Miikka Ermes

Activity classification using realistic data from wearable sensors

Automatic classification of everyday activities can be used for promotion of health-enhancing physical activities and a healthier lifestyle. In this paper, methods used for classification of everyday activities like walking, running, and cycling are described. The aim of the study was to find out how to recognize activities, which sensors are useful and what kind of signal processing and classification is required. A large and realistic data library of sensor data was collected. Sixteen test persons took part in the data collection, resulting in approximately 31 h of annotated, 35-channel data recorded in an everyday environment. The test persons carried a set of wearable sensors while performing several activities during the 2-h measurement session. Classification results of three classifiers are shown: custom decision tree, automatically generated decision tree, and artificial neural network. The classification accuracies using leave-one-subject-out cross validation range from 58 to 97% for custom decision tree classifier, from 56 to 97% for automatically generated decision tree, and from 22 to 96% for artificial neural network. Total classification accuracy is 82% for custom decision tree classifier, 86% for automatically generated decision tree, and 82% for artificial neural network

Detection of Daily Activities and Sports With Wearable Sensors in Controlled and Uncontrolled Conditions

Physical activity has a positive impact on peoples well-being, and it may also decrease the occurrence of chronic diseases. Activity recognition with wearable sensors can provide feedback to the user about his/her lifestyle regarding physical activity and sports, and thus, promote a more active lifestyle. So far, activity recognition has mostly been studied in supervised laboratory settings. The aim of this study was to examine how well the daily activities and sports performed by the subjects in unsupervised settings can be recognized compared to supervised settings. The activities were recognized by using a hybrid classifier combining a tree structure containing a priori knowledge and artificial neural networks, and also by using three reference classifiers. Activity data were collected for 68 h from 12 subjects, out of which the activity was supervised for 21 h and unsupervised for 47 h. Activities were recognized based on signal features from 3-D accelerometers on hip and wrist and GPS information. The activities included lying down, sitting and standing, walking, running, cycling with an exercise bike, rowing with a rowing machine, playing football, Nordic walking, and cycling with a regular bike. The total accuracy of the activity recognition using both supervised and unsupervised data was 89% that was only 1% unit lower than the accuracy of activity recognition using only supervised data. However, the accuracy decreased by 17% unit when only supervised data were used for training and only unsupervised data for validation, which emphasizes the need for out-of-laboratory data in the development of activity-recognition systems. The results support a vision of recognizing a wider spectrum, and more complex activities in real life settings.

Hypothermia-treated cardiac arrest patients with good neurological outcome differ early in quantitative variables of EEG suppression and epileptiform activity.

Objective:To evaluate electroencephalogram-derived quantitative variables after out-of-hospital cardiac arrest. Design:Prospective study. Setting:University hospital intensive care unit. Patients:Thirty comatose adult patients resuscitated from a witnessed out-of-hospital ventricular fibrillation cardiac arrest and treated with induced hypothermia (33°C) for 24 hrs. Interventions:None. Measurements and Main Results:Electroencephalography was registered from the arrival at the intensive care unit until the patient was extubated or transferred to the ward, or 5 days had elapsed from cardiac arrest. Burst-suppression ratio, response entropy, state entropy, and wavelet subband entropy were derived. Serum neuron-specific enolase and protein 100B were measured. The Pulsatility Index of Transcranial Doppler Ultrasonography was used to estimate cerebral blood flow velocity. The Glasgow-Pittsburgh Cerebral Performance Categories was used to assess the neurologic outcome during 6 mos after cardiac arrest. Twenty patients had Cerebral Performance Categories of 1 to 2, one patient had a Cerebral Performance Categories of 3, and nine patients had died (Cerebral Performance Categories of 5). Burst-suppression ratio, response entropy, and state entropy already differed between good (Cerebral Performance Categories 1–2) and poor (Cerebral Performance Categories 3–5) outcome groups (p = .011, p = .011, p = .008) during the first 24 hrs after cardiac arrest. Wavelet subband entropy was higher in the good outcome group between 24 and 48 hrs after cardiac arrest (p = .050). All patients with status epilepticus died, and their wavelet subband entropy values were lower (p = .022). Protein 100B was lower in the good outcome group on arrival at ICU (p = .010). After hypothermia treatment, neuron-specific enolase and protein 100B values were lower (p = .002 for both) in the good outcome group. The Pulsatility Index was also lower in the good outcome group (p = .004). Conclusions:Quantitative electroencephalographic variables may be used to differentiate patients with good neurologic outcomes from those with poor outcomes after out-of-hospital cardiac arrest. The predictive values need to be determined in a larger, separate group of patients.

Mobile Mental Wellness Training for Stress Management: Feasibility and Design Implications Based on a One-Month Field Study

Background Prevention and management of work-related stress and related mental problems is a great challenge. Mobile applications are a promising way to integrate prevention strategies into the everyday lives of citizens. Objective The objectives of this study was to study the usage, acceptance, and usefulness of a mobile mental wellness training application among working-age individuals, and to derive preliminary design implications for mobile apps for stress management. Methods Oiva, a mobile app based on acceptance and commitment therapy (ACT), was designed to support active learning of skills related to mental wellness through brief ACT-based exercises in the daily life. A one-month field study with 15 working-age participants was organized to study the usage, acceptance, and usefulness of Oiva. The usage of Oiva was studied based on the usage log files of the application. Changes in wellness were measured by three validated questionnaires on stress, satisfaction with life (SWLS), and psychological flexibility (AAQ-II) at the beginning and at end of the study and by user experience questionnaires after one week’s and one month’s use. In-depth user experience interviews were conducted after one month’s use to study the acceptance and user experiences of Oiva. Results Oiva was used actively throughout the study. The average number of usage sessions was 16.8 (SD 2.4) and the total usage time per participant was 3 hours 12 minutes (SD 99 minutes). Significant pre-post improvements were obtained in stress ratings (mean 3.1 SD 0.2 vs mean 2.5 SD 0.1, P=.003) and satisfaction with life scores (mean 23.1 SD 1.3 vs mean 25.9 SD 0.8, P=.02), but not in psychological flexibility. Oiva was perceived easy to use, acceptable, and useful by the participants. A randomized controlled trial is ongoing to evaluate the effectiveness of Oiva on working-age individuals with stress problems. Conclusions A feasibility study of Oiva mobile mental wellness training app showed good acceptability, usefulness, and engagement among the working-age participants, and provided increased understanding on the essential features of mobile apps for stress management. Five design implications were derived based on the qualitative findings: (1) provide exercises for everyday life, (2) find proper place and time for challenging content, (3) focus on self-improvement and learning instead of external rewards, (4) guide gently but do not restrict choice, and (5) provide an easy and flexible tool for self-reflection.

Advancing from offline to online activity recognition with wearable sensors

Activity recognition with wearable sensors could motivate people to perform a variety of different sports and other physical exercises. We have earlier developed algorithms for offline analysis of activity data collected with wearable sensors. In this paper, we present our current progress in advancing the platform for the existing algorithms to an online version, onto a PDA. Acceleration data are obtained from wireless motion bands which send the 3D raw acceleration signals via a Bluetooth link to the PDA which then performs the data collection, feature extraction and activity classification. As a proof-of-concept, the online activity system was tested with three subjects. All of them performed at least 5 minutes of each of the following activities: lying, sitting, standing, walking, running and cycling with an exercise bike. The average second-by-second classification accuracies for the subjects were 99%, 97%, and 82 %. These results suggest that earlier developed offline analysis methods for the acceleration data obtained from wearable sensors can be successfully implemented in an online activity recognition application.

Estimating Intensity of Physical Activity: A Comparison of Wearable Accelerometer and Gyro Sensors and 3 Sensor Locations

Automatic estimation of physical activity using wearable sensors can be used for promotion of a healthier lifestyle. In this study, accelerometers and gyroscopes attached to ankle, wrist and hip were used to estimate intensity of physical activity. The estimates are compared to metabolic equivalent (MET) obtained from a portable cardiopulmonary exercise testing system. Data from common everyday tasks and exercise were collected with 11 subjects. The tasks include, e.g., ironing, vacuuming, walking, running and cycling on exercise bicycle (ergometer). The strongest linear correlation with metabolic equivalent was obtained with the tri-axial accelerometer attached to the ankle (r=0.86).

Automatic feature selection for context recognition in mobile devices

In mobile devices there exist several in-built sensor units and sources which provide data for context reasoning. More context sources can be attached via wireless network connections. Usually, the mobile devices and the context sources are battery powered and their computational and space resources are limited. This sets special requirements for the context recognition algorithms. In this paper, several classification and automatic feature selection algorithms are compared in the context recognition domain. The main goal of this study is to investigate how much advantage can be achieved by using sophisticated and complex classification methods compared with a simple method that can easily be implemented in mobile devices. The main result is that even a simple linear classification algorithm can achieve a reasonably good accuracy if the features calculated from raw data are selected in a suitable way. Usually context recognition algorithms are fitted to a particular problem instance in an off-line manner and modifying methods for on-line learning is difficult or impossible. An on-line version of the Minimum-distance classifier is presented in this paper and it is justified that it leads to considerably higher classification accuracies compared with the static off-line version of the algorithm. Moreover, we report superior performance for the Minimum-distance classifier compared to other classifiers from the view point of computational load and power consumption of a smart phone.

The effectiveness and applicability of different lifestyle interventions for enhancing wellbeing: the study design for a randomized controlled trial for persons with metabolic syndrome risk factors and psychological distress

BackgroundObesity and stress are among the most common lifestyle-related health problems. Most of the current disease prevention and management models are not satisfactorily cost-effective and hardly reach those who need them the most. Therefore, novel evidence-based controlled interventions are necessary to evaluate models for prevention and treatment based on self-management. This randomized controlled trial examines the effectiveness, applicability, and acceptability of different lifestyle interventions with individuals having symptoms of metabolic syndrome and psychological distress. The offered interventions are based on cognitive behavioral approaches, and are designed for enhancing general well-being and supporting personalized lifestyle changes.Methods/Design339 obese individuals reporting stress symptoms were recruited and randomized to either (1) a minimal contact web-guided Cognitive Behavioral Therapy-based (CBT) intervention including an approach of health assessment and coaching methods, (2) a mobile-guided intervention comprising of mindfulness, acceptance and value-based exercises, (3) a face-to-face group intervention using mindfulness, acceptance and value-based approach, or (4) a control group. The participants were measured three times during the study (pre = week 0, post = week 10, and follow-up = week 36). Psychological well-being, lifestyles and habits, eating behaviors, and user experiences were measured using online surveys. Laboratory measurements for physical well-being and general health were performed including e.g. liver function, thyroid glands, kidney function, blood lipids and glucose levels and body composition analysis. In addition, a 3-day ambulatory heart rate and 7-day movement data were collected for analyzing stress, recovery, physical activity, and sleep patterns. Food intake data were collected with a 48 -hour diet recall interview via telephone. Differences in the effects of the interventions would be examined using multiple-group modeling techniques, and effect-size calculations.DiscussionThis study will provide additional knowledge about the effects of three low intensity interventions for improving general well-being among individuals with obesity and stress symptoms. The study will show effects of two technology guided self-help interventions as well as effect of an acceptance and value–based brief group intervention. Those who might benefit from the aforesaid interventions will increase knowledge base to better understand what mechanisms facilitate effects of the interventions.Trial registrationCurrent Clinical Trials NCT01738256, Registered 17 August, 2012.

Quantification of Epileptiform Electroencephalographic Activity during Sevoflurane Mask Induction

Background:Sevoflurane may induce epileptiform electroencephalographic activity leading to unstable Bispectral Index numbers, underestimating the hypnotic depth of anesthesia. The authors developed a method for the quantification of epileptiform electroencephalographic activity during sevoflurane anesthesia. Methods:Electroencephalographic data from 60 patients under sevoflurane mask induction were used in the analysis. Electroencephalographic data were visually classified. A novel electroencephalogram-derived quantity, wavelet subband entropy (WSE), was developed. WSE variables were calculated from different frequency bands. Performance of the WSE in detection and quantification of epileptiform electroencephalographic activity and the ability of the WSE to recognize misleading Bispectral Index readings caused by epileptiform activity were evaluated. Results:Two WSE variables were found to be sufficient for the quantification of epileptiform activity: WSE from the frequency bands 4–16 and 16–32 Hz. The lower frequency band was used for monophasic pattern monitoring, and the higher frequency band was used for spike activity monitoring. WSE values of the lower and higher bands followed the time evolution of epileptiform activity with prediction probabilities of 0.809 (SE, 0.007) and 0.804 (SE, 0.007), respectively. In deep anesthesia with epileptiform activity, WSE detected electroencephalographic patterns causing Bispectral Index readings greater than 60, with event sensitivity of 97.1%. Conclusions:The developed method proved useful in detection and quantification of epileptiform electroencephalographic activity during sevoflurane anesthesia. In the future, it may improve the understanding of electroencephalogram-derived information by assisting in recognizing misleading readings of depth-of-anesthesia monitors. The method also may assist in minimizing the occurrence of epileptiform activity and seizures during sevoflurane anesthesia.

Weight rhythms: weight increases during weekends and decreases during weekdays.

Background/Aims: The weeks cycle influences sleep, exercise, and eating habits. An accurate description of weekly weight rhythms has not been reported yet - especially across people who lose weight versus those who maintain or gain weight. Methods: The daily weight in 80 adults (BMI 20.0-33.5 kg/m2; age, 25-62 years) was recorded and analysed to determine if a group-level weekly weight fluctuation exists. This was a retrospective study of 4,657 measurements during 15-330 monitoring days. Semi-parametric regression was used to model the rhythm. Results: A pattern of daily weight changes was found (p < 0.05), with higher weight early in the week (Sunday and Monday) and decreasing weight during the week. Increases begin on Saturday and decreases begin on Tuesday. This compensation pattern was strongest for those who lost or maintained weight and weakest for those who slowly gained weight. Conclusion: Weight variations between weekends and weekdays should be considered as normal instead of signs of weight gain. Those who compensate the most are most likely to either lose or maintain weight over time. Long-term habits may make more of a difference than short-term splurges. People prone to weight gain could be counselled about the importance of weekday compensation.