Temporal feature extraction and clustering analysis of electromyographic linear envelopes in gait studies

Movement pattern and upper extremity muscle activation during fast and slow continuous steering movement

Continuous steering movement (CSM) is an essential component of the upper extremity (UE) task during vehicle driving, and could be a suitable candidate for multi-joint rehabilitation programs for patients with UE disabilities. This study aims to evaluate the UE muscle activation during CSM and how the rotating speed and direction affect CSM's kinematic and kinetic performance. Surface electromyography (EMG), hand contact information, and steering torque were measured under fast (180°/s) and slow (60°/s) constant-velocity CSM to reveal the activation of shoulder and elbow muscles, temporal characteristics, and force exertion during the stance and swing phases of a CSM cycle. Data from 24 normal young adults showed that shorter contact duration but higher force exertion occurred in the hand moving in an outward steering direction during only fast CSM in either the clockwise (CW) or counterclockwise (CCW) direction. During a steering cycle (either fast or slow speed), the triceps brachii, sternal part of the pectoralis major (PS), and posterior deltoid play major roles in generating steering torque in the CW direction of the CSM. In contrast, the PS, clavicular part of the pectoralis major (PC), and anterior deltoid (AD) largely contribute to torque generation during the CCW CSM. During the swing phase of CSM, AD, PC, and PS are the major muscles that move the hand for the next grasping of the steering wheel in all four conditions. Using the mean activation profiles of the major contributing muscles, the functional roles of these elbow and shoulder muscles were analyzed and are discussed herein. These findings help us to further understand the activation patterns of UE muscles and the kinematic and kinetic changes during two rotating directions and two speeds of CSM, and suggest important implications for future practice in clinical training.

EMG analysis across different tasks improves prevention screenings in diabetes: a cluster analysis approach

The aim of this work was twofold: on one side to determine the most suitable parameters of surface electromyography (sEMG) to classify diabetic subjects with and without neuropathy and discriminate them from healthy controls and second to assess the role of the task acquired in the classification process. For this purpose 30 subjects were examined (10 controls, 10 diabetics with and 10 without neuropathy) whilst walking and stair ascending and descending. The electrical activity of six muscles was recorded bilaterally through a 16-channel sEMG system synchronised with a stereophotogrammetric system: Rectus Femoris, Gluteus Medius, Tibialis Anterior, Peroneus Longus, Gastrocnemius Lateralis and Extensor Digitorum. Spatiotemporal parameters of gait and stair climbing and the following sEMG parameters were extracted: signal envelope, activity duration, timing of activation and deactivation. A hierarchical clustering algorithm was applied to the whole set of parameters with different distances and linkage methods. Results showed that only by applying the Ward agglomerative hierarchical clustering (Hamming distance) to the all set of parameters extracted from both tasks, 5 well-separated clusters were obtained: cluster 3 included only DS subjects, cluster 2 and 4 only controls and cluster 1 and 5 only DNS subjects. This method could be used for planning rehabilitation treatments.

Graphical abstract

A Novel Phonology- and Radical-Coded Chinese Sign Language Recognition Framework Using Accelerometer and Surface Electromyography Sensors

Sign language recognition (SLR) is an important communication tool between the deaf and the external world. It is highly necessary to develop a worldwide continuous and large-vocabulary-scale SLR system for practical usage. In this paper, we propose a novel phonology- and radical-coded Chinese SLR framework to demonstrate the feasibility of continuous SLR using accelerometer (ACC) and surface electromyography (sEMG) sensors. The continuous Chinese characters, consisting of coded sign gestures, are first segmented into active segments using EMG signals by means of moving average algorithm. Then, features of each component are extracted from both ACC and sEMG signals of active segments (i.e., palm orientation represented by the mean and variance of ACC signals, hand movement represented by the fixed-point ACC sequence, and hand shape represented by both the mean absolute value (MAV) and autoregressive model coefficients (ARs)). Afterwards, palm orientation is first classified, distinguishing “Palm Downward” sign gestures from “Palm Inward” ones. Only the “Palm Inward” gestures are sent for further hand movement and hand shape recognition by dynamic time warping (DTW) algorithm and hidden Markov models (HMM) respectively. Finally, component recognition results are integrated to identify one certain coded gesture. Experimental results demonstrate that the proposed SLR framework with a vocabulary scale of 223 characters can achieve an averaged recognition accuracy of 96.01% ± 0.83% for coded gesture recognition tasks and 92.73% ± 1.47% for character recognition tasks. Besides, it demonstrats that sEMG signals are rather consistent for a given hand shape independent of hand movements. Hence, the number of training samples will not be significantly increased when the vocabulary scale increases, since not only the number of the completely new proposed coded gestures is constant and limited, but also the transition movement which connects successive signs needs no training samples to model even though the same coded gesture performed in different characters. This work opens up a possible new way to realize a practical Chinese SLR system.

Taxonomy of golf putting: do different golf putting techniques exist?

This study is a preliminary investigation into the use of cluster analysis to determine if different putting techniques existed in a group of club level golfers. Putting at a hole 4 m away, the performances of 34 experienced golfers (age 55.3 ± 17.8 years and handicap 15.3 ± 6.9, range 3-27) were analysed using putter head kinematic and centre of pressure data. Two distinct putting techniques were identified (named as Arm putting and Body putting), this being the first time different putting techniques have been reported in the research literature. These techniques were defined by parameters related solely to movement of the centre of pressure along the line of the putt. Some players (17 of 34) moved between techniques when performing their putting trials. Neither technique produced more accurate putt results (P = 0.783). Putting technique was further analysed after grouping players according to handicap (similar skill level) or accuracy (similar putting performance). The lack of significant findings when players were re-analysed according to handicap or accuracy highlights the importance of the correct methodological approach to detecting technique differences.

A study of various normalization procedures for within day electromyographic data

Normalization of electromyographic (EMG) data has been described in the scientific literature as crucial for comparisons between subjects and between muscles. The reference value used in the normalization equation has, however, varied across reports. Comparison between studies could be facilitated by use of a common value. We propose the best way to select the common value is through a reliability approach. Accordingly, the purpose of this study was to identify which of three EMG normalization values provided the most reproducible data set. The gastrocnemius EMG results from 20 normal persons and 20 individuals with anterior cruciate deficiency who were participating in a larger study were normalized to a maximum voluntary isometric contraction (MVIC) EMG, peak dynamic EMG, and mean dynamic EMG. Values were then subjected to evaluation using four statistical measures: inter and intrasubject coefficients of variation (CV), variance ratio (VR), and intraclass correlation coefficient (ICC). The CV measures, while not being reflective of reliability were included for comprehensive consideration in view of other literature. The intersubject CV which measures group variability and the intrasubject CV which measures precision were lower for the dynamic conditions, however, the VR and ICC suggested reproducibility was best with EMG from the MVIC. Given that other studies have advocated normalizing EMG by taking data from the dynamic event, reconsideration may be warranted if high reproducibility is desired. Interpretations of the findings given the population, muscle and condition studied are discussed.

Classification of rhythmic locomotor patterns in electromyographic signals using fuzzy sets

Background

Locomotor control is accomplished by a complex integration of neural mechanisms including a central pattern generator, spinal reflexes and supraspinal control centres. Patterns of muscle activation during walking exhibit an underlying structure in which groups of muscles seem to activate in united bursts. Presented here is a statistical approach for analyzing Surface Electromyography (SEMG) data with the goal of classifying rhythmic "burst" patterns that are consistent with a central pattern generator model of locomotor control.

Methods

A fuzzy model of rhythmic locomotor patterns was optimized and evaluated using SEMG data from a convenience sample of four able-bodied individuals. As well, two subjects with pathological gait participated: one with Parkinson's Disease, and one with incomplete spinal cord injury. Subjects walked overground and on a treadmill while SEMG was recorded from major muscles of the lower extremities. The model was fit to half of the recorded data using non-linear optimization and validated against the other half of the data. The coefficient of determination, R², was used to interpret the model's goodness of fit.

Results

Using four fuzzy burst patterns, the model was able to explain approximately 70-83% of the variance in muscle activation during treadmill gait and 74% during overground gait. When five burst functions were used, one function was found to be redundant. The model explained 81-83% of the variance in the Parkinsonian gait, and only 46-59% of the variance in spinal cord injured gait.

Conclusions

The analytical approach proposed in this article is a novel way to interpret multichannel SEMG signals by reducing the data into basic rhythmic patterns. This can help us better understand the role of rhythmic patterns in locomotor control.

Myoelectric control of a computer animated hand: A new concept based on the combined use of a tree-structured artificial neural network and a data glove

This paper proposes a new learning set-up in the field of control systems for multifunctional hand prostheses. Two male subjects with a traumatic one-hand amputation performed simultaneous symmetric movements with the healthy and the phantom hand. A data glove on the healthy hand was used as a reference to train the system to perform natural movements. Instead of a physical prosthesis with limited degrees of freedom, a virtual (computer-animated) hand was used as the target tool. Both subjects successfully performed seven different motoric actions with the fingers and wrist. To reduce the training time for the system, a tree-structured, self-organizing, artificial neural network was designed. The training time never exceeded 30 seconds for any of the configurations used, which is three to four times faster than most currently used artificial neural network (ANN) architectures.

Analysis of EMG patterns of control subjects and subjects with ACL deficiency during an unanticipated walking cut task

PURPOSE

The purpose of this study was to describe the muscle activation patterns of the vastus lateralis (VL), medial hamstrings (MH) and lateral hamstrings (LH) associated with subjects that were anterior cruciate ligament (ACL) deficient and controls.

METHODS

A total of 54 subjects participated in this study including 25 ACL deficient subjects subdivided into copers (n=9) and non-copers (n=16) using clinical criteria. Muscle activation patterns were recorded at 1000 Hz during an unanticipated side step cut task. The root mean square processed data (time constant 11 ms) were ensemble averaged from 20% of stance before heel strike to toe off. Using the first five harmonics of the Fourier Coefficients as features, muscle activation patterns were divided using a cluster analysis algorithm.

RESULTS

A majority (76-93%) of control subjects used three muscle activation patterns for each muscle. The coper group preferentially used a particular VL and MH activation pattern >2 times more frequently than controls. The non-coper group also preferentially used a MH activation pattern >2 times more frequently than controls and utilized a unique MH and LH activation pattern, distinct from the copers and controls.

CONCLUSIONS

Specific muscle activation patterns distinguish subsets of subjects that are healthy and injured, suggesting possible patterns of muscle activation that contribute to coping status.

Toward an objective interpretation of surface EMG patterns: a voluntary response index (VRI)

Individuals with incomplete spinal cord injuries (SCI) retain varying degrees of voluntary motor control. The complexity of the motor control system and the nature of the recording biophysics have inhibited efforts to develop objective measures of voluntary motor control. This paper proposes the definition and use of a voluntary response index (VRI) calculated from quantitative analysis of surface electromyographic (sEMG) data recorded during defined voluntary movement as a sensitive measure of voluntary motor control in such individuals. The VRI is comprised of two numeric values, one derived from the total muscle activity recorded for the voluntary motor task (magnitude), and the other from the sEMG distribution across the recorded muscles (similarity index (SI)). Calculated as a vector, the distribution of sEMG from the test subject is compared to the average vector calculated from sEMG recordings of the same motor task from 10 neurologically intact subjects in a protocol called brain motor control assessment (BMCA). To evaluate the stability of the VRI, a group of five healthy subjects were individually compared to the prototype, average healthy-subject vectors for all of the maneuvers. To evaluate the sensitivity of this method, the VRI was obtained from two SCI subjects participating in other research studies. One was undergoing supported treadmill ambulation training, and the other a controlled withdrawal of anti-spasticity medications. The supported treadmill training patient's VRI, calculated from pre- and post-training BMCA recordings, reflected the qualitative changes in sEMG patterns and functional improvement of motor control. The VRI of the patient followed by serial BMCA during medication withdrawal also reflected changes in the motor control as a result of changes in anti-spasticity medication. To validate this index for clinical use, serial studies using larger numbers of subjects with compromised motor control should be performed.

Job enlargement and mechanical exposure variability in cyclic assembly work

Cyclic assembly work is known to imply a high risk for musculoskeletal disorders. To have operators rotate between work tasks is believed to be one way of decreasing this risk, since it is expected to increase variation in mechanical and psychological exposures (physical and mental loads). This assumption was investigated by assessing mechanical exposure variability in three assembly tasks in an electronics assembly plant, each on a separate workstation, as well as in a 'job enlargement' scenario combining all three stations. Five experienced operators worked for 1 h on each station. Data on upper trapezius and forearm extensor muscle activity were obtained by means of electromyography (EMG), and working postures of the head and upper arms were assessed by inclinometry. The cycle-to-cycle variance of parameters representing the three exposure dimensions: level, frequency and duration was estimated using ANOVA algorithms for each workstation separately as well as for a balanced combination of all three. For a particular station, the variability of trapezius EMG activity levels relative to the mean was higher than for extensor EMG: between-cycles coefficients of variation (CV) about 0.15 and 0.10, respectively. A similar relationship between CV applied to the parameter describing frequency of EMG activity. Except for head inclination levels, the between-cycles CV was larger for posture parameters than for EMG. The between-cycles variance increased up to six fold in the job enlargement scenario, as compared to working at only one station. The difference in mean exposure between workstations was larger for trapezius EMG parameters than for forearm extensor EMG and postures, and hence the effect of job enlargement on exposure variability was more pronounced for the trapezius. For some stations, job enlargement even implied less cycle-to-cycle variability in forearm extensor EMG parameters than working at that station only. Whether the changes in exposure variability associated with job enlargement were sufficient to imply a decreased risk for musculoskeletal disorders is not known.

Multidimensional EMG-based assessment of walking dynamics

The electromyogram (EMG) provides a measure of a muscle's involvement in the execution of a motor task. Successful completion of an activity, such as walking, depends on the efficient motor control of a group of muscles. In this paper, we present a method to quantify the intricate phasing and activation levels of a group of muscles during gait. At the core of our method is a multidimensional representation of the EMG activity observed during a single stride. This representation is referred to as a "trajectory." A hierarchical clustering procedure is used to identify representative classes of muscle activity patterns. The relative frequencies with which these motor patterns occur during a session (i.e., a series of consecutive strides) are expressed as histograms. Changes in walking strategy will be reflected as changes in the relative frequency with which specific gait patterns occur. This method was evaluated using EMG data obtained during walking on a level and a moderately-inclined treadmill. It was found that the histogram changes due to artificially altered gait are significantly larger than the changes due to normal day-to-day variability.

Electromyographic analysis of locomotion for healthy and hemiparetic subjects--study of performance variability and rail effect on treadmill

This study quantified the performance variability and effect of rail support on the ankle joint for normal and hemiparetic subjects during treadmill walking. Muscle activities of the anterior tibialis (TA) and medial gastrocnemius (MG) from six hemiparetic patients and 14 healthy subjects were assessed with EMG linear envelope and variance ratio of consecutive strides at self-selected cadences. Our results indicate that (1) performance consistency of the hemiparetic patients was significantly undermined; (2) habituated process during treadmill walking was notable in the MG but not in the TA; and (3) rail support could reduce performance variability of the ankle antagonist pairs.

Use of cluster analysis for gait pattern classification of patients in the early and late recovery phases following stroke

The mixture of gait deviations seen in patients following a stroke is remarkably variable. An objective system for classification of gait patterns for this population could be used to guide treatment planning. Quantitated gait analysis was conducted for 47 individuals at admission to in-patient rehabilitation and again at 6 months post-stroke for 42 subjects. Non-hierarchical cluster analysis was used to classify the gait patterns of patients based on the temporal-spatial and kinematic parameters of walking. Four clusters of patients were identified at both assessment intervals. At the admission test walking velocity, peak knee extension in mid stance and peak dorsiflexion in swing were the three factors that best characterized the groups. At 6 months the explanatory variables were velocity, knee extension in terminal stance, and knee flexion in pre swing. Differences in muscle strength and muscle activation patterns during walking were identified between groups.

[Postural strategies and falls in elderly and in parkinsonism]

OBJECTIVE

To use a posture analysis to show the evolution of postural pattern connected with falls.

MATERIAL AND METHOD

It is a prospective study on two groups of 16 persons of more than 60 years. A group concerns 16 small disability off drug parkinsonian patients, a group concerns 16 healthy witnesses. All the persons benefited from a posture recording by means of a force platform and were followed during 1 year.

RESULTS

Data analysis underlines three groups of persons corresponding to three postural patterns, independently of the presence of Parkinson disease. A group (n = 18) did not contain fallers, the second (n = 10 ) contained 20% of fallers, the third (n = 4) contained 100% of fallers. Differences between the groups were identified on 16 posturographic parameters.

DISCUSSION

A group has a good functional value and one does not record any fall. Its characteristics, which correspond to a category of persons who compensate well for the phenomena of ageing, are found in the literature. A group has an intermediate functional value and regrets 20% of fallers. Kinetic profile reveals a tendency to the stiffness of the posture. This group is going to operate rather ankle strategies. A group has an inferior functional value and regrets 100% of fallers. Kinetic profile seems disrupted and not to be able to adapt itself in a satisfactory way to the situation otherwise than by stereotypical reactions. This group is going to operate systematically much less stabilizing hip strategies.

CONCLUSION

A close determinism between physiological neuromotor ageing and Parkinson disease does exist. We showed with a prospective follow-up, the arisen of fall and showed the evolution of postural patterns related to fall. It appears as well that evolution mainly follows three stages leading from a small risk of fall gait pattern to a major risk of fall gait pattern.

Abdominal and erector spinae muscle activity during gait: the use of cluster analysis to identify patterns of activity

OBJECTIVE

To describe patterns of muscle activation during gait in selected abdominal and lumbar muscles using cluster analysis.

PARTICIPANTS

A sample of convenience of 38 healthy adult volunteers. Outcome measures. Electromyographic activity from the right internal and external obliques, rectus abdominis and lumbar erector spinae were recorded, and the root mean square values for each muscle were calculated throughout the stride in 5% epochs. These values were normalised to maximum effort isometric muscle contractions. Cluster analysis was used to identify groups of subjects with similar patterns of activity and activation levels.

RESULTS

Cluster analysis identified two patterns of activity for the internal oblique, external oblique and rectus abdominis muscles. In the lumbar erector spinae, three patterns of activity were observed. In most instances, the patterns observed for each muscle differed in the magnitude of the activation levels. In rectus abdominis and external oblique muscles, the majority of subjects had low levels of activity (<5.0% of a maximum voluntary contraction) that were relatively constant throughout the stride cycle. In the internal oblique and the erector spinae muscles, more distinct bursts of activity were observed, most often close to foot-strike. The different algorithms used for the cluster analysis yielded similar results and a discriminant function analysis provided further evidence to support the patterns observed.

CONCLUSIONS

Cluster analysis was useful in grouping subjects who had similar patterns of muscle activity. It provided evidence that there were subgroups that might otherwise not be observed if a group ensemble was presented as the "norm" for any particular muscle's role during gait.

RELEVANCE

The identification of common variations in muscle activity may prove valuable in identifying individuals with electromyographic patterns that might influence their chances of sustaining injury. Alternatively, clusters may provide important information related to muscle activity in those that do well or otherwise after a particular injury.

Determination of generic body-seat interface shapes by cluster analysis

The purpose of this study was to determine typical or generic shape patterns of the buttock-seat interface for elderly wheelchair users. The group of subjects was composed of 30 elderly people (aged 65 or older) and the shapes of the body-seat interface were measured by the electronic shape sensor (ESS). By analyzing the dissimilarity in geometrical shape descriptors or parameters, four distinct generic shapes were identified by means of the cluster analysis method. The results suggest that the generic shapes were mainly characterized by the lateral symmetry of the shapes. The determination of elderly people's seat interface shapes into distinct clusters may lead to a more comprehensive understanding of the seat support interface and more effective seat cushion designs.

An application of pattern recognition for the comparison of trunk muscles EMG waveforms between subjects with and without chronic low back pain during flexion-extension and lateral bending tasks

The purpose of the study was twofold: (1) to evaluate the reliability of distance measures computed from a principal component analysis (PCA) of electromyographic (EMG) waveforms of trunk muscles recorded during standardized trunk movements and (2) to evaluate their sensitivity to the load lifted, the trunk range of motion (ROM) and to low back status. Thirty-three male subjects (18 normals, 15 suffering from non specific CLBP) aged between 35-45 years participated. The EMG signals from 12 trunk muscles and the kinematics of trunk segments were recorded during 12 tasks. The subjects performed flexion-extension and lateral bending (left and right) tasks (three complete cycles) with and without a 12 kg load and at different trunk ROM (maximal or at defined submaximal angles). Distance measures locating each subject relative to a reference PCA model were computed for each muscle and task. The reliability of these distance measures was evaluated for 10 subjects (five normals and five CLBP) who performed two tasks on three different days. The reliability of distance measures was acceptable for agonist muscles only. The distance measures were sensitive to the load lifted and to the trunk ROM for different muscles and tasks but poorly sensitive to low back status. Several reasons that could explain the low sensitivity of these measures to low back status are discussed and potential solutions are proposed. A procedure based on a reliability analysis is proposed to select the number of principal components to include in the reference PCA model. It is expected that the refinement of the method used in this study could provide an effective clinical tool to assess EMG waveforms of trunk muscles during dynamic tasks.

Gait pattern classification of healthy elderly men based on biomechanical data

OBJECTIVES

To distinguish the gait patterns of young subjects from those of elderly men using three-dimensional (3D) gait data, to determine if elderly subjects displayed other than a typical gait pattern, and to identify which parameters best describe them.

DESIGN

Nonrandomized study in which video and force plate data were collected at the subject's own free walking speed and used in a 3D inverse dynamic model. Cluster analysis was chosen to identify the gait families, and analyses of variance were performed to determine which parameters were different.

SETTING

A gait laboratory.

PARTICIPANTS

The sample of convenience involved a single but mixed group consisting of 16 able-bodied elderly subjects (mean age, 62yrs) and 16 able-bodied young subjects aged between 20 and 35 years.

MAIN OUTCOME MEASURES

Phasic and temporal gait parameters, as well as the 3D muscle powers developed in the joints of the right lower limb during the gait cycle.

RESULTS

The walking patterns in elderly subjects were found to be different from those of the young adults. Three elderly gait families or groups forming a specific gait pattern were identified, and differences were found in the phasic and temporal parameters as well as in 6 peak muscle powers. Four of the peak powers occurred in the sagittal plane, and half of them were related to the hip.

CONCLUSIONS

Biomechanical parameters can be used to classify the gait patterns of young and elderly men using cluster analysis rather than age alone. The muscle powers in elderly subjects are perturbed throughout the gait cycle and not only at push-off. It appears that the plane in which the peak powers occurred was related to their occurrence in the gait cycle. Variability in the gait patterns of elderly subjects could reflect natural adaptations or compensations. These should not be indicative of a deficient gait or be misconstrued as some age-related pathology.

Components of EMG symmetry and variability in parkinsonian and healthy elderly gait

Variability and bilateral symmetry of EMG gait-cycle profiles were studied in parkinsonian and healthy elderly subjects in the gastrocnemius, tibialis anterior, and vastus lateralis muscles. Components reflecting shape and timing were defined by the magnitude and phase of the cross-correlation function between individual stride profiles and the latency corrected ensemble average (LCEA) (variability), and between bilateral LCEAs (symmetry). Statistical significance was set at a confidence level of 0.01 reflecting a Bonferroni adjustment due to multiple measures. Parkinsonian gait was significantly different from the healthy elderly in several measures: increased shape variability and asymmetry in the gastrocnemius and tibialis anterior muscles, and reduced timing variability in the gastrocnemius. A portion of the parkinsonian group participated in a 3 week therapy program where they walked to rhythmic auditory stimulation. Gait parameters shifted toward healthy elderly values in each measure where population differences were found. Significant changes were observed in decreased tibialis anterior shape variability and asymmetry, and gastrocnemius shape variability. Strong trends were also observed in increased gastrocnemius timing variability and reduced bilateral asymmetry. In addition to the expected decreased in variability and asymmetry of healthy elderly, increased timing variability in the gastrocnemius was associated with a more normal gait, possibly reflecting feedback adaptability of muscle activity which may be useful in generating stable locomotion.

Tauberian-prony feature extraction technique for esophageal motility patterns

For the esophageal contractile activity recorded during swallowing, a feature extraction scheme has been developed. It recognizes the time, duration, and amplitudes of local peaks for each peristaltic wave. The method is based on the Tauberian approximation for modeling waveforms as a sum of identically shaped pulses with different time delays and amplitudes. Initial conditions on the pulse properties are set and an optimal solution is sought. The method is completely automated and can be utilized for characterization and classification purposes.

Clustering analysis and pattern discrimination of EMG linear envelopes

A technique has been developed for performing pattern analysis of EMG activities generated during locomotion. In this development it was found that the shapes of the EMG linear envelopes (LE) are mainly determined by their phase spectra; their magnitude spectra are much less important. Autoregressive (AR) parametric models and discrete Fourier transform (DFT) approaches were tested and compared. The latter was proved to be a better way to describe the EMG LE's. Feature extraction and clustering were performed by doing DFT of EMG LE's, extracting part of the phase and magnitude spectra (in less important degree) as features, and using the percent powers to weigh the corresponding harmonics. The approach was applied to the clustering analysis of EMG LE's of normal and anterior cruciate ligament (ACL) injured subjects during walking.