Helping Parkinson’s Patients

B. Cole (PhD ’12), S. Chang (PhD ’09), S. Ganesan (MS ’10), S. Abu Bakr (MS ’11), P. Ozdemir (MS ’11)
Prof. S. Hamid Nawab, Prof. Carlo De Luca (BME), Prof. Serge Roy (NMRC)

Sensor signals from the combined sEMG/ACC sensor attached to the dominant arm as illustrated in top picture. The signals in panel (a) are from a control subject with no dyskinesia or tremor. The signals in panel (b) are from a PD subject with tremor. The signals in panel (c) are from a PD subject with dyskinesia. The shaded regions in panels (b) and (c) indicate intervals where the corresponding movement disorders are absent.
Sensor signals from the combined sEMG/ACC sensor attached to the dominant arm as illustrated in top picture. The signals in panel (a) are from a control subject with no dyskinesia or tremor. The signals in panel (b) are from a PD subject with tremor. The signals in panel (c) are from a PD subject with dyskinesia. The shaded regions in panels (b) and (c) indicate intervals where the corresponding movement disorders are absent.

Background: Uncontrollable movement activity is a major problem in long-term management of approximately 50% of patients with Parkinson’s disease (PD). Currently, physicians rely on patient diaries to monitor these complications. Diaries need to be recorded every 15 minutes, are often inaccurate, have poor time resolution, and are difficult for patients to manage. We have previously built custom hybrid sensors which can be conveniently worn by a PD patient to provide electromyographic (EMG) and triaxial accelerometer (ACC) signals while the patient goes about their normal, everyday life activities. It is of interest to develop algorithms that can accurately detect and characterize movement abnormalities in the midst of the unpredictable real-life intentional movements of the PD patient.

Funding: National Institutes of Health

Description: We have developed algorithms that adaptively invoke dynamic neural networks in order to detect and characterize movement abnormalities in the midst of unpredictable, real-life activities of sensor-wearing PD patients. In Figure 1, we illustrate the type of sensor signals that are obtained when a PD patient is wearing one of our sensors on his forearm. Our algorithms require only a few strategically worn sensors and require no patient-specific training. These algorithms were developed using the IPUS framework for Integrated Processing and Understanding of Signals. Development and application of the IPUS framework has been a major focus of our research group for more than a decade.

Results: The per-second accuracy levels of our algorithms are > 85% for each category of motor abnormality. This stands in marked contrast to previous work where such accuracy levels were (1) obtained only on a per-minute or per-task basis, (2) obtained using patient-specific training, (3) obtained from patients wearing a large number of sensors, and/or (4) obtained from patients who were required to periodically perform standardized activities.

Publications:

Cole BT, Roy SH, De Luca CJ, Nawab, SH, Dynamic neural network detection of tremor and dyskinesia from wearable sensor data. Proc. 32nd Annual International Conference IEEE EMBS, Buenos Aires, Argentina, Sep. 1-4, 2010.

Roy, SH, Cheng, MS, Chang, SS, Moore, J, De Luca, G, Nawab, SH, De Luca, CJ A combined sEMG and accelerometer system for monitoring functional activity in stroke. IEEE Trans Neural Systems Rehab Eng, 17: 1-10. Dec. 2009.