputs to five commonly used machine learning classifiers, including random forest, support vector machines, artificial neural networks, decision tree, and naïve Bayes to classify the 24-hour activity behaviour categories.

RESULTS

Using the same training, validation, and test data and window size, the classification accuracy of AccNet24 outperformed the accuracy of the other five machine learning classification algorithms by 16%-30% on unseen data.

CONCLUSION

AccNet24, relying on signal-to-image conversion, deep feature extraction, and BiLSTM achieved consistently high accuracy (>95 %) in classifying the 24-hour activity behaviour categories as sleep, sedentary, LPA, and MVPA. The next generation accelerometry analytics may rely on deep learning techniques for activity prediction.

Examination of physical activity development in early childhood: protocol for a longitudinal cohort study of mother-toddler dyads

BACKGROUND

Physical activity (PA) development in toddlers (age 1 and 2 years) is not well understood, partly because of a lack of analytic tools for accelerometer-based data processing that can accurately evaluate PA among toddlers. This has led to a knowledge gap regarding how parenting practices around PA, mothers' PA level, mothers' parenting stress, and child developmental and behavioral problems influence PA development in early childhood.

METHODS

The Child and Mother Physical Activity Study is a longitudinal study to observe PA development in toddlerhood and examine the influence of personal and parental characteristics on PA development. The study is designed to refine and validate an accelerometer-based machine learning algorithm for toddler activity recognition (Aim 1), apply the algorithm to compare the trajectories of toddler PA levels in males and females age 1-3 years (Aim 2), and explore the association between gross motor development and PA development in toddlerhood, as well as how parenting practices around PA, mothers' PA, mothers' parenting stress, and child developmental and behavioral problems are associated with toddlerhood PA development (Exploratory Aims 3a-c).

DISCUSSION

This study will be one of the first to use longitudinal data to validate a machine learning activity recognition algorithm and apply the algorithm to quantify free-living ambulatory movement in toddlers. The study findings will help fill a significant methodological gap in toddler PA measurement and expand the body of knowledge on the factors influencing early childhood PA development.

Conceptual Structure and Current Trends in Artificial Intelligence, Machine Learning, and Deep Learning Research in Sports: A Bibliometric Review

Artificial intelligence and its subcategories of machine learning and deep learning are gaining increasing importance and attention in the context of sports research. This has also meant that the number of corresponding publications has become complex and unmanageably large in human terms. In the current state of the research field, there is a lack of bibliometric analysis, which would prove useful for obtaining insights into the large amounts of available literature. Therefore, the present work aims to identify important research issues, elucidate the conceptual structure of the research field, and unpack the evolutionary trends and the direction of hot topics regarding key themes in the research field of artificial intelligence in sports. Using the Scopus database, 1215 documents (reviews and articles) were selected. Bibliometric analysis was performed using VOSviewer and bibliometrix R package. The main findings are as follows: (a) the literature and research interest concerning AI and its subcategories is growing exponentially; (b) the top 20 most cited works comprise 32.52% of the total citations; (c) the top 10 journals are responsible for 28.64% of all published documents; (d) strong collaborative relationships are present, along with small, isolated collaboration networks of individual institutions; (e) the three most productive countries are China, the USA, and Germany; (f) different research themes can be characterized using author keywords with current trend topics, e.g., in the fields of biomechanics, injury prevention or prediction, new algorithms, and learning approaches. AI research activities in the fields of sports pedagogy, sports sociology, and sports economics seem to have played a subordinate role thus far. Overall, the findings of this study expand knowledge on the research situation as well as the development of research topics regarding the use of artificial intelligence in sports, and may guide researchers to identify currently relevant topics and gaps in the research.

Joint Profiles of Sedentary Time and Physical Activity in Adults and Their Associations with Cardiometabolic Health

PURPOSE

This study aimed to identify and characterize joint profiles of sedentary time and physical activity among adults and to investigate how these profiles are associated with markers of cardiometabolic health.

METHODS

The participants included 3702 of the Northern Finland Birth Cohort 1966 at age 46 yr, who wore a hip-worn accelerometer during waking hours and provided seven consecutive days of valid data. Sedentary time, light-intensity physical activity, and moderate- to vigorous-intensity physical activity on each valid day were obtained, and a data-driven clustering approach ("KmL3D") was used to characterize distinct joint profiles of sedentary time and physical activity intensities. Participants self-reported their sleep duration and performed a submaximal step test with continuous heart rate measurement to estimate their cardiorespiratory fitness (peak heart rate). Linear regression was used to determine the association between joint profiles of sedentary time and physical activities with cardiometabolic health markers, including adiposity markers and blood lipid, glucose, and insulin levels.

RESULTS

Four distinct groups were identified: "active couch potatoes" ( n = 1173), "sedentary light movers" ( n = 1199), "sedentary exercisers" ( n = 694), and "movers" ( n = 636). Although sufficiently active, active couch potatoes had the highest daily sedentary time (>10 h) and lowest light-intensity physical activity. Compared with active couch potatoes, sedentary light movers, sedentary exercisers, and movers spent less time in sedentary by performing more physical activity at light-intensity upward and had favorable differences in their cardiometabolic health markers after accounting for potential confounders (1.1%-25.0% lower values depending on the health marker and profile).

CONCLUSIONS

After accounting for sleep duration and cardiorespiratory fitness, waking activity profiles characterized by performing more physical activity at light-intensity upward, resulting in less time spent in sedentary, were associated with better cardiometabolic health.

Machine Learning Models for Weight-Bearing Activity Type Recognition Based on Accelerometry in Postmenopausal Women

Hip-worn triaxial accelerometers are widely used to assess physical activity in terms of energy expenditure. Methods for classification in terms of different types of activity of relevance to the skeleton in populations at risk of osteoporosis are not currently available. This publication aims to assess the accuracy of four machine learning models on binary (standing and walking) and tertiary (standing, walking, and jogging) classification tasks in postmenopausal women. Eighty women performed a shuttle test on an indoor track, of which thirty performed the same test on an indoor treadmill. The raw accelerometer data were pre-processed, converted into eighteen different features and then combined into nine unique feature sets. The four machine learning models were evaluated using three different validation methods. Using the leave-one-out validation method, the highest average accuracy for the binary classification model, 99.61%, was produced by a k-NN Manhattan classifier using a basic statistical feature set. For the tertiary classification model, the highest average accuracy, 94.04%, was produced by a k-NN Manhattan classifier using a feature set that included all 18 features. The methods and classifiers within this study can be applied to accelerometer data to more accurately characterize weight-bearing activity which are important to skeletal health.

Historical development of accelerometry measures and methods for physical activity and sedentary behavior research worldwide: A scoping review of observational studies of adults

This scoping review identified observational studies of adults that utilized accelerometry to assess physical activity and sedentary behavior. Key elements on accelerometry data collection were abstracted to describe current practices and completeness of reporting. We searched three databases (PubMed, Web of Science, and SPORTDiscus) on June 1, 2021 for articles published up to that date. We included studies of non-institutionalized adults with an analytic sample size of at least 500. The search returned 5686 unique records. After reviewing 1027 full-text publications, we identified and abstracted accelerometry characteristics on 155 unique observational studies (154 cross-sectional/cohort studies and 1 case control study). The countries with the highest number of studies included the United States, the United Kingdom, and Japan. Fewer studies were identified from the continent of Africa. Five of these studies were distributed donor studies, where participants connected their devices to an application and voluntarily shared data with researchers. Data collection occurred between 1999 to 2019. Most studies used one accelerometer (94.2%), but 8 studies (5.2%) used 2 accelerometers and 1 study (0.6%) used 4 accelerometers. Accelerometers were more commonly worn on the hip (48.4%) as compared to the wrist (22.3%), thigh (5.4%), other locations (14.9%), or not reported (9.0%). Overall, 12.7% of the accelerometers collected raw accelerations and 44.6% were worn for 24 hours/day throughout the collection period. The review identified 155 observational studies of adults that collected accelerometry, utilizing a wide range of accelerometer data processing methods. Researchers inconsistently reported key aspects of the process from collection to analysis, which needs addressing to support accurate comparisons across studies.

Predicting vertical ground reaction forces from 3D accelerometry using reservoir computers leads to accurate gait event detection

Accelerometers are low-cost measurement devices that can readily be used outside the lab. However, determining isolated gait events from accelerometer signals, especially foot-off events during running, is an open problem. We outline a two-step approach where machine learning serves to predict vertical ground reaction forces from accelerometer signals, followed by force-based event detection. We collected shank accelerometer signals and ground reaction forces from 21 adults during comfortable walking and running on an instrumented treadmill. We trained one common reservoir computer using segmented data using both walking and running data. Despite being trained on just a small number of strides, this reservoir computer predicted vertical ground reaction forces in continuous gait with high quality. The subsequent foot contact and foot off event detection proved highly accurate when compared to the gold standard based on co-registered ground reaction forces. Our proof-of-concept illustrates the capacity of combining accelerometry with machine learning for detecting isolated gait events irrespective of mode of locomotion.

Energy Expenditure Estimation in Children, Adolescents and Adults by Using a Respiratory Magnetometer Plethysmography System and a Deep Learning Model

PURPOSE

Energy expenditure is a key parameter in quantifying physical activity. Traditional methods are limited because they are expensive and cumbersome. Additional portable and cheaper devices are developed to estimate energy expenditure to overcome this problem. It is essential to verify the accuracy of these devices. This study aims to validate the accuracy of energy expenditure estimation by a respiratory magnetometer plethysmography system in children, adolescents and adults using a deep learning model.

METHODS

Twenty-three healthy subjects in three groups (nine adults (A), eight post-pubertal (PP) males and six pubertal (P) females) first sat or stood for six minutes and then performed a maximal graded test on a bicycle ergometer until exhaustion. We measured energy expenditure, oxygen uptake, ventilatory thresholds 1 and 2 and maximal oxygen uptake. The respiratory magnetometer plethysmography system measured four chest and abdomen distances using magnetometers sensors. We trained the models to predict energy expenditure based on the temporal convolutional networks model.

RESULTS

The respiratory magnetometer plethysmography system provided accurate energy expenditure estimation in groups A (R² = 0.98), PP (R² = 0.98) and P (R² = 0.97). The temporal convolutional networks model efficiently estimates energy expenditure under sitting, standing and high levels of exercise intensities.

CONCLUSION

Our results proved the respiratory magnetometer plethysmography system's effectiveness in estimating energy expenditure for different age populations across various intensities of physical activity.

A theory-based model of cumulative activity

Energy expenditure can be used to examine the health of individuals and the impact of environmental factors on physical activity. One of the more common ways to quantify energy expenditure is to process accelerometer data into some unit of measurement for this expenditure, such as Actigraph activity counts, and bin those measures into physical activity levels. However, accepted thresholds can vary between demographics, and some units of energy measurements do not currently have agreed upon thresholds. We present an approach which computes unique thresholds for each individual, using piecewise exponential functions to model the characteristics of their overall physical activity patterns corresponding to well established sedentary, light, moderate and vigorous activity levels from the literature. Models are fit using existing piecewise fitting techniques and software. Most participants’ activity intensity profile is exceptionally well modeled as piecewise exponential decay. Using this model, we find emergent groupings of participant behavior and categorize individuals into non-vigorous, consistent, moderately active, or extremely active activity intensity profiles. In the supplemental materials, we demonstrate that the parameters of the model correlate with demographics of age, household size, and level of education, inform behavior change under COVID lockdown, and are reasonably robust to signal frequency.

Technological evaluation of strategies to get out of bed by people with Parkinson's disease: Insights from multisite wearable sensors

Background

Difficulty getting out of bed is a common night-time and early morning manifestation of Parkinson's disease (PD), rated by 40% of the patients as their most concerning motor symptoms. However, current assessment methods are based on clinical interviews, video analysis, and clinical scales as objective outcome measures are not yet available.

Objective

To study the technical feasibility of multisite wearable sensors in the assessment of the supine-to-stand (STS) task as a determinant of the ability to get out of bed in patients with PD and age-matched control subjects, and develop relevant objective outcome measures.

Methods

The STS task was assessed in 32 patients with PD (mean Hoehn and Yahr; HY = 2.5) in the early morning before their first dopaminergic medication, and in 14 control subjects, using multisite wearable sensors (NIGHT-Recorder®; trunk, both wrists, and both ankles) in a sleep laboratory. Objective getting out of bed parameters included duration, onset, velocity and acceleration of truncal rotation, and angle deviation (a°) from the z-axis when subjects rose from the bed at different angles from the x-axis (10°, 15°, 30°, 45°, and 60°) as measures of truncal lateral flexion. Movement patterns were identified from the first body part or parts that moved. Correlation analysis was performed between these objective outcomes and standard clinical rating scales.

Results

Compared to control subjects, the duration of STS was significantly longer in patients with PD (p = 0.012), which is associated with a significantly slower velocity of truncal rotation (p = 0.003). Moderate and significant correlations were observed between the mean STS duration and age, and the Nocturnal Hypokinesia Questionnaire. The velocity of truncal rotation negatively and significantly correlated with HY staging. Any arm and leg moved together as the first movement significantly correlated with UPDRS-Axial and item #28. Several other correlations were also observed.

Conclusion

Our study was able to demonstrate the technical feasibility of using multisite wearable sensors to quantitatively assess early objective outcome measures of the ability of patients with PD to get out of bed, which significantly correlated with axial severity scores, suggesting that axial impairment could be a contributing factor in difficulty getting out of bed. Future studies are needed to refine these outcome measures for use in therapeutic trials related to nocturia or early morning akinesia in PD.

Methods for estimating physical activity and energy expenditure using raw accelerometry data or novel analytical approaches: a repository, framework, and reporting guidelines

The proliferation of approaches for analyzing accelerometer data using raw acceleration or novel analytic approaches like machine learning ('novel methods') outpaces their implementation in practice. This may be due to lack of accessibility, either because authors do not provide their developed models or because these models are difficult to find when included as supplementary material. Additionally, when access to a model is provided, authors may not include example data or instructions on how to use the model. This further hinders use by other researchers, particularly those who are not experts in statistics or writing computer code.

OBJECTIVE

We created a repository of novel methods of analyzing accelerometer data for the estimation of energy expenditure and/or physical activity intensity and a framework and reporting guidelines to guide future work.

APPROACH

Methods were identified from a recent scoping review. Available code, models, sample data, and instructions were compiled or created.

MAIN RESULTS

Sixty-three methods are hosted in the repository, in preschoolers (n=6), children/adolescents (n=20), and adults (n=42), using hip (n=45), wrist (n=25), thigh (n=4), chest (n=4), ankle (n=6), other (n=4), or a combination of monitor wear locations (n=9). Fifteen models are implemented in R, while 48 are provided as cut-points, equations, or decision trees.

SIGNIFICANCE

The developed tools should facilitate the use and development of novel methods for analyzing accelerometer data, thus improving data harmonization and consistency across studies. Future advances may involve including models that authors did not link to the original published article or those which identify activity type.

Accessibility and use of novel methods for predicting physical activity and energy expenditure using accelerometry: a scoping review

Use of raw acceleration data and/or "novel" analytic approaches like machine learning for physical activity measurement will not be widely implemented if methods are not accessible to researchers.

OBJECTIVE

This scoping review characterizes the validation approach, accessibility and use of novel analytic techniques for classifying energy expenditure and/or physical activity intensity using raw or count-based accelerometer data.

APPROACH

Three databases were searched for articles published between January 2000 and February 2021. Use of each method was coded from a list of citing articles compiled from Google Scholar. Authors' provision of access to the model (e.g., by request, sample code) was recorded.

MAIN RESULTS

Studies (N=168) included adults (n=143), and/or children (n=38). Model use ranged from 0 to 27 uses/year (average 0.83) with 101 models that have never been used. Approximately half of uses occurred in a free-living setting (52%) and/or by other authors (56%). Over half of included articles (n=107) did not provide complete access to their model. Sixty-one articles provided access to their method by including equations, coefficients, cut-points, or decision trees in the paper (n=48) and/or by providing access to code (n=13).

SIGNIFICANCE

The proliferation of approaches for analyzing accelerometer data outpaces the use of these models in practice. As less than half of the developed models are made accessible, it is unsurprising that so many models are not used by other researchers. We encourage researchers to make their models available and accessible for better harmonization of methods and improved capabilities for device-based physical activity measurement.

Cross-Sectional Associations of Sedentary Behavior and Sitting with Serum Lipid Biomarkers in Midlife

INTRODUCTION

Physical inactivity, excessive total time spent in sedentary behavior (SB) and prolonged sedentary bouts have been proposed to be risk factors for chronic disease morbidity and mortality worldwide. However, which patterns and postures of SB have the most negative impacts on health outcomes is still unclear. This population-based study aimed to investigate the independent associations of the patterns of accelerometer-based overall SB and sitting with serum lipid biomarkers at different moderate- to vigorous-intensity physical activity (MVPA) levels.

METHODS

Physical activity and SB were measured in a birth cohort sample ( N = 3272) at 46 yr using a triaxial hip-worn accelerometer in free-living conditions for 14 d. Raw acceleration data were classified into SB and PA using a machine learning-based model, and the bouts of overall SB and sitting were identified from the classified data. The participants also answered health-related questionnaires and participated in clinical examinations. Associations of overall SB (lying and sitting) and sitting patterns with serum lipid biomarkers were investigated using linear regression.

RESULTS

The overall SB patterns were more consistently associated with serum lipid biomarkers than the sitting patterns after adjustments. Among the participants with the least and the most MVPA, high total time spent in SB and SB bouts of 15-29.99 and ≥30 min were associated with impaired lipid metabolism. Among those with moderate amount of MVPA, higher time spent in SB and SB bouts of 15-29.99 min was unfavorably associated with serum lipid biomarkers.

CONCLUSIONS

The associations between SB patterns and serum lipid biomarkers were dependent on MVPA level, which should be considered when planning evidence-based interventions to decrease SB in midlife.

Monitoring Physical Behavior in Rehabilitation Using a Machine Learning-Based Algorithm for Thigh-Mounted Accelerometers: Development and Validation Study

BACKGROUND

Physical activity is emerging as an outcome measure. Accelerometers have become an important tool in monitoring physical behavior, and newer analytical approaches of recognition methods increase the degree of details. Many studies have achieved high performance in the classification of physical behaviors through the use of multiple wearable sensors; however, multiple wearables can be impractical and lower compliance.

OBJECTIVE

The aim of this study was to develop and validate an algorithm for classifying several daily physical behaviors using a single thigh-mounted accelerometer and a supervised machine-learning scheme.

METHODS

We collected training data by adding the behavior classes-running, cycling, stair climbing, wheelchair ambulation, and vehicle driving-to an existing algorithm with the classes of sitting, lying, standing, walking, and transitioning. After combining the training data, we used a random forest learning scheme for model development. We validated the algorithm through a simulated free-living procedure using chest-mounted cameras for establishing the ground truth. Furthermore, we adjusted our algorithm and compared the performance with an existing algorithm based on vector thresholds.

RESULTS

We developed an algorithm to classify 11 physical behaviors relevant for rehabilitation. In the simulated free-living validation, the performance of the algorithm decreased to 57% as an average for the 11 classes (F-measure). After merging classes into sedentary behavior, standing, walking, running, and cycling, the result revealed high performance in comparison to both the ground truth and the existing algorithm.

CONCLUSIONS

Using a single thigh-mounted accelerometer, we obtained high classification levels within specific behaviors. The behaviors classified with high levels of performance mostly occur in populations with higher levels of functioning. Further development should aim at describing behaviors within populations with lower levels of functioning.

Assessment of 24-hour physical behaviour in children and adolescents via wearables: a systematic review of free-living validation studies

Objectives

Studies that assess all three dimensions of the integrative 24-hour physical behaviour (PB) construct, namely, intensity, posture/activity type and biological state, are on the rise. However, reviews on validation studies that cover intensity, posture/activity type and biological state assessed via wearables are missing.

Design

Systematic review. The risk of bias was evaluated by using the QUADAS-2 tool with nine signalling questions separated into four domains (ie, patient selection/study design, index measure, criterion measure, flow and time).

Data sources

Peer-reviewed validation studies from electronic databases as well as backward and forward citation searches (1970–July 2021).

Eligibility criteria for selecting studies

Wearable validation studies with children and adolescents (age <18 years). Required indicators: (1) study protocol must include real-life conditions; (2) validated device outcome must belong to one dimension of the 24-hour PB construct; (3) the study protocol must include a criterion measure; (4) study results must be published in peer-reviewed English language journals.

Results

Out of 13 285 unique search results, 76 articles with 51 different wearables were included and reviewed. Most studies (68.4%) validated an intensity measure outcome such as energy expenditure, but only 15.9% of studies validated biological state outcomes, while 15.8% of studies validated posture/activity type outcomes. We identified six wearables that had been used to validate outcomes from two different dimensions and only two wearables (ie, ActiGraph GT1M and ActiGraph GT3X+) that validated outcomes from all three dimensions. The percentage of studies meeting a given quality criterion ranged from 44.7% to 92.1%. Only 18 studies were classified as ‘low risk’ or ‘some concerns’.

Summary

Validation studies on biological state and posture/activity outcomes are rare in children and adolescents. Most studies did not meet published quality principles. Standardised protocols embedded in a validation framework are needed.

PROSPERO registration number

CRD42021230894.

Recent Machine Learning Progress in Lower Limb Running Biomechanics With Wearable Technology: A Systematic Review

With the emergence of wearable technology and machine learning approaches, gait monitoring in real-time is attracting interest from the sports biomechanics community. This study presents a systematic review of machine learning approaches in running biomechanics using wearable sensors. Electronic databases were retrieved in PubMed, Web of Science, SPORTDiscus, Scopus, IEEE Xplore, and ScienceDirect. A total of 4,068 articles were identified via electronic databases. Twenty-four articles that met the eligibility criteria after article screening were included in this systematic review. The range of quality scores of the included studies is from 0.78 to 1.00, with 40% of articles recruiting participant numbers between 20 and 50. The number of inertial measurement unit (IMU) placed on the lower limbs varied from 1 to 5, mainly in the pelvis, thigh, distal tibia, and foot. Deep learning algorithms occupied 57% of total machine learning approaches. Convolutional neural networks (CNN) were the most frequently used deep learning algorithm. However, the validation process for machine learning models was lacking in some studies and should be given more attention in future research. The deep learning model combining multiple CNN and recurrent neural networks (RNN) was observed to extract different running features from the wearable sensors and presents a growing trend in running biomechanics.

Predicting lying, sitting and walking at different intensities using smartphone accelerometers at three different wear locations: hands, pant pockets, backpack

Objective

This study uses machine learning (ML) to develop methods for estimating activity type/intensity using smartphones, to evaluate the accuracy of these models for classifying activity, and to evaluate differences in accuracy between three different wear locations.

Method

Forty-eight participants were recruited to complete a series of activities while carrying Samsung phones in three different locations: backpack, right hand and right pocket. They were asked to sit, lie down, walk and run three Metabolic Equivalent Task (METs), five METs and at seven METs. Raw accelerometer data were collected. We used the R, activity counts package, to calculate activity counts and generated new features based on the raw accelerometer data. We evaluated and compared several ML algorithms; Random Forest (RF), Support Vector Machine, Naïve Bayes, Decision Tree, Linear Discriminant Analysis and k-Nearest Neighbours using the caret package (V.6.0-86). Using the combination of the raw accelerometer data and the computed features leads to high model accuracy.

Results

Using raw accelerometer data, RF models achieved an accuracy of 92.90% for the right pocket location, 89% for the right hand location and 90.8% for the backpack location. Using activity counts, RF models achieved an accuracy of 51.4% for the right pocket location, 48.5% for the right hand location and 52.1% for the backpack location.

Conclusion

Our results suggest that using smartphones to measure physical activity is accurate for estimating activity type/intensity and ML methods, such as RF with feature engineering techniques can accurately classify physical activity intensity levels in laboratory settings.

Enhancing Nutrition Care Through Real-Time, Sensor-Based Capture of Eating Occasions: A Scoping Review

As food intake patterns become less structured, different methods of dietary assessment may be required to capture frequently omitted snacks, smaller meals, and the time of day when they are consumed. Incorporating sensors that passively and objectively detect eating behavior may assist in capturing these eating occasions into dietary assessment methods. The aim of this study was to identify and collate sensor-based technologies that are feasible for dietitians to use to assist with performing dietary assessments in real-world practice settings. A scoping review was conducted using the PRISMA extension for scoping reviews (PRISMA-ScR) framework. Studies were included if they were published between January 2016 and December 2021 and evaluated the performance of sensor-based devices for identifying and recording the time of food intake. Devices from included studies were further evaluated against a set of feasibility criteria to determine whether they could potentially be used to assist dietitians in conducting dietary assessments. The feasibility criteria were, in brief, consisting of an accuracy ≥80%; tested in settings where subjects were free to choose their own foods and activities; social acceptability and comfort; a long battery life; and a relatively rapid detection of an eating episode. Fifty-four studies describing 53 unique devices and 4 device combinations worn on the wrist (n = 18), head (n = 16), neck (n = 9), and other locations (n = 14) were included. Whilst none of the devices strictly met all feasibility criteria currently, continuous refinement and testing of device software and hardware are likely given the rapidly changing nature of this emerging field. The main reasons devices failed to meet the feasibility criteria were: an insufficient or lack of reporting on battery life (91%), the use of a limited number of foods and behaviors to evaluate device performance (63%), and the device being socially unacceptable or uncomfortable to wear for long durations (46%). Until sensor-based dietary assessment tools have been designed into more inconspicuous prototypes and are able to detect most food and beverage consumption throughout the day, their use will not be feasible for dietitians in practice settings.

The Validity of the Energy Expenditure Criteria Based on Open Source Code through two Inertial Sensors

Through this study, we developed and validated a system for energy expenditure calculation, which only requires low-cost inertial sensors and open source R software. Five healthy subjects ran at ten different speeds while their kinematic variables were recorded on the thigh and wrist. Two ActiGraph wireless inertial sensors and a low-cost Bluetooth-based inertial sensor (Lis2DH12), assembled by SensorID, were used. Ten energy expenditure equations were automatically calculated in a developed open source R software (our own creation). A correlation analysis was used to compare the results of the energy expenditure equations. A high interclass correlation coefficient of estimated energy expenditure on the thigh and wrist was observed with an Actigraph and Sensor ID accelerometer; the corrected Freedson equation showed the highest values, and the Santos-Lozano vector magnitude equation and Sasaki equation demonstrated the lowest one. Energy expenditure was compared between the wrist and thigh and showed low correlation values. Despite the positive results obtained, it was necessary to design specific equations for the estimation of energy expenditure measured with inertial sensors on the thigh. The use of the same formula equation in two different placements did not report a positive interclass correlation coefficient.

The Best Choice of Oxygen Cost Prediction Equation for Computing Post-Stroke Walking Energy Expenditure Using an Accelerometer

BACKGROUND

The integration of oxygen cost into the accelerometer's algorithms improves accuracy of total energy expenditure (TEE) values as post-stroke individuals walk. Recent work has shown that oxygen cost can be estimated from specific prediction equations for stroke patients.

OBJECTIVE

The objective is to the validity of the different oxygen cost estimation equations available in the literature for calculating TEE using ActigraphGT3x as individuals with stroke sequelae walk.

METHOD

Individuals with stroke sequelae who were able to walk without human assistance were included. The TEE was calculated by multiplying the walking distance provided by an ActigraphGT3x worn on the healthy ankle and the patient's oxygen cost estimated from the selected prediction equations. The TEE values from each equation were compared to the TEE values measured by indirect calorimetry. The validity of the prediction methods was evaluated by Bland-Altman analysis (mean bias (MB) and limits of agreement (LoA) values).

RESULTS

We included 26 stroke patients (63.5 years). Among the selected equations, those of Compagnat and Polese obtained the best validity parameters for the ActigraphGT3x: MB_Compagnat = 1.2 kcal, 95% LoA_Compagnat = [-12.0; 14.3] kcal and MB_Polese = 3.5 kcal, 95% LoA_Polese = [-9.2; 16.1] kcal. For comparison, the estimated TEE value according to the manufacturer's algorithm reported MB_Manufacturer = -15 kcal, 95% LoA_Manufacturer = [-52.9; 22.8] kcal.

CONCLUSION

The Polese and Compagnat equations offer the best validity parameters in comparison with the criterion method. Using oxygen cost prediction equations is a promising approach to improving assessment of TEE by accelerometers in post-stroke individuals.

Perspective: Opportunities and Challenges of Technology Tools in Dietary and Activity Assessment: Bridging Stakeholder Viewpoints

The science and tools of measuring energy intake and output in humans have rapidly advanced in the last decade. Engineered devices such as wearables and sensors, software applications, and Web-based tools are now ubiquitous in both research and consumer environments. The assessment of energy expenditure in particular has progressed from reliance on self-report instruments to advanced technologies requiring collaboration across multiple disciplines, from optics to accelerometry. In contrast, assessing energy intake still heavily relies on self-report mechanisms. Although these tools have improved, moving from paper-based to online reporting, considerable room for refinement remains in existing tools, and great opportunities exist for novel, transformational tools, including those using spectroscopy and chemo-sensing. This report reviews the state of the science, and the opportunities and challenges in existing and emerging technologies, from the perspectives of 3 key stakeholders: researchers, users, and developers. Each stakeholder approaches these tools with unique requirements: researchers are concerned with validity, accuracy, data detail and abundance, and ethical use; users with ease of use and privacy; and developers with high adherence and utilization, intellectual property, licensing rights, and monetization. Cross-cutting concerns include frequent updating and integration of the food and nutrient databases on which assessments rely, improving accessibility and reducing disparities in use, and maintaining reliable technical assistance. These contextual challenges are discussed in terms of opportunities and further steps in the direction of personalized health.

Population and Age-Based Cardiorespiratory Fitness Level Investigation and Automatic Prediction

Maximal oxygen consumption (VO₂max) reflects aerobic capacity and is crucial for assessing cardiorespiratory fitness and physical activity level. The purpose of this study was to classify and predict the population-based cardiorespiratory fitness based on anthropometric parameters, workload, and steady-state heart rate (HR) of the submaximal exercise test. Five hundred and seventeen participants were recruited into this study. This study initially classified aerobic capacity followed by VO₂max predicted using an ordinary least squares regression model with measured VO₂max from a submaximal cycle test as ground truth. Furthermore, we predicted VO₂max in the age ranges 21–40 and above 40. For the support vector classification model, the test accuracy was 75%. The ordinary least squares regression model showed the coefficient of determination (R²) between measured and predicted VO₂max was 0.83, mean absolute error (MAE) and root mean square error (RMSE) were 3.12 and 4.24 ml/kg/min, respectively. R² in the age 21–40 and above 40 groups were 0.85 and 0.75, respectively. In conclusion, this study provides a practical protocol for estimating cardiorespiratory fitness of an individual in large populations. An applicable submaximal test for population-based cohorts could evaluate physical activity levels and provide exercise recommendations.

A comparison of accelerometer cut-points for measuring physical activity and sedentary time in adolescents with Down syndrome

BACKGROUND

No cut-points have been developed for youth with Down syndrome; there is concern that altered gait patterns, decreased energy expenditure and exercise capacity of individuals with Down syndrome may produce inaccurate physical activity data if accelerometer data are analyzed using cut-points from populations with typical development and other IDD diagnoses.

AIM

To compare physical activity and sedentary time across existing accelerometer cut-point methods in adolescents with Down syndrome.

METHODS

In this cross-sectional analysis, participants diagnosed with Down syndrome (n = 37; 15.5 ± 1.9 years; 57 % female) wore an accelerometer on their non-dominant hip for seven-days. Data were analyzed and compared across four physical activity intensity cut-points: Evenson, Freedson 4-MET, McGarty, and Romanizi.

OUTCOMES & RESULTS

Differences in time spent in each intensity across cut-point methods were evident for sedentary (448-615 min/day), light (72-303 min/day) and moderate-to-vigorous (12-77 min/day) activities. Between 0.0-67.6 % of the sample met the physical activity guidelines, depending on the cut-point method selected.

CONCLUSIONS & IMPLICATIONS

This study presents the wide variation of accumulated physical activity minutes when different cut-points are applied to individuals with Down syndrome. There is a critical need to establish Down syndrome-specific measures of physical activity assessment rather than applying methods developed for their peers with typical development.

WHAT THIS PAPER ADDS

This paper highlights concerns over the application of objective measurements of physical activity in youth with Down syndrome from measurement methods derived from populations with typical development. This is the first manuscript to examine this issue in a sample comprised solely of youth with Down syndrome. Results demonstrate the large variation in time spent in each activity intensity that arise due to the application of different cut-point methods.

The CNN Hip Accelerometer Posture (CHAP) Method for Classifying Sitting Patterns from Hip Accelerometers: A Validation Study

INTRODUCTION

Sitting patterns predict several healthy aging outcomes. These patterns can potentially be measured using hip-worn accelerometers, but current methods are limited by an inability to detect postural transitions. To overcome these limitations, we developed the Convolutional Neural Network Hip Accelerometer Posture (CHAP) classification method.

METHODS

CHAP was developed on 709 older adults who wore an ActiGraph GT3X+ accelerometer on the hip, with ground-truth sit/stand labels derived from concurrently worn thigh-worn activPAL inclinometers for up to 7 d. The CHAP method was compared with traditional cut-point methods of sitting pattern classification as well as a previous machine-learned algorithm (two-level behavior classification).

RESULTS

For minute-level sitting versus nonsitting classification, CHAP performed better (93% agreement with activPAL) than did other methods (74%-83% agreement). CHAP also outperformed other methods in its sensitivity to detecting sit-to-stand transitions: cut-point (73%), TLBC (26%), and CHAP (83%). CHAP's positive predictive value of capturing sit-to-stand transitions was also superior to other methods: cut-point (30%), TLBC (71%), and CHAP (83%). Day-level sitting pattern metrics, such as mean sitting bout duration, derived from CHAP did not differ significantly from activPAL, whereas other methods did: activPAL (15.4 min of mean sitting bout duration), CHAP (15.7 min), cut-point (9.4 min), and TLBC (49.4 min).

CONCLUSION

CHAP was the most accurate method for classifying sit-to-stand transitions and sitting patterns from free-living hip-worn accelerometer data in older adults. This promotes enhanced analysis of older adult movement data, resulting in more accurate measures of sitting patterns and opening the door for large-scale cohort studies into the effects of sitting patterns on healthy aging outcomes.

Open-source Longitudinal Sleep Analysis From Accelerometer Data (DPSleep): Algorithm Development and Validation

BACKGROUND

Wearable devices are now widely available to collect continuous objective behavioral data from individuals and to measure sleep.

OBJECTIVE

This study aims to introduce a pipeline to infer sleep onset, duration, and quality from raw accelerometer data and then quantify the relationships between derived sleep metrics and other variables of interest.

METHODS

The pipeline released here for the deep phenotyping of sleep, as the DPSleep software package, uses a stepwise algorithm to detect missing data; within-individual, minute-based, spectral power percentiles of activity; and iterative, forward-and-backward-sliding windows to estimate the major Sleep Episode onset and offset. Software modules allow for manual quality control adjustment of the derived sleep features and correction for time zone changes. In this paper, we have illustrated the pipeline with data from participants studied for more than 200 days each.

RESULTS

Actigraphy-based measures of sleep duration were associated with self-reported sleep quality ratings. Simultaneous measures of smartphone use and GPS location data support the validity of the sleep timing inferences and reveal how phone measures of sleep timing can differ from actigraphy data.

CONCLUSIONS

We discuss the use of DPSleep in relation to other available sleep estimation approaches and provide example use cases that include multi-dimensional, deep longitudinal phenotyping, extended measurement of dynamics associated with mental illness, and the possibility of combining wearable actigraphy and personal electronic device data (eg, smartphones and tablets) to measure individual differences across a wide range of behavioral variations in health and disease. A new open-source pipeline for deep phenotyping of sleep, DPSleep, analyzes raw accelerometer data from wearable devices and estimates sleep onset and offset while allowing for manual quality control adjustments.

Reallocation of time between device-measured movement behaviours and risk of incident cardiovascular disease

OBJECTIVE

To improve classification of movement behaviours in free-living accelerometer data using machine-learning methods, and to investigate the association between machine-learned movement behaviours and risk of incident cardiovascular disease (CVD) in adults.

METHODS

Using free-living data from 152 participants, we developed a machine-learning model to classify movement behaviours (moderate-to-vigorous physical activity behaviours (MVPA), light physical activity behaviours, sedentary behaviour, sleep) in wrist-worn accelerometer data. Participants in UK Biobank, a prospective cohort, were asked to wear an accelerometer for 7 days, and we applied our machine-learning model to classify their movement behaviours. Using compositional data analysis Cox regression, we investigated how reallocating time between movement behaviours was associated with CVD incidence.

RESULTS

In leave-one-participant-out analysis, our machine-learning method classified free-living movement behaviours with mean accuracy 88% (95% CI 87% to 89%) and Cohen's kappa 0.80 (95% CI 0.79 to 0.82). Among 87 498 UK Biobank participants, there were 4105 incident CVD events. Reallocating time from any behaviour to MVPA, or reallocating time from sedentary behaviour to any behaviour, was associated with lower CVD risk. For an average individual, reallocating 20 min/day to MVPA from all other behaviours proportionally was associated with 9% (95% CI 7% to 10%) lower risk, while reallocating 1 hour/day to sedentary behaviour from all other behaviours proportionally was associated with 5% (95% CI 3% to 7%) higher risk.

CONCLUSION

Machine-learning methods classified movement behaviours accurately in free-living accelerometer data. Reallocating time from other behaviours to MVPA, and from sedentary behaviour to other behaviours, was associated with lower risk of incident CVD, and should be promoted by interventions and guidelines.

Machine-learning models for activity class prediction: A comparative study of feature selection and classification algorithms

PURPOSE

Machine-learning (ML) approaches have been repeatedly coupled with raw accelerometry to classify physical activity classes, but the features required to optimize their predictive performance are still unknown. Our aim was to identify appropriate combination of feature subsets and prediction algorithms for activity class prediction from hip-based raw acceleration data.

METHODS

The hip-based raw acceleration data collected from 27 participants was split into training (70 %) and validation (30 %) subsets. A total of 206 time- (TD) and frequencydomain (FD) features were extracted from 6-second non-overlapping windows of the signal. Feature selection was done using seven filter-based, two wrapper-based, and one embedded algorithm, and classification was performed with artificial neural network (ANN), support vector machine (SVM), and random forest (RF). For every combination between the feature selection method and the classifiers, the most appropriate feature subsets were found and used for model training within the training set. These models were then validated with the left-out validation set.

RESULTS

The appropriate number of features for the ANN, SVM, and RF ranged from 20 to 45. Overall, the accuracy of all the three classifiers was higher when trained with feature subsets generated using filter-based methods compared with when they were trained with wrapper-based methods (range: 78.1 %-88 % vs. 66 %-83.5 %). TD features that reflect how signals vary around the mean, how they differ with one another, and how much and how often they change were more frequently selected via the feature selection methods.

CONCLUSIONS

A subset of TD features from raw accelerometry could be sufficient for ML-based activity classification if properly selected from different axes.

Application of Convolutional Neural Network Algorithms for Advancing Sedentary and Activity Bout Classification

Background

Machine learning has been used for classification of physical behavior bouts from hip-worn accelerometers; however, this research has been limited due to the challenges of directly observing and coding human behavior "in the wild." Deep learning algorithms, such as convolutional neural networks (CNNs), may offer better representation of data than other machine learning algorithms without the need for engineered features and may be better suited to dealing with free-living data. The purpose of this study was to develop a modeling pipeline for evaluation of a CNN model on a free-living data set and compare CNN inputs and results with the commonly used machine learning random forest and logistic regression algorithms.

Method

Twenty-eight free-living women wore an ActiGraph GT3X+accelerometer on their right hip for 7 days. A concurrently worn thigh-mounted activPAL device captured ground truth activity labels. The authors evaluated logistic regression, random forest, and CNN models for classifying sitting, standing, and stepping bouts. The authors also assessed the benefit of performing feature engineering for this task.

Results

The CNN classifier performed best (average balanced accuracy for bout classification of sitting, standing, and stepping was 84%) compared with the other methods (56% for logistic regression and 76% for random forest), even without performing any feature engineering.

Conclusion

Using the recent advancements in deep neural networks, the authors showed that a CNN model can outperform other methods even without feature engineering. This has important implications for both the model's ability to deal with the complexity of free-living data and its potential transferability to new populations.

Individual versus Group Calibration of Machine Learning Models for Physical Activity Assessment Using Body-Worn Accelerometers

Modeling approaches for translating accelerometer data into physical activity metrics are often developed using a group calibration approach. However, it is unknown if models developed for specific individuals will improve measurement accuracy.

PURPOSE

We sought to determine if individually calibrated machine learning models yielded higher accuracy than a group calibration approach for physical activity intensity assessment.

METHODS

Participants (n = 48) wore accelerometers on the right hip and non-dominant wrist while performing activities of daily living in a semi-structured laboratory and/or free-living setting. Criterion measures of activity intensity (sedentary, light, moderate, vigorous) were determined using direct observation. Data were reintegrated into 30-second epochs, and eight random forest models were created to determine physical activity intensity by using all possible conditions of training data (individual vs. group), protocol (laboratory vs. free-living), and placement (hip vs. wrist). A 2x2x2 repeated-measures analysis of variance was used to compare epoch-level accuracy statistics (% accuracy, kappa [k]) of the models when used to determine activity intensity in an independent sample of free-living participants.

RESULTS

Main effects were significant for the type of training data (group: accuracy = 80%, k = 0.59; individual: accuracy = 74% [p = 0.02], k = 0.50 [p = 0.01]) and protocol (free-living: accuracy = 81%, k = 0.63; laboratory: accuracy = 74% [p = 0.04], k = 0.47 [p < 0.01]). Main effects were not significant for placement (hip: accuracy = 79%, k = 0.58; wrist: accuracy = 75% [p = 0.18]; k = 0.52 [p = 0.18]). Point estimates for mean absolute error were generally lowest for the group training, free-living protocol, and hip placement.

CONCLUSION

Contrary to expectations, individually calibrated machine learning models yielded poorer accuracy than a traditional group approach. Additionally, models should be developed in free-living settings when possible to optimize predictive accuracy.

How Accurate and Precise Can We Measure the Posture and the Energy Expenditure Component of Sedentary Behaviour with One Sensor?

Sedentary behaviour is an emergent public health topic, but there is still no method to simultaneously measure both components of sedentary behaviour-posture and energy expenditure-with one sensor. This study investigated the accuracy and precision of measuring sedentary time when combining the proprietary processing of a posture sensor (activPAL) with a new energy expenditure algorithm and the proprietary processing of a movement sensor (ActiGraph) with a published posture algorithm. One hundred office workers wore both sensors for an average of 7 days. The activPAL algorithm development used 38 and the subsequent independent method comparison 62 participants. The single sensor sedentary estimates were compared with Bland-Atman statistics to the Posture and Physical Activity Index, a combined measurement with both sensors. All single-sensor methods overestimated sedentary time. However, adding the algorithms reduced the overestimation from 129 to 21 (activPAL) and from 84 to 7 min a day (ActiGraph), with far narrower 95% limits of agreements. Thus, combining the proprietary data with the algorithms is an easy way to increase the accuracy and precision of the single sensor sedentary estimates and leads to sedentary estimates that are more precise at the individual level than those of the proprietary processing are at the group level.

Measurement of Physical Activity by Shoe-Based Accelerometers-Calibration and Free-Living Validation

There is conflicting evidence regarding the health implications of high occupational physical activity (PA). Shoe-based accelerometers could provide a feasible solution for PA measurement in workplace settings. This study aimed to develop calibration models for estimation of energy expenditure (EE) from shoe-based accelerometers, validate the performance in a workplace setting and compare it to the most commonly used accelerometer positions. Models for EE estimation were calibrated in a laboratory setting for the shoe, hip, thigh and wrist worn accelerometers. These models were validated in a free-living workplace setting. Furthermore, additional models were developed from free-living data. All sensor positions performed well in the laboratory setting. When the calibration models derived from laboratory data were validated in free living, the shoe, hip and thigh sensors displayed higher correlation, but lower agreement, with measured EE compared to the wrist sensor. Using free-living data for calibration improved the agreement of the shoe, hip and thigh sensors. This study suggests that the performance of a shoe-based accelerometer is similar to the most commonly used sensor positions with regard to PA measurement. Furthermore, it highlights limitations in using the relationship between accelerometer output and EE from a laboratory setting to estimate EE in a free-living setting.

Optimization and Validation of a Classification Algorithm for Assessment of Physical Activity in Hospitalized Patients

Low amounts of physical activity (PA) and prolonged periods of sedentary activity are common in hospitalized patients. Objective PA monitoring is needed to prevent the negative effects of inactivity, but a suitable algorithm is lacking. The aim of this study is to optimize and validate a classification algorithm that discriminates between sedentary, standing, and dynamic activities, and records postural transitions in hospitalized patients under free-living conditions. Optimization and validation in comparison to video analysis were performed in orthopedic and acutely hospitalized elderly patients with an accelerometer worn on the upper leg. Data segmentation window size (WS), amount of PA threshold (PA Th) and sensor orientation threshold (SO Th) were optimized in 25 patients, validation was performed in another 25. Sensitivity, specificity, accuracy, and (absolute) percentage error were used to assess the algorithm’s performance. Optimization resulted in the best performance with parameter settings: WS 4 s, PA Th 4.3 counts per second, SO Th 0.8 g. Validation showed that all activities were classified within acceptable limits (>80% sensitivity, specificity and accuracy, ±10% error), except for the classification of standing activity. As patients need to increase their PA and interrupt sedentary behavior, the algorithm is suitable for classifying PA in hospitalized patients.

Correlation of quality of life with instrumented analysis of a total knee arthroplasty series at the long-term follow-up

PURPOSE

The relationship between instrumented knee measurements and patient-reported outcome measures is a newer field that continues to evolve. The aim of this study was to evaluate long-term quality of life (QoL) post-total knee arthroplasty (TKA) surgery correlating validated self-reported questionnaires, clinical examination and instrumented analysis, using baropodometry and accelerometry.

METHODS

Thirty-six patients who underwent primary unilateral TKA between 1999 and 2006 were evaluated at 11.3 ± 2.3 years following surgery. Clinical examination included range of motion (ROM) and instrumented knee laxity measurements with the Rolimeter device. The visual analogue scale (VAS) for pain was also recorded. The utilised subjective outcome scores were the Knee Injury and Osteoarthritis Outcome Score (KOOS) and the short form of World Health Organisation Quality of Life (WHOQOL-BREF). Instrumented analysis was performed with baropodometry and accelerometry. QoL was assessed correlating clinical, subjective and instrumented results. Univariate analysis included the Spearman's Rho correlation coefficient and Mann-Whitney tests.

RESULTS

At the long-term follow-up all patients had relatively high quality of life measurements, as well as functional scores, except for the Sport/Rec dimension of the KOOS score. Only cadence (p = 0.008) and velocity (p = 0.026) affected the WHOQOL psychology domain no matter the age, follow-up and gender of the patients. The domain was unaffected by VAS and Rolimeter measurements. WHOQOL Social domain was unaffected by all instrumentation measurements except for stance phase (p = 0.025), VAS (p = 0.005) and ROM (p = 0.028). KOOS physical domain was not affected by any parameter. KOOS pain was reversely affected by VAS (p = 0.004), KOOS symptom by ROM (p = 0.000 and median maximum pressure (p = 0.033).

CONCLUSION

Quality of life for the TKA patient can be correlated and assessed reliably with instrumented analysis using pedobarography and accelerometry, at the long-term follow-up.

LEVEL OF EVIDENCE

III.

Calibrating Wrist-Worn Accelerometers for Physical Activity Assessment in Preschoolers: Machine Learning Approaches

BACKGROUND

Physical activity (PA) level is associated with multiple health benefits during early childhood. However, inconsistency in the methods for quantification of PA levels among preschoolers remains a problem.

OBJECTIVE

This study aimed to develop PA intensity cut points for wrist-worn accelerometers by using machine learning (ML) approaches to assess PA in preschoolers.

METHODS

Wrist- and hip-derived acceleration data were collected simultaneously from 34 preschoolers on 3 consecutive preschool days. Two supervised ML models, receiver operating characteristic curve (ROC) and ordinal logistic regression (OLR), and one unsupervised ML model, k-means cluster analysis, were applied to establish wrist-worn accelerometer vector magnitude (VM) cut points to classify accelerometer counts into sedentary behavior, light PA (LPA), moderate PA (MPA), and vigorous PA (VPA). Physical activity intensity levels identified by hip-worn accelerometer VM cut points were used as reference to train the supervised ML models. Vector magnitude counts were classified by intensity based on three newly established wrist methods and the hip reference to examine classification accuracy. Daily estimates of PA were compared to the hip-reference criterion.

RESULTS

In total, 3600 epochs with matched hip- and wrist-worn accelerometer VM counts were analyzed. All ML approaches performed differently on developing PA intensity cut points for wrist-worn accelerometers. Among the three ML models, k-means cluster analysis derived the following cut points: ≤2556 counts per minute (cpm) for sedentary behavior, 2557-7064 cpm for LPA, 7065-14532 cpm for MPA, and ≥14533 cpm for VPA; in addition, k-means cluster analysis had the highest classification accuracy, with more than 70% of the total epochs being classified into the correct PA categories, as examined by the hip reference. Additionally, k-means cut points exhibited the most accurate estimates on sedentary behavior, LPA, and VPA as the hip reference. None of the three wrist methods were able to accurately assess MPA.

CONCLUSIONS

This study demonstrates the potential of ML approaches in establishing cut points for wrist-worn accelerometers to assess PA in preschoolers. However, the findings from this study warrant additional validation studies.