A method for classification of movements in bed

Distinguishing positions and movements in bed from load cell signals

OBJECTIVE

To characterize and classify six positions and movements for individuals in a bed using the output signals of four load cell sensors.

APPROACH

A bed equipped with four load cell sensors and synchronized video was used to assess the load cell response of 54 healthy individuals in prescribed positions and as they moved between positions. Stationary positions were characterized by the signals from the four load cells and the coordinates of the center of mass (CoM). Movements were characterized by the changes in load cell signals, four parameters associated with the trajectory of the CoM between the initial and final position (Euclidean distance, length of the trajectory, and the x- and y- variances), and the initial position's CoM coordinates. Classification and decision tree models were used to assess the ability of these parameters to identify specific positions or movements.

MAIN RESULTS

Six positions were classified with an accuracy of 74.9% and six movements were classified with an accuracy of 79.7%.

SIGNIFICANCE

This study demonstrates the feasibility of distinguishing certain positions and movements with load cell sensors. The identification of positions and movements for individuals in bed can be used as a tool in a variety of clinical settings.

Nonnegative matrix factorization and sparse representation for the automated detection of periodic limb movements in sleep

Stroke is a leading cause of death and disability in adults, and incurs a significant economic burden to society. Periodic limb movements (PLMs) in sleep are repetitive movements involving the great toe, ankle, and hip. Evolving evidence suggests that PLMs may be associated with high blood pressure and stroke, but this relationship remains underexplored. Several issues limit the study of PLMs including the need to manually score them, which is time-consuming and costly. For this reason, we developed a novel automated method for nocturnal PLM detection, which was shown to be correlated with (a) the manually scored PLM index on polysomnography, and (b) white matter hyperintensities on brain imaging, which have been demonstrated to be associated with PLMs. Our proposed algorithm consists of three main stages: (1) representing the signal in the time-frequency plane using time-frequency matrices (TFM), (2) applying K-nonnegative matrix factorization technique to decompose the TFM matrix into its significant components, and (3) applying kernel sparse representation for classification (KSRC) to the decomposed signal. Our approach was applied to a dataset that consisted of 65 subjects who underwent polysomnography. An overall classification of 97 % was achieved for discrimination of the aforementioned signals, demonstrating the potential of the presented method.

What Engineering Technology Could Do for Quality of Life in Parkinson's Disease: A Review of Current Needs and Opportunities

Parkinson's disease (PD) involves well-known motor symptoms such as tremor, rigidity, bradykinesia, and altered gait, but there are also nonlocomotory motor symptoms (e.g., changes in handwriting and speech) and even nonmotor symptoms (e.g., disrupted sleep, depression) that can be measured, monitored, and possibly better managed through activity-based monitoring technologies. This will enhance quality of life (QoL) in PD through improved self-monitoring and also provide information that could be shared with a healthcare provider to help better manage treatment. Until recently, nonmotor symptoms ("soft signs") had been generally overlooked in clinical management, yet these are of primary importance to patients and their QoL. Day-to-day variability of the condition, the high variability in symptoms between patients, and the isolated snapshots of a patient in periodic clinic visits make better monitoring essential to the proper management of PD. Continuously monitored patterns of activity, social interactions, and daily activities could provide a rich source of information on status changes, guiding self-correction and clinical management. The same tools can be useful in earlier detection of PD and will improve clinical studies. Remote medical communications in the form of telemedicine, sophisticated tracking of medication use, and assistive technologies that directly compensate for disease-related challenges are examples of other near-term technology solutions to PD problems. Ultimately, a sensor technology is not good if it is not used. The Parkinson's community is a sophisticated early adopter of useful technologies and a group for which engineers can provide near-term gratifying benefits.

Analyzing center of pressure progression during bed exits

This paper presents a new approach for analyzing center of pressure (COP) progression using pressure data collected from a pressure-sensitive array placed under the bed mattress. Pressure data were collected from a young female participant who was healthy and an older 78 year old female participant who had a history of falls. Information relevant to movement direction, time, path trajectory, magnitude and frequency was presented in three dimensional plots and color differentiated displays. When tested on data collected from an older participant who experienced a fall, this method of analyzing COP was able to illustrate distinct differences in bed exit patterns used pre and post fall episode. This analysis approach shows the potential to detect changes in bed exit patterns indicative of a critical health event. Future applications include home monitoring to assist with early intervention in the event of bed mobility decline.

Unobtrusive classification of sleep and wakefulness using load cells under the bed

Poor quality of sleep increases the risk of many adverse health outcomes. Some measures of sleep, such as sleep efficiency or sleep duration, are calculated from periods of time when a patient is asleep and awake. The current method for assessing sleep and wakefulness is based on polysomnography, an expensive and inconvenient method of measuring sleep in a clinical setting. In this paper, we suggest an alternative method of detecting periods of sleep and wake that can be obtained unobtrusively in a patient's own home by placing load cells under the supports of their bed. Specifically, we use a support vector machine to classify periods of sleep and wake in a cohort of patients admitted to a sleep lab. The inputs to the classifier are subject demographic information, a statistical characterization of the load cell derived signals, and several sleep parameters estimated from the load cell data that are related to movement and respiration. Our proposed classifier achieves an average sensitivity of 0.808 and specificity of 0.812 with 90% confidence intervals of (0.790, 0.821) and (0.798, 0.826), respectively, when compared to the "gold-standard" sleep/wake annotations during polysomnography. As this performance is over 27 sleep patients with a wide variety of diagnosis levels of sleep disordered breathing, age, body mass index, and other demographics, our method is robust and works well in clinical practice.

Automated assessment of mobility in bedridden patients

Immobility in older patients is a costly problem for both patients and healthcare workers. The Hierarchical Assessment of Balance and Mobility (HABAM) is a clinical tool able to assess immobile patients and predict morbidity, yet could become more reliable and informative through automation. This paper proposes an algorithm to automatically determine which of three enacted HABAM scores (associated with bedridden patients) had been performed by volunteers. A laptop was used to gather pressure data from three mats placed on a standard hospital bed frame while five volunteers performed three enactments each. A system of algorithms was created, consisting of three subsystems. The first subsystem used mattress data to calculate individual sensor sums and eliminate the weight of the mattress. The second subsystem established a baseline pressure reading for each volunteer and used percentage change to identify and distinguish between two enactments. The third subsystem used calculated weight distribution ratios to determine if the data represented the remaining enactment. The system was tested for accuracy by inputting the volunteer data and recording the assessment output (a score per data set). The system identified 13 of 15 sets of volunteer data as expected. Examination of these results indicated that the two sets of data were not misidentified; rather, the volunteers had made mistakes in performance. These results suggest that this system of algorithms is effective in distinguishing between the three HABAM score enactments examined here, and emphasizes the potential for pervasive computing to improve traditional healthcare.