Mark V. Albert

Fall Classification by Machine Learning Using Mobile Phones

Fall prevention is a critical component of health care; falls are a common source of injury in the elderly and are associated with significant levels of mortality and morbidity. Automatically detecting falls can allow rapid response to potential emergencies; in addition, knowing the cause or manner of a fall can be beneficial for prevention studies or a more tailored emergency response. The purpose of this study is to demonstrate techniques to not only reliably detect a fall but also to automatically classify the type. We asked 15 subjects to simulate four different types of falls–left and right lateral, forward trips, and backward slips–while wearing mobile phones and previously validated, dedicated accelerometers. Nine subjects also wore the devices for ten days, to provide data for comparison with the simulated falls. We applied five machine learning classifiers to a large time-series feature set to detect falls. Support vector machines and regularized logistic regression were able to identify a fall with 98% accuracy and classify the type of fall with 99% accuracy. This work demonstrates how current machine learning approaches can simplify data collection for prevention in fall-related research as well as improve rapid response to potential injuries due to falls.

Wii Fit Balance Board Playing Improves Balance and Gait in Parkinson Disease

To assess the effect of exercise training by using the Nintendo Wii Fit video game and balance board system on balance and gait in adults with Parkinson disease (PD).

Using mobile phones for activity recognition in Parkinson's patients

Mobile phones with built-in accelerometers promise a convenient, objective way to quantify everyday movements and classify those movements into activities. Using accelerometer data we estimate the following activities of 18 healthy subjects and eight patients with Parkinson’s disease: walking, standing, sitting, holding, or not wearing the phone. We use standard machine learning classifiers (support vector machines, regularized logistic regression) to automatically select, weigh, and combine a large set of standard features for time series analysis. Using cross validation across all samples we are able to correctly identify 96.1% of the activities of healthy subjects and 92.2% of the activities of Parkinson’s patients. However, when applying the classification parameters derived from the set of healthy subjects to Parkinson’s patients, the percent correct lowers to 60.3%, due to different characteristics of movement. For a fairer comparison across populations we also applied subject-wise cross validation, identifying healthy subject activities with 86.0% accuracy and 75.1% accuracy for patients. We discuss the key differences between these populations, and why algorithms designed for and trained with healthy subject data are not reliable for activity recognition in populations with motor disabilities.

Hand, belt, pocket or bag: Practical activity tracking with mobile phones

For rehabilitation and diagnoses, an understanding of patient activities and movements is important. Modern smartphones have built in accelerometers which promise to enable quantifying minute-by-minute what patients do (e.g. walk or sit). Such a capability could inform recommendations of physical activities and improve medical diagnostics. However, a major problem is that during everyday life, we carry our phone in different ways, e.g. on our belt, in our pocket, in our hand, or in a bag. The recorded accelerations are not only affected by our activities but also by the phones location. Here we develop a method to solve this kind of problem, based on the intuition that activities change rarely, and phone locations change even less often. A hidden Markov model (HMM) tracks changes across both activities and locations, enabled by a static support vector machine (SVM) classifier that probabilistically identifies activity-location pairs. We find that this approach improves tracking accuracy on healthy subjects as compared to a static classifier alone. The obtained method can be readily applied to patient populations. Our research enables the use of phones as activity tracking devices, without the need of previous approaches to instruct subjects to always carry the phone in the same location.

Measuring generalization of visuomotor perturbations in wrist movements using mobile phones

Recent studies in motor control have shown that visuomotor rotations for reaching have narrow generalization functions: what we learn during movements in one direction only affects subsequent movements into close directions. Here we wanted to measure the generalization functions for wrist movement. To do so we had 7 subjects performing an experiment holding a mobile phone in their dominant hand. The mobile phones built in acceleration sensor provided a convenient way to measure wrist movements and to run the behavioral protocol. Subjects moved a cursor on the screen by tilting the phone. Movements on the screen toward the training target were rotated and we then measured how learning of the rotation in the training direction affected subsequent movements in other directions. We find that generalization is local and similar to generalization patterns of visuomotor rotation for reaching.

Monitoring Functional Capability of Individuals with Lower Limb Amputations Using Mobile Phones

To be effective, a prescribed prosthetic device must match the functional requirements and capabilities of each patient. These capabilities are usually assessed by a clinician and reported by the Medicare K-level designation of mobility. However, it is not clear how the K-level designation objectively relates to the use of prostheses outside of a clinical environment. Here, we quantify participant activity using mobile phones and relate activity measured during real world activity to the assigned K-levels. We observe a correlation between K-level and the proportion of moderate to high activity over the course of a week. This relationship suggests that accelerometry-based technologies such as mobile phones can be used to evaluate real world activity for mobility assessment. Quantifying everyday activity promises to improve assessment of real world prosthesis use, leading to a better matching of prostheses to individuals and enabling better evaluations of future prosthetic devices.

The Applicability of Inertial Motion Sensors for Locomotion and Posture

Wearable sensors are now ubiquitous consumer devices, enabling the general population to track their physical activities - providing estimates of time in each activity and measures such as steps and calories. Clinicians can benefit from this growth in wearable devices to better evaluate patient activity and posture for diagnosis and evaluating outcomes. The technologies involved range from low-grade, inexpensive consumer-oriented sensors to high-grade clinically validated activity monitors with associated analytics suites. In this chapter, we review the relevant technologies and how they can be applied to track locomotion and posture in clinical populations. We will review the form for these devices, the enabling analytics technology, and patient-specific applications. Beyond the limitations of more traditional measures, we will see that wearable devices enable convenient, objective, and continuous information that can assist clinicians in better diagnostics to quantify the impact of therapeutic interventions.

In-lab versus at-home activity recognition in ambulatory subjects with incomplete spinal cord injury

BackgroundAlthough commercially available activity trackers can aid in tracking therapy and recovery of patients, most devices perform poorly for patients with irregular movement patterns. Standard machine learning techniques can be applied on recorded accelerometer signals in order to classify the activities of ambulatory subjects with incomplete spinal cord injury in a way that is specific to this population and the location of the recording—at home or in the clinic.MethodsSubjects were instructed to perform a standardized set of movements while wearing a waist-worn accelerometer in the clinic and at-home. Activities included lying, sitting, standing, walking, wheeling, and stair climbing. Multiple classifiers and validation methods were used to quantify the ability of the machine learning techniques to distinguish the activities recorded in-lab or at-home.ResultsIn the lab, classifiers trained and tested using within-subject cross-validation provided an accuracy of 91.6%. When the classifier was trained on data collected in the lab but tested on at home data, the accuracy fell to 54.6% indicating distinct movement patterns between locations. However, the accuracy of the at-home classifications, when training the classifier with at-home data, improved to 85.9%.ConclusionIndividuals with unique movement patterns can benefit from using tailored activity recognition algorithms easily implemented using modern machine learning methods on collected movement data.

Saccadic gain adaptation is predicted by the statistics of natural fluctuations in oculomotor function

Due to multiple factors such as fatigue, muscle strengthening, and neural plasticity, the responsiveness of the motor apparatus to neural commands changes over time. To enable precise movements the nervous system must adapt to compensate for these changes. Recent models of motor adaptation derive from assumptions about the way the motor apparatus changes. Characterizing these changes is difficult because motor adaptation happens at the same time, masking most of the effects of ongoing changes. Here, we analyze eye movements of monkeys with lesions to the posterior cerebellar vermis that impair adaptation. Their fluctuations better reveal the underlying changes of the motor system over time. When these measured, unadapted changes are used to derive optimal motor adaptation rules the prediction precision significantly improves. Among three models that similarly fit single-day adaptation results, the model that also matches the temporal correlations of the non-adapting saccades most accurately predicts multiple day adaptation. Saccadic gain adaptation is well matched to the natural statistics of fluctuations of the oculomotor plant.

Determining posture from physiological tremor.

The measurement of body and limb posture is important to many clinical and research studies. Current approaches either directly measure posture (e.g., using optical or magnetic methods) or more indirectly measure it by integrating changes over time (e.g., using gyroscopes and/or accelerometers). Here, we introduce a way of estimating posture from movements without requiring integration over time and the resulting complications. We show how the almost imperceptible tremor of the hand is affected by posture in an intuitive way and therefore can be used to estimate the posture of the arm. We recorded postures and tremor of the arms of volunteers. By using only the minor axis in the covariance of hand tremor, we could estimate the angle of the forearm with a standard deviation of about 4° when the subject’s elbow is resting on a table and about 10° when it is off the table. This technique can also be applied as a post hoc analysis on other hand-position data sets to extract posture. This new method allows the estimation of body posture from tremor, is complementary to other techniques, and so can become a useful tool for future research and clinical applications.