Wearable-Sensor System for Monitoring Motor Function

NIH/NIBIB EB007163
PI: Carlo J. De Luca, PhD1 Co-PIs: Serge H. Roy, ScD1, S. Hamid Nawab, PhD1,2, Joe Jabre, MD3 Other Key Persons: L. Gilmore, ABEE1, Samuel Chang, MS1,2, Peter Novak, MD3, Cathi Thomas, RN, MS3
2

BOSTON
UNIVERSITY

NeuroMuscular Research Center, Boston University Electrical and Computer Engineering, College of Engineering, Boston University 3 Department of Neurology, Boston Medical Center
1

Medicine

SCHOOL of

Goal:

To develop a Personal Status Monitor (PSM) that automatically identifies and tracks movement disorders, medication states, and mobility in patients with Parkinsons disease. Project 2: Clinical Application to PD
Data Collection Protocol: patients are monitored using the Personal Status Monitor (PSM) and are videotaped during
this approximately 4-hour period that is timed to coincide with a complete medication cycle (from On to wearing Off). Activities are conducted in a laboratory configured like an apartment. The activities include standardized motor tests used for clinical assessment (e.g. motor scales from UPDRS), scripted functionl tasks (e.g. sit-stand-and-walk) and freeroaming activity.
Movement Disorders
W earable Wireless ireless E M GM S ystem W earable W E G S ystem

Project 1: Technical Infrastructure

Problem: Uncontrollable movement activity is a major problem in long-term management of approximately 50% of patients

with Parkinsons disease. Physicians rely on patient diaries to monitor these complications. Diaries need to be recorded every 15 min, are often inaccurate, have poor time resolution, and are difficult for patients to manage.

Proposed Solution

Medication States
Levodopa Intake Onset Dyskinesia Peak-Dose Dyskinesia End-of-Dose Dyskinesia

Mobility States Sitting Standing Walking Lying Down The PSM will automatically identify transitions between these four mobility states.

Tremor: Oscillatory, rhythmic movement Bradykinesia: Slow movement Akinesia: Inability to initiate a movement Dyskinesia: Spasmodic movement ment t h a

OFF
Off-Period Freezing disorders Bradykinesia t Tremor

The four different move-

OFF
Off-Period Freezing Bradykinesia Tremor

ON

WEARING-OFF

Hardware developed by research team members to acquire and store multi-channel signals from hybrid sensors using wireless technology. Hybrid sensors under development in this project will detect EMG and accelerometer (ACC) signals.

The PSM will automatically identify these four different movement disorders associated with Parkinsons disease and their severity.

The PSM will automatically identify the onset and duration of the patients On, Off, and On-with-Dyskinesia medication phases.

Sensor Data from PD Patient

Tremor While Sitting
E MG 500 ( V) 0 Deltoid -500
E MG 500 (V ) 0 Deltoid -500

Dyskinesia While Sitting

E MG (mV) Quad

Walking While On
1 0 -1 0 3

Walking While Off

1 0 Quad -1 0 AC C -X (G) -3
E MG (mV) AC C -Y (G) 0

A schematic of the proposed Personal Status Monitor (PSM) for identifying movement disorders, medication states, and mobility in patients with Parkinsons disease by the automatic analysis and interpretation of electromyographic (EMG) and accelerometer (ACC) signals recorded from the surface of the body. In this example, the PSM device is monitoring the subject while walking. The proposed system will provide a continuous history and statistical summarization that can be made available to the clinician to help manage

-1 AC C -X (G)-2 -3 0

AC C -Y (G)-1

-2

-1 -2 -3 1 AC C -Y (G ) 0 -1
AC C -X (G )

AC C -X (G) -3

AC C -Y (G) 0

-3

-3

0 AC C -Z (G)-1 -2
E MG 500 ( V) 0 B iceps -500

AC C -Z (G )

1 0 -1

3 AC C -Z (G) 0 -3 1 0 TA -1 3 AC C -X (G) 0 -3
E MG (mV) AC C -Y (G) 0

3 AC C -Z (G) 0 -3 1 0 TA -1 3 AC C -X (G) 0 -3
E MG (mV) AC C -Y (G) 0

E MG (V ) B iceps

500 0 -500 1 0 -1 1 0 -1 0

AC C -X (G) 0

1
AC C -X (G )

-1

Algorithm Development
Planner

AC C -Y (G)-1

AC C -Y (G )

-2 0 0 2 4 6 8 10 12 14

-3

-3

AC C -Z (G)-1

AC C -Z (G ) -1

-2

-2 0 2 4 6 8 Time (s) 10 12 14

3 AC C -Z (G) 0 -3 4 6 8 10 12 14 Time (s) 16 18 20 22

3 0 AC C -Z (G) -3 8 10 12 14 16

T ime (s )

Time (s)

18

20

22

24

26

Knowledge Sources

Assessment level C E D Abnormality level F E D C B A Mobility Epoch. Abnorm. Mobility Assess. Abnorm. Abnorm. Assess. Mobility Abnorm. Epoch Mobility
Raw data

Artificial Neural Network

q1
1 f (q ) 1
r lte Fi

x1
Feature Extraction

Filter
r lte Fi

Sample data from a patient asked to sit quietly and not move. The figure on the left was taken during his Off period when he was experiencing Tremor and on the right during his On w/ Dyskinesia period. Data were recorded from the Anterior Deltoid (Ch1-4) and Biceps brachii (Ch 5-8). EMG signals are in black; ACC signals are in blue.

-1

Sample data from the same patient while walking. The figure on the left was taken during his On period when he walked normally without movement disorders and on the right during his Off period when he had Akinesia or Freezing. Data were recorded from the Quadraceps (Ch1-4) and Tibialis Anterior muscles (Ch 5-8).

...

Fi

Mobility level A F Epoch level

lte r

y q2
-1 1 f (q ) 2
Fi r lte

1 -1

f (y)

P A 44 45 46

Sitting

xN

47

48

49 mins

Preliminary Results
Results from the algorithms provide automatic classification of mobility states and movement disorders in a patient with PD transitioning from Off to On-with-Dyskinesia. Classification is made solely on the basis of EMG and ACC data. Only 4 of the 8 sensors were used for the mobility analysis. No attempts were made to reduce the number of requisite sensors (to less than 8) for identifying movement disorders at this stage of algorithm development. Classification resolution is 1s and the sensitivity is 95% based on the interpretation of videotaped data by movement disorder specialists. The patient is a 73 y.o. male with advanced PD characterized by severe motor fluctuations.

Filter

Standing

Hidden Layer

Output Layer

44

45

46

47

48

49 mins

DATA Acquisition

Blackboard Database

P A

Walking

44

45

46

47

48

49 mins

Blackboard architecture for the PSM application. The blackboard database shown here contains four data representation levels (Signal Epoch, Mobility, Movement Abnormality, and Medication State). Transformations between levels are carried out by a collection of independent programs that are typically called Knowledge Sources and are invoked according to algorithmically defined control plans.

Features from the raw EMG and ACC signals are analyzed using Artificial Neural Networks (ANNs). More specifically, dynamically evolving features for each activity are characterized through TimeDependent Neural Networks. ANN technology is being used in conjunction with Rule-Based Systems technology. Iterative Correlation Analysis is used for correcting initial identifications. These diverse techniques operate collectively through the Blackboard architecture in the previous figure.

P A

Bradykinesia

44

45

46

47

48

49 mins

P A

Dyskinesia

44

45

46

47

48

49 mins

P A

Tremor

44

45

46

47

48

49 mins P = Present A = Absent

Summary Results - Year 2

We have built a custom hybrid surface sensor acquisition system for acquiring EMG and ACC signals from patients and control subjects. Algorithm development has been enhanced for standardized and singular activities using a blackboard-based signal processing system and time-dependent neural networks. Preliminary classification results for patients with Parkinson's disease resulted in 95% sensitivity for mobility and movement disorders during standardized and singular activities. We are currently preparing to acquire sufficient data to enhance the algorithms for monitoring patients during unconstrained free-form activities.

Special thanks to the patients and staff of the Boston University Parkinsons Disease Center