III. METHOD
A. Experiments
In this study, we use wireless inertial sensors attached to
the foot to continuously monitor the motion of the foot
during walking. The sensors used are assembled from the
Micro-AHRS model of sensors manufactured by
MicroStrain, Inc [4]. They are small and lightweight,
measuring 41 mm x 63 mm x 32 mm and weighing 39 grams.
The sensors include a 50-g triaxial accelerometer and 1200
deg/s triaxial rate gyroscope, outputting raw 3D acceleration
and angular rate data. This raw data is transformed into 3D
displacements as previously described using the principal of
“zero velocity updating” [5]. This technique assesses the
sensor drift at the end of every step, i.e. when foot velocity is
zero, and allows the removal of cumulative effects of drift
over one step to obtain accurate integration results [6].
Data were collected from 23 subjects with a clinical
diagnosis of PD attending the UCSF Parkinson’s Disease
Clinic and Research Center in San Francisco, and from 16
age-matched control subjects without history of neurological
disease. Of the subjects diagnosed with PD, 11 had a
clinically significant disturbance of gait, and 12 had no such
disturbance. The wireless inertial sensors were attached to
subjects’ feet using a foot mount as shown in Figure 1(a).
Subjects then walked along a predetermined path along a
hallway in the UCSF clinic. Data were transmitted over
Bluetooth to a nearby handheld PDA and stored on the
device for later processing. In total, collecting data from
sensors on both feet in some cases, 21 PD with significant
gait disturbance, 24 PD with no significant gait disturbance,
and 24 control data points were obtained.Figure 1(b) shows an example of the 3D displacement
signal for a 10-second time segment during the course of an
experiment for one subject. The pitch, roll, and yaw
directions are as given in Figure 1(a).B. Data Analysis
The data analysis is performed for two separate
classification tasks. The first is to detect the presence of PD,
i.e. a binary classification task of distinguishing between PD
and control. The second is to characterize parkinsonian gait,
i.e. a multi-class problem of distinguishing among PD with
significant gait disturbance, PD with no significant gait
disturbance, and control. A similar data analysis algorithm is
employed for both tasks. This algorithm involves multiple
modules (Figure 2). Each analysis module is discussed in
detail in their following subsections.
III. METHODA. ExperimentsIn this study, we use wireless inertial sensors attached tothe foot to continuously monitor the motion of the footduring walking. The sensors used are assembled from theMicro-AHRS model of sensors manufactured byMicroStrain, Inc [4]. They are small and lightweight,measuring 41 mm x 63 mm x 32 mm and weighing 39 grams.The sensors include a 50-g triaxial accelerometer and 1200deg/s triaxial rate gyroscope, outputting raw 3D accelerationand angular rate data. This raw data is transformed into 3Ddisplacements as previously described using the principal of“zero velocity updating” [5]. This technique assesses thesensor drift at the end of every step, i.e. when foot velocity iszero, and allows the removal of cumulative effects of driftover one step to obtain accurate integration results [6].Data were collected from 23 subjects with a clinicaldiagnosis of PD attending the UCSF Parkinson’s DiseaseClinic and Research Center in San Francisco, and from 16age-matched control subjects without history of neurologicaldisease. Of the subjects diagnosed with PD, 11 had aclinically significant disturbance of gait, and 12 had no suchdisturbance. The wireless inertial sensors were attached tosubjects’ feet using a foot mount as shown in Figure 1(a).Subjects then walked along a predetermined path along ahallway in the UCSF clinic. Data were transmitted overBluetooth to a nearby handheld PDA and stored on thedevice for later processing. In total, collecting data fromsensors on both feet in some cases, 21 PD with significantgait disturbance, 24 PD with no significant gait disturbance,and 24 control data points were obtained.Figure 1(b) shows an example of the 3D displacementsignal for a 10-second time segment during the course of anexperiment for one subject. The pitch, roll, and yawdirections are as given in Figure 1(a).B. Data AnalysisThe data analysis is performed for two separateclassification tasks. The first is to detect the presence of PD,i.e. a binary classification task of distinguishing between PDand control. The second is to characterize parkinsonian gait,i.e. a multi-class problem of distinguishing among PD withsignificant gait disturbance, PD with no significant gaitdisturbance, and control. A similar data analysis algorithm isemployed for both tasks. This algorithm involves multiplemodules (Figure 2). Each analysis module is discussed indetail in their following subsections.
การแปล กรุณารอสักครู่..
