Advances in Neural Signal Processing
40
specific joint based data for discriminating male and female characteristics in gait.
Reliably using EEG to predict swing and stance will include comparisons of temporal and spectral components although the resolutions and accuracies are not so
reliable beyond basic gait changes, we find the positive and negative amplitudes of
the MRCPs can serve as good discriminators.
Gait data was classified using machine learning algorithms with percentage split
cross-validations. As with many datasets, with increase in training data samples, a
consequential increase in the accuracy was observed. Among the algorithms Naïve
Bayes, SVM and tree-based algorithms showed high accuracy across the data with
validations based on different percentage splits of training data. The data from
accelerometers may be used in the BCI-related predictive algorithms for gait phase
estimates.
The study computed joint torques in order to understand relationship of joint
rotations during gait phases. As indicated, generated torque amplitude was sufficient to test classification algorithms on accelerometer-based gait data. We analysed
the data grouped based on the subject weight since average torque amplitude of
each subject was dependent on the weight of the subject. As the weight of the
subject increased, increments in the joint torques were observed across the subjects.
The torques and forces within subjects during different gait cycles showed little
difference.
In terms of gait data from accelerometers, male subjects showed variations in
the frontal and sagittal axes and estimates suggested higher joint movement correlated to higher torque amplitude changes with respect to motion. Hip and ankle
joints served as strong discriminators in classification of subject gender based on
data. Rather than acceleration, torques classified variations of gait across male and
female subjects.
EEG-gait methodology allowed to map cortical organization relationships and
between the contralateral and ipsilateral joints during gait. During stance when
compared to swing, there was higher activity in the delta and theta bands in the
frontal and parietal regions, whereas decreased activity in beta band in the parietal
regions. Using delta and beta rhythms in the fronto-parietal cortical microzones, it
may be possible to classify swing and stance. Additionally, gait-based assessments
need to rely on motor related cortical potentials and their amplitudes. Temporal
analysis of gait related potentials has shown positive and negative motor potentials
for stance and swing and their significant variety could be related as a marker
discriminating stance and swing.
The significance of such assessments is many; with gait categorization using
torque, it may now be possible to employ mobile phone accelerometers to estimate
swing and stance variations as a preclinical step for estimating medical disorders.
The variations could also allow gait as a biometric information especially in validating male and female subjects and their upright walking capabilities. Although EEG
data is far from assessing gait intent, initialization, swing and stance phases may
be explored for correlations related to neurophysiological changes attributing such
data for classifying neurological disorders in the future.
5. Conclusions
Spatio-temporal reconstruction of swing and stance from triaxial accelerometers allow an understanding of how multi-position accelerometer data accounts for
healthy gait before developing optimizations and methods to assess dysfunctional
gait. The study suggests quantifying specific torque patterns during gait may
facilitate cheaply and easily detecting gait phase changes. Although a more detailed
40
specific joint based data for discriminating male and female characteristics in gait.
Reliably using EEG to predict swing and stance will include comparisons of temporal and spectral components although the resolutions and accuracies are not so
reliable beyond basic gait changes, we find the positive and negative amplitudes of
the MRCPs can serve as good discriminators.
Gait data was classified using machine learning algorithms with percentage split
cross-validations. As with many datasets, with increase in training data samples, a
consequential increase in the accuracy was observed. Among the algorithms Naïve
Bayes, SVM and tree-based algorithms showed high accuracy across the data with
validations based on different percentage splits of training data. The data from
accelerometers may be used in the BCI-related predictive algorithms for gait phase
estimates.
The study computed joint torques in order to understand relationship of joint
rotations during gait phases. As indicated, generated torque amplitude was sufficient to test classification algorithms on accelerometer-based gait data. We analysed
the data grouped based on the subject weight since average torque amplitude of
each subject was dependent on the weight of the subject. As the weight of the
subject increased, increments in the joint torques were observed across the subjects.
The torques and forces within subjects during different gait cycles showed little
difference.
In terms of gait data from accelerometers, male subjects showed variations in
the frontal and sagittal axes and estimates suggested higher joint movement correlated to higher torque amplitude changes with respect to motion. Hip and ankle
joints served as strong discriminators in classification of subject gender based on
data. Rather than acceleration, torques classified variations of gait across male and
female subjects.
EEG-gait methodology allowed to map cortical organization relationships and
between the contralateral and ipsilateral joints during gait. During stance when
compared to swing, there was higher activity in the delta and theta bands in the
frontal and parietal regions, whereas decreased activity in beta band in the parietal
regions. Using delta and beta rhythms in the fronto-parietal cortical microzones, it
may be possible to classify swing and stance. Additionally, gait-based assessments
need to rely on motor related cortical potentials and their amplitudes. Temporal
analysis of gait related potentials has shown positive and negative motor potentials
for stance and swing and their significant variety could be related as a marker
discriminating stance and swing.
The significance of such assessments is many; with gait categorization using
torque, it may now be possible to employ mobile phone accelerometers to estimate
swing and stance variations as a preclinical step for estimating medical disorders.
The variations could also allow gait as a biometric information especially in validating male and female subjects and their upright walking capabilities. Although EEG
data is far from assessing gait intent, initialization, swing and stance phases may
be explored for correlations related to neurophysiological changes attributing such
data for classifying neurological disorders in the future.
5. Conclusions
Spatio-temporal reconstruction of swing and stance from triaxial accelerometers allow an understanding of how multi-position accelerometer data accounts for
healthy gait before developing optimizations and methods to assess dysfunctional
gait. The study suggests quantifying specific torque patterns during gait may
facilitate cheaply and easily detecting gait phase changes. Although a more detailed