Leave-one-subject-out Research Articles

Generative Adversarial Networks-Based Data Augmentation for Brain-Computer Interface.

The performance of a classifier in a brain-computer interface (BCI) system is highly dependent on the quality and quantity of training data. Typically, the training data are collected in a laboratory where the users perform tasks in a controlled environment. However, users' attention may be diverted in real-life BCI applications and this may decrease the performance of the classifier. To improve the robustness of the classifier, additional data can be acquired in such conditions, but it is not practical to record electroencephalogram (EEG) data over several long calibration sessions. A potentially time- and cost-efficient solution is artificial data generation. Hence, in this study, we proposed a framework based on the deep convolutional generative adversarial networks (DCGANs) for generating artificial EEG to augment the training set in order to improve the performance of a BCI classifier. To make a comparative investigation, we designed a motor task experiment with diverted and focused attention conditions. We used an end-to-end deep convolutional neural network for classification between movement intention and rest using the data from 14 subjects. The results from the leave-one subject-out (LOO) classification yielded baseline accuracies of 73.04% for diverted attention and 80.09% for focused attention without data augmentation. Using the proposed DCGANs-based framework for augmentation, the results yielded a significant improvement of 7.32% for diverted attention ( ) and 5.45% for focused attention ( ). In addition, we implemented the method on the data set IVa from BCI competition III to distinguish different motor imagery tasks. The proposed method increased the accuracy by 3.57% ( ). This study shows that using GANs for EEG augmentation can significantly improve BCI performance, especially in real-life applications, whereby users' attention may be diverted.

Combining deep learning and model-based method using Bayesian Inference for walking speed estimation

In this paper, deep learning and model-based method were combined using Bayesian Inference to realize high accuracy and good generalization capability stride-by-stride walking speed estimation by low cost inertial measurement unit (IMU) sensor. Long Short-Term Memory (LSTM) network was applied to train the prediction model because of its ability to consider the temporal correlation of multi-dimensional kinematic parameters during one stride. To improve the performance with unseen subjects, a model-based method was introduced for its relatively good generalization capability. Fusion strategy based on Bayesian Inference was applied to take advantage of the two methods which considered the estimation derived from different methods as abstract sensors. Six healthy young adults performed treadmill walking with shank-mounted IMU and the range of the walking speed was 2.5 km/h to 5 km/h at an increment of 0.5 km/h. Leave-one-subject-out (LOSO) cross-validation was performed to analyze the generalization capability. For deep learning method, the root mean square error (RMSE) of the model trained by all available subjects was 0.026 m/s and the RMSE of LOSO cross-validation was 0.066 m/s which indicated a low generalization capability. After the fusion strategy was applied, RMSE of the model trained by all available subjects was 0.023 m/s which was slightly improved, while the RMSE of LOSO cross-validation was reduced to 0.036 m/s which indicated that accuracy and the generalization capability was greatly improved. In addition, this accurate estimation can be easily realized online which is essential for locomotion interactive systems (e.g. self-paced treadmill).

An Attention Based CNN-LSTM Approach for Sleep-Wake Detection With Heterogeneous Sensors

In this article, we propose an attention based convolutional neural network long short-term memory (CNN-LSTM) approach for sleep-wake detection with heterogeneous sensor data, i.e., acceleration and heart rate variability (HRV). Since the three-dimensional acceleration data was sampled with a high frequency, we firstly design a CNN-LSTM structure to effectively learn latent features from the acceleration. Meanwhile, considering the unique format of the HRV data, some effective features are extracted based on domain knowledge. Next, we design a unified architecture to efficiently merge the features learned by CNN-LSTM approach from the acceleration and the extracted features from the HRV, which enables us to make full use of all the available information from these two heterogeneous sources. Taking into consideration that these two heterogeneous sources may have distinct contributions for the sleep and wake states, we propose an attention network to dynamically adjust the importance of features from the two sources. Real-world experiments have been conducted to verify the effectiveness of the proposed approach for sleep-wake detection. The results demonstrate that the proposed method outperforms all existing approaches for sleep-wake classification. In the evaluation of leave-one-subject-out (LOSO) cross-validation which is more challenging and practical, the proposed method achieves remarkable improvements ranging from 5% to 46% over the benchmark approaches.

Improving physical activity recognition using a new deep learning architecture and post-processing techniques

This paper proposes a Human Activity Recognition system composed of three modules. The first one segments the acceleration signals into overlapped windows and extracts information from each window in the frequency domain. The second module detects the performed activity at each window using a deep learning structure based on Convolutional Neural Networks (CNNs). The first part of this structure has several layers associated to each sensor independently and the second part combines the outputs from all sensors in order to classify the physical activity. The third module integrates the window-level decision in longer periods of time, obtaining a significant performance improvement (from 89.83% to 96.62%). These are the best classification results on the PAMAP2 dataset with a Leave-One-Subject-Out (LOSO) evaluation.

Deep Learning Approaches for Detecting Freezing of Gait in Parkinson's Disease Patients through On-Body Acceleration Sensors.

Freezing of gait (FOG) is one of the most incapacitating motor symptoms in Parkinson’s disease (PD). The occurrence of FOG reduces the patients’ quality of live and leads to falls. FOG assessment has usually been made through questionnaires, however, this method can be subjective and could not provide an accurate representation of the severity of this symptom. The use of sensor-based systems can provide accurate and objective information to track the symptoms’ evolution to optimize PD management and treatments. Several authors have proposed specific methods based on wearables and the analysis of inertial signals to detect FOG in laboratory conditions, however, its performance is usually lower when being used at patients’ homes. This study presents a new approach based on a recurrent neural network (RNN) and a single waist-worn triaxial accelerometer to enhance the FOG detection performance to be used in real home-environments. Also, several machine and deep learning approaches for FOG detection are evaluated using a leave-one-subject-out (LOSO) cross-validation. Results show that modeling spectral information of adjacent windows through an RNN can bring a significant improvement in the performance of FOG detection without increasing the length of the analysis window (required to using it as a cue-system).

Motion-to-BMI: Using Motion Sensors to Predict the Body Mass Index of Smartphone Users.

Obesity has become a widespread health problem worldwide. The body mass index (BMI) is a simple and reliable index based on weight and height that is commonly used to identify and classify adults as underweight, normal, overweight (pre-obesity), or obese. In this paper, we propose a hybrid deep neural network for predicting the BMI of smartphone users, based only on the characteristics of body movement captured by the smartphone’s built-in motion sensors without any other sensitive data. The proposed deep learning model consists of four major modules: a transformation module for data preprocessing, a convolution module for extracting spatial features, a long short-term memory (LSTM) module for exploring temporal dependency, and a fully connected module for regression. We define motion entropy (MEn), which is a measure of the regularity and complexity of the motion sensor, and propose a novel MEn-based filtering strategy to select parts of sensor data that met certain thresholds for training the model. We evaluate this model using two public datasets in comparison with baseline conventional feature-based methods using leave-one-subject-out (LOSO) cross-validation. Experimental results show that the proposed model with the MEn-based filtering strategy outperforms the baseline approaches significantly. The results also show that jogging may be a more suitable activity of daily living (ADL) for BMI prediction than walking and walking upstairs. We believe that the conclusions of this study will help to develop a long-term remote health monitoring system.

Exploring Deep Physiological Models for Nociceptive Pain Recognition

Standard feature engineering involves manually designing measurable descriptors based on some expert knowledge in the domain of application, followed by the selection of the best performing set of designed features for the subsequent optimisation of an inference model. Several studies have shown that this whole manual process can be efficiently replaced by deep learning approaches which are characterised by the integration of feature engineering, feature selection and inference model optimisation into a single learning process. In the following work, deep learning architectures are designed for the assessment of measurable physiological channels in order to perform an accurate classification of different levels of artificially induced nociceptive pain. In contrast to previous works, which rely on carefully designed sets of hand-crafted features, the current work aims at building competitive pain intensity inference models through autonomous feature learning, based on deep neural networks. The assessment of the designed deep learning architectures is based on the BioVid Heat Pain Database (Part A) and experimental validation demonstrates that the proposed uni-modal architecture for the electrodermal activity (EDA) and the deep fusion approaches significantly outperform previous methods reported in the literature, with respective average performances of and for the binary classification experiment consisting of the discrimination between the baseline and the pain tolerance level ( vs. ) in a Leave-One-Subject-Out (LOSO) cross-validation evaluation setting. Moreover, the experimental results clearly show the relevance of the proposed approaches, which also offer more flexibility in the case of transfer learning due to the modular nature of deep neural networks.

A Comparison of SVM and CNN-LSTM Based Approach for Detecting Smoke Inhalations from Respiratory signal.

Wearable sensors have successfully been used in recent studies to monitor cigarette smoking events and analyze people's smoking behavior. Respiratory inductive plethysmography (RIP) has been employed to track breathing and to identify characteristic breathing pattern specific to smoking. Pattern recognition algorithms such as Support Vector Machine (SVM), Hidden Markov Model, Decision tree, or ensemble approaches have been used to identify smoke inhalations. However, no deep learning approaches, which have been proved effective to many time series datasets, have ever been tested yet. Hence, a Convolutional Neural Network (CNN) and Long Term Short Memory (LSTM) based approach is presented in this paper to detect smoke inhalations in the breathing signal. To illustrate the effectiveness of this deep learning approach, a traditional machine learning (SVM) based approach was used for comparison. On the validation dataset of 120 smoking sessions performed in a laboratory setting by 30 moderate-to-heavy smokers, the CNN-LSTM approach achieved an F1-score of 72% in leave-one-subject-out (LOSO) cross-validation method whereas the classical SVM approach scored 63%. These results suggest that deep learning-based approaches might provide a better analytical method for detection of smoke inhalations than more conventional machine learning approaches.

Cigarette Smoking Detection with An Inertial Sensor and A Smart Lighter.

In recent years, a number of wearable approaches have been introduced for objective monitoring of cigarette smoking based on monitoring of hand gestures, breathing or cigarette lighting events. However, non-reactive, objective and accurate measurement of everyday cigarette consumption in the wild remains a challenge. This study utilizes a wearable sensor system (Personal Automatic Cigarette Tracker 2.0, PACT2.0) and proposes a method that integrates information from an instrumented lighter and a 6-axis Inertial Measurement Unit (IMU) on the wrist for accurate detection of smoking events. The PACT2.0 was utilized in a study of 35 moderate to heavy smokers in both controlled (1.5–2 h) and unconstrained free-living conditions (~24 h). The collected dataset contained approximately 871 h of IMU data, 463 lighting events, and 443 cigarettes. The proposed method identified smoking events from the cigarette lighter data and estimated puff counts by detecting hand-to-mouth gestures (HMG) in the IMU data by a Support Vector Machine (SVM) classifier. The leave-one-subject-out (LOSO) cross-validation on the data from the controlled portion of the study achieved high accuracy and F1-score of smoking event detection and estimation of puff counts (97%/98% and 93%/86%, respectively). The results of validation in free-living demonstrate 84.9% agreement with self-reported cigarettes. These results suggest that an IMU and instrumented lighter may potentially be used in studies of smoking behavior under natural conditions.

Increasing Robustness in the Detection of Freezing of Gait in Parkinson’s Disease

This paper focuses on detecting freezing of gait in Parkinson’s patients using body-worn accelerometers. In this study, we analyzed the robustness of four feature sets, two of which are new features adapted from speech processing: mel frequency cepstral coefficients and quality assessment metrics. For classification based on these features, we compared random forest, multilayer perceptron, hidden Markov models, and deep neural networks. These algorithms were evaluated using a leave-one-subject-out (LOSO) cross validation to match the situation where a system is being constructed for patients for whom there is no training data. This evaluation was performed using the Daphnet dataset, which includes recordings from ten patients using three accelerometers situated on the ankle, knee, and lower back. We obtained a reduction from 17.3% to 12.5% of the equal error rate compared to the previous best results using this dataset and LOSO testing. For high levels of sensitivity (such as 0.95), the specificity increased from 0.63 to 0.75. The biggest improvement across all of the feature sets and algorithms tested in this study was obtained by integrating information from longer periods of time in a deep neural network with convolutional layers.

Automated Detection of Parkinson's Disease Based on Multiple Types of Sustained Phonations Using Linear Discriminant Analysis and Genetically Optimized Neural Network.

Objective: Parkinson’s disease (PD) is a serious neurodegenerative disorder. It is reported that most of PD patients have voice impairments. But these voice impairments are not perceptible to common listeners. Therefore, different machine learning methods have been developed for automated PD detection. However, these methods either lack generalization and clinically significant classification performance or face the problem of subject overlap. Methods: To overcome the problems discussed above, we attempt to develop a hybrid intelligent system that can automatically perform acoustic analysis of voice signals in order to detect PD. The proposed intelligent system uses linear discriminant analysis (LDA) for dimensionality reduction and genetic algorithm (GA) for hyperparameters optimization of neural network (NN) which is used as a predictive model. Moreover, to avoid subject overlap, we use leave one subject out (LOSO) validation. Results: The proposed method namely LDA-NN-GA is evaluated in numerical experiments on multiple types of sustained phonations data in terms of accuracy, sensitivity, specificity, and Matthew correlation coefficient. It achieves classification accuracy of 95% on training database and 100% on testing database using all the extracted features. However, as the dataset is imbalanced in terms of gender, thus, to obtain unbiased results, we eliminated the gender dependent features and obtained accuracy of 80% for training database and 82.14% for testing database, which seems to be more unbiased results. Conclusion: Compared with the previous machine learning methods, the proposed LDA-NN-GA method shows better performance and lower complexity. Clinical Impact: The experimental results suggest that the proposed automated diagnostic system has the potential to classify PD patients from healthy subjects. Additionally, in future the proposed method can also be exploited for prodromal and differential diagnosis, which are considered challenging tasks.

Classifying Parkinson's Disease Based on Acoustic Measures Using Artificial Neural Networks.

In recent years, neural networks have become very popular in all kinds of prediction problems. In this paper, multiple feed-forward artificial neural networks (ANNs) with various configurations are used in the prediction of Parkinson’s disease (PD) of tested individuals, based on extracted features from 26 different voice samples per individual. Results are validated via the leave-one-subject-out (LOSO) scheme. Few feature selection procedures based on Pearson’s correlation coefficient, Kendall’s correlation coefficient, principal component analysis, and self-organizing maps, have been used for boosting the performance of algorithms and for data reduction. The best test accuracy result has been achieved with Kendall’s correlation coefficient-based feature selection, and the most relevant voice samples are recognized. Multiple ANNs have proven to be the best classification technique for diagnosis of PD without usage of the feature selection procedure (on raw data). Finally, a neural network is fine-tuned, and a test accuracy of 86.47% was achieved.

A Novel Gesture Recognition System for Intelligent Interaction with a Nursing-Care Assistant Robot

The expansion of nursing-care assistant robots in smart infrastructure has provided more applications for homecare services, which has raised new demands for smart and natural interaction between humans and robots. This article proposed an innovative hand motion trajectory (HMT) gesture recognition system based on background velocity features. Here, a new wearable wrist-worn camera prototype for gesture’s video collection was designed, and a new method for the segmentation of continuous gestures was shown. Meanwhile, a nursing-care assistant robot prototype was designed for assisting the elderly, which is capable of carrying the elderly with omnidirectional motion and grabbing the specified object at home. In order to evaluate the performance of the gesture recognition system, 10 special gestures were defined as the move commands for interaction with the robot, and 1000 HMT gesture samples were obtained from five subjects for leave-one-subject-out (LOSO) cross-validation classification with an average recognition accuracy of up to 97.34%. Moreover, the performance and practicability of the proposed system were further demonstrated by controlling the omnidirectional movement of the nursing-care assistant robot using the predefined gesture commands.

Wheeze type classification using non-dyadic wavelet transform based optimal energy ratio technique

Background and objectiveWheezes in pulmonary sounds are anomalies which are often associated with obstructive type of lung diseases. The previous works on wheeze-type classification focused mainly on using fixed time-frequency/scale resolution based on Fourier and wavelet transforms. The main contribution of the proposed method, in which the time-scale resolution can be tuned according to the signal of interest, is to discriminate monophonic and polyphonic wheezes with higher accuracy than previously suggested time and time-frequency/scale based methods. MethodsAn optimal Rational Dilation Wavelet Transform (RADWT) based peak energy ratio (PER) parameter selection method is proposed to discriminate wheeze types. Previously suggested Quartile Frequency Ratios, Mean Crossing Irregularity, Multiple Signal Classification, Mel-frequency Cepstrum and Dyadic Discrete Wavelet Transform approaches are also applied and the superiority of the proposed method is demonstrated in leave-one-out (LOO) and leave-one-subject-out (LOSO) cross validation schemes with support vector machine (SVM), k nearest neighbor (k-NN) and extreme learning machine (ELM) classifiers. ResultsThe results show that the proposed RADWT based method outperforms the state-of-the-art time, frequency, time-frequency and time-scale domain approaches for all classifiers in both LOO and LOSO cross validation settings. The highest accuracy values are obtained as 86% and 82.9% in LOO and LOSO respectively when the proposed PER features are fed into SVM. ConclusionsIt is concluded that time and frequency domain characteristics of wheezes are not steady and hence, tunable time-scale representations are more successful in discriminating polyphonic and monophonic wheezes when compared with conventional fixed resolution representations.

A hierarchical multimodal system for motion analysis in patients with epilepsy

During seizures, a myriad of clinical manifestations may occur. The analysis of these signs, known as seizure semiology, gives clues to the underlying cerebral networks involved. When patients with drug-resistant epilepsy are monitored to assess their suitability for epilepsy surgery, semiology is a vital component to the presurgical evaluation. Specific patterns of facial movements, head motions, limb posturing and articulations, and hand and finger automatisms may be useful in distinguishing between mesial temporal lobe epilepsy (MTLE) and extratemporal lobe epilepsy (ETLE). However, this analysis is time-consuming and dependent on clinical experience and training. Given this limitation, an automated analysis of semiological patterns, i.e., detection, quantification, and recognition of body movement patterns, has the potential to help increase the diagnostic precision of localization. While a few single modal quantitative approaches are available to assess seizure semiology, the automated quantification of patients' behavior across multiple modalities has seen limited advances in the literature. This is largely due to multiple complicated variables commonly encountered in the clinical setting, such as analyzing subtle physical movements when the patient is covered or room lighting is inadequate. Semiology encompasses the stepwise/temporal progression of signs that is reflective of the integration of connected neuronal networks. Thus, single signs in isolation are far less informative. Taking this into account, here, we describe a novel modular, hierarchical, multimodal system that aims to detect and quantify semiologic signs recorded in 2D monitoring videos. Our approach can jointly learn semiologic features from facial, body, and hand motions based on computer vision and deep learning architectures. A dataset collected from an Australian quaternary referral epilepsy unit analyzing 161 seizures arising from the temporal (n = 90) and extratemporal (n = 71) brain regions has been used in our system to quantitatively classify these types of epilepsy according to the semiology detected. A leave-one-subject-out (LOSO) cross-validation of semiological patterns from the face, body, and hands reached classification accuracies ranging between 12% and 83.4%, 41.2% and 80.1%, and 32.8% and 69.3%, respectively. The proposed hierarchical multimodal system is a potential stepping-stone towards developing a fully automated semiology analysis system to support the assessment of epilepsy.

Mental Fatigue Estimation Using EEG in a Vigilance Task and Resting States.

Mental fatigue induced by long time mental work can cause deterioration in task performance and increase the risk of accidents. Recently, electroencephalogram (EEG)-based monitoring of mental fatigue has received increasing attention in the field of brain-computer interfaces (BCI). This study aims to employ EEG signals to measure the mental fatigue level by estimating reaction time (RT) in a psychomotor vigilance task (PVT). In a 36-hour sleep deprivation experiment, EEG data from 18 subjects were recorded every four hours in nine blocks, each consisting of three tasks: a 6-minute PVT task and two 3-minute resting states (eyes closed and eyes open). The mean RT in the PVT task showed a generally increasing trend during the 36-hour awake period, reflecting the increase of fatigue over time. For each task, multiple EEG features were extracted and selected to better estimate RT using a multiple linear regression (MLR) method. The correlation between predicted RT and actual RT was evaluated using a leave-one-subject-out (LOSO) validation strategy. After parameter optimization, EEG data from the PVT task obtained a mean correlation coefficient of $0.81 \pm 0.16$ across all subjects. Resting-state EEG data showed lower correlations (eyes-closed: $0.65 \pm 0.20$, eyes-open: $0.50 \pm 0.30)$ partially due to the involvement of shorter data lengths. These results demonstrate the feasibility and robustness of the EEG-based fatigue monitoring method, which could be potential for applications in operational environments.

Motion descriptors for micro-expression recognition

Micro-expression analysis is an interesting and challenging task in computer vision. It has inspired a series of possible applications in many areas such as police–criminal interrogation and important business negotiation. One of the most crucial step in a micro-expression recognition system is the extraction of well discriminating features. In this paper, we propose a new feature based on the fusion of motion boundary histograms (FMBH). This feature is generated by combining both the horizontal and the vertical components of the differential of optical flow as inspired from the motion boundary histograms (MBH). The proposed feature is then validated and evaluated through the leave-one-subject-out (LOSO) protocol for micro-expression recognition. Moreover, the proposed method is compared to state-of-the-art methods on four well-known databases CASME, CASME II, SMIC and CAS(ME)2. Comparative experimental results demonstrate that the proposed FMBH feature descriptor yields promising performance.

Computer-Aided Theragnosis Based on Tumor Volumetric Information in Breast Cancer.

A computer-assisted technology has recently been proposed for the assessment of therapeutic responses to neoadjuvant chemotherapy in patients with locally advanced breast cancer (LABC). The system, however, extracted features from individual scans in a tumor irrespective of its relation to the other scans of the same patient, ignoring the volumetric information. This study addresses this problem by introducing a novel engineered texton-based method in order to account for volumetric information in the design of textural descriptors to represent tumor scans. A noninvasive computer-aided-theragnosis (CAT) system was developed by employing multiparametric QUS spectral and backscatter coefficient maps. The proceeding was composed of two subdictionaries: one built on the "pretreatment" and another on "week " scans, where was 1, 4, or 8. The learned dictionary of each patient was subsequently used to compute the model (histogram of textons) for each scan of the patient. Advanced machine learning techniques including a kernel-based dissimilarity measure to estimate the distances between "pretreatment" and "mid-treatment" scans as an indication of treatment effectiveness, learning from imbalanced data, and supervised learning were subsequently employed on the texton-based features. The performance of the CAT system was tested using statistical tests of significance and leave-one-subject-out (LOSO) classification on 56 LABC patients. The proposed texton-based CAT system indicated significant differences in changes between the responding and nonresponding patient populations and achieved high accuracy, sensitivity, and specificity in discriminating between the two patient groups early after the start of treatment, i.e., on weeks 1 and 4 of several months of treatment. Specifically, the CAT system achieved the area under curve of 0.81, 0.83, and 0.85 on weeks 1, 4, and 8, respectively. The proposed texton-based CAT system accounted for the volumetric information in "pretreatment" and "mid-treatment" scans of each patient. It was demonstrated that this attribute of the CAT system could boost its performance compared to the cases that the features were extracted from solely individual scans.

Automated Detection of Wheelchair Propulsion Using a Single Wrist Accelerometer

PURPOSE: Physical activity (PA) provides important health benefits such as improved cardio-metabolic health, mental health, and cognitive functioning. However, the majority of this evidence is based on research conducted in ambulatory populations. Research informing the relationship between PA and health among manual wheelchair users (MWU’s) is limited. One of the barriers is the lack of valid and reliable PA measures for the population. In the current study, machine learning (ML) techniques were used to develop activity recognition models to automatically identify episodes of active self-propulsion in manual MWU’s wearing a single wrist-mounted accelerometer. METHODS: 11 adult MWU’s (males= 8; 7 paraplegic; 4 tetraplegic) completed a series of activity trials while wearing an ActiGraph GT9X accelerometer on the non-dominant wrist. Activities included: sitting quietly, being pushed, self-propulsion, and completing manual tasks such as drinking water, working on an iPad, and folding laundry. Trials were categorised into 3 classes: sedentary (SED), manual tasks (MT), and self-propulsion (SP). 15 time-domain features from the X, Y, and Z axis were extracted from 1 s windows with 50% overlap and inputted into 3 supervised learning algorithms Decision Tree (DT), Random Forest (RF), and Support Vector Machine (SVM). Performance was evaluated using leave-one-subject-out (LOSO) cross validation. To determine if the resultant models generalized to new data, performance was also evaluated in an independent sample of MWU’s (n = 14). RESULTS: Cross-validation F1-scores for the DT, RF, and SVM classifiers were 0.83, 0.84 and 0.85, respectively. Classification accuracy was consistently good to excellent for SED (86.0% - 92.7%), MT (76.0% - 82.4%), and SP (76.0% - 76.8%). In the independent sample, F1-scores for the DT, RF, and SVM classifiers were, 0.80, 0.81, and 0.82, respectively. Classification accuracy remained good to excellent for SED (83.9% - 92.0%), MT (70.5% - 79.3%), and SP (74.2% - 77.6%) CONCLUSION: ML models trained on simple time-domain features from a single wrist-worn accelerometer can be used to differentiate active self-propulsion from other activities in MWU’s. The models generalized well to new data and could help researchers evaluate the effectiveness of interventions to promote PA in MWU’s.

GeriActive: Wearable App for Monitoring and Encouraging Physical Activity among Older Adults.

The ability to monitor a person's level of daily activity can inform self-management of physical activity and assist in augmenting behavioral interventions. For older adults, the importance of regular physical activity is critical to reduce the risk of long-term disability. In this work, we present GeriActive, an application on the Amulet wrist-worn device that monitors in real time older adults' daily activity levels (low, moderate and vigorous), which we categorized using metabolic equivalents (METs). The app implements an activity-level detection model we developed using a linear Support Vector Machine (SVM). We trained our model using data from volunteer subjects (n=29) who performed common physical activities (sit, stand, lay down, walk and run) and obtained an accuracy of 94.3% with leave-one-subject-out (LOSO) cross-validation. We ran a week-long field study to evaluate the usability and battery life of the GeriActive system where 5 older adults wore the Amulet as it monitored their activity level. Their feedback showed that our system has the potential to be usable and useful. Our evaluation further revealed a battery life of at least 1 week. The results are promising, indicating that the app may be used for activity-level monitoring by individuals or researchers for health delivery interventions that could improve the health of older adults.