Ernest Nlandu Kamavuako

Optimal wavelets for biomedical signal compression.

Signal compression is gaining importance in biomedical engineering due to the potential applications in telemedicine. In this work, we propose a novel scheme of signal compression based on signal-dependent wavelets. To adapt the mother wavelet to the signal for the purpose of compression, it is necessary to define (1) a family of wavelets that depend on a set of parameters and (2) a quality criterion for wavelet selection (i.e., wavelet parameter optimization). We propose the use of an unconstrained parameterization of the wavelet for wavelet optimization. A natural performance criterion for compression is the minimization of the signal distortion rate given the desired compression rate. For coding the wavelet coefficients, we adopted the embedded zerotree wavelet coding algorithm, although any coding scheme may be used with the proposed wavelet optimization. As a representative example of application, the coding/encoding scheme was applied to surface electromyographic signals recorded from ten subjects. The distortion rate strongly depended on the mother wavelet (for example, for 50% compression rate, optimal wavelet, mean±SD, 5.46±1.01%; worst wavelet 12.76±2.73%). Thus, optimization significantly improved performance with respect to previous approaches based on classic wavelets. The algorithm can be applied to any signal type since the optimal wavelet is selected on a signal-by-signal basis. Examples of application to ECG and EEG signals are also reported.

Real-Time, Simultaneous Myoelectric Control Using Force and Position-Based Training Paradigms

In this paper, the simultaneous real-time control of multiple degrees of freedom (DOF) for myoelectric systems is investigated. The goal of this study, in which ten able-bodied subjects participated, was to directly compare three control paradigms of constrained (force targeted), unconstrained (position targeted) and resisted unconstrained (position targeted) limb contractions. Artificial neural networks (ANNs) were trained for simultaneous myoelectric control of the three degrees of freedom (DOFs) (wrist flexion-extension, abduction-adduction, and pronation-supination) using mirrored bilateral contractions. In the resisted unconstrained experiment, some resistance to movement was provided using flexible wrist braces in order to increase the required level of muscle activation. The force, in constrained experiments, and position, in unconstrained and resisted unconstrained experiments, were measured. The three protocols were compared off-line using estimation accuracies (R2) and online using a real-time computer-based target acquisition test. The constrained control paradigm outperformed the unconstrained method in the abduction-adduction DOF (Rconstrained2 = 90.8 ± 0.6, Runconstrained2 = 85.6 ± 1.6) and pronation-supination DOF ( Rconstrained2 = 88.5 ± 0.9, Runconstrained2 = 82.3 ± 1.6), but no significant difference was found in the flexion-extension DOF. The constrained control method outperformed unconstrained control in two real-time testing metrics including completion time and path efficiency. The constrained method results, however, were not significantly different than those of the resisted unconstrained method (with braces) in both off-line and real-time tests. This suggests that the quality of control using constrained and unconstrained contraction-based myoelectric schemes is not appreciably different when using comparable levels of muscle activation.

Relationship between grasping force and features of single-channel intramuscular EMG signals

The surface electromyographic (sEMG) signal can be used for force prediction and control in prosthetic devices. Because of technological advances on implantable sensors, the use of intramuscular EMG (iEMG) is becoming a potential alternative to sEMG for the control of multiple degrees-of-freedom (DOF). An invasive system is not affected by crosstalk, typical of sEMG, and provides more stable and independent control sites. However, intramuscular recordings provide more local information because of their high selectivity, and may thus be less representative of the global muscle activity with respect to sEMG. This study investigates the capacity of selective single-channel iEMG recordings to represent the grasping force with respect to the use of sEMG with the aim of assessing if iEMG can be an effective method for proportional myoelectric control. sEMG and iEMG were recorded concurrently from 10 subjects who exerted six grasping force profiles from 0 to 25/50N. The linear correlation coefficient between features extracted from iEMG and force was approximately 0.9 and was not significantly different from the degree of correlation between sEMG and force. This result indicates that a selective iEMG recording is representative of the applied grasping force and can be used for proportional control.

Simultaneous and Proportional Force Estimation in Multiple Degrees of Freedom From Intramuscular EMG

This letter investigates simultaneous and proportional estimation of force in 2 degree-of-freedoms (DoFs) from intramuscular electromyography (EMG). Intramuscular EMG signals from three able-bodied subjects were recorded along with isometric forces in multiple DoF from the right arm. The association between five EMG features and force profiles was modeled using an artificial neural network. Correlation coefficients between the measured and the estimated forces were 0.85 ± 0.056 and 0.88 ± 0.05 without and with post processing, respectively. The results showed that force can be estimated in 2 DoFs with high accuracy and that the degree of performance depended on the force function (task) to be estimated.

Influence of the feature space on the estimation of hand grasping force from intramuscular EMG

Abstract The study compares the performance of different combinations of nine features extracted from intramuscular electromyogram (EMG) recordings for the estimation of grasping force within the range 0–100% maximum voluntary contraction (MVC). Single-channel intramuscular EMG was recorded from the flexor digitorum profundus (FDP) muscle from 11 subjects who exerted three force profiles during power grasping. The ability of the features to estimate force with a 1st order polynomial (poly1) and an artificial neural network (ANN) model was assessed using the adjusted coefficient of determination (R2). Willison amplitude (WAMP) and root mean square (RMS) showed the highest R2 (∼0.88) values for poly1. The performance of all the features to predict force significantly increased (P

Surface Versus Untargeted Intramuscular EMG Based Classification of Simultaneous and Dynamically Changing Movements

The pattern recognition-based myoelectric control scheme is in the process of being implemented in clinical settings, but it has been mainly tested on sequential and steady state data. This paper investigates the ability of pattern recognition to resolve movements that are simultaneous and dynamically changing and compares the use of surface and untargeted intramuscular EMG signals for this purpose. Ten able-bodied subjects participated in the study. Both EMG types were recorded concurrently from the right forearm. The subjects were instructed to track dynamic contraction profiles using single and combined degrees of freedom in three trials. During trials one and two, the amplitude and the frequency of the profile were kept constant (nonmodulated data), and during trial three, the two parameters were modulated (modulated data). The results showed that the performance was up to 93% for nonmodulated tasks, but highly depended on the nature of the data used. Surface and untargeted intramuscular EMG had equal performance for data of similar nature (nonmodulated), but the performance of intramuscular EMG decreased, compared to surface, when tested on modulated data. However, the results of intramuscular recordings obtained in this study are promising for future use of implantable electrodes, because, besides the value added in terms of potential chronic implantation, the performance is theoretically the same as for surface EMG provided that enough information is captured in the recordings. Nevertheless, care should be taken when training the system since data obtained from selective recordings probably need more training data to generalize to new signals.

Wrist torque estimation during simultaneous and continuously changing movements: surface vs. untargeted intramuscular EMG

In this paper, the predictive capability of surface and untargeted intramuscular electromyography (EMG) was compared with respect to wrist-joint torque to quantify which type of measurement better represents joint torque during multiple degrees-of-freedom (DoF) movements for possible application in prosthetic control. Ten able-bodied subjects participated in the study. Surface and intramuscular EMG was recorded concurrently from the right forearm. The subjects were instructed to track continuous contraction profiles using single and combined DoF in two trials. The association between torque and EMG was assessed using an artificial neural network. Results showed a significant difference between the two types of EMG (P < 0.007) for all performance metrics: coefficient of determination (R(2)), Pearson correlation coefficient (PCC), and root mean square error (RMSE). The performance of surface EMG (R(2) = 0.93 ± 0.03; PCC = 0.98 ± 0.01; RMSE = 8.7 ± 2.1%) was found to be superior compared with intramuscular EMG (R(2) = 0.80 ± 0.07; PCC = 0.93 ± 0.03; RMSE = 14.5 ± 2.9%). The higher values of PCC compared with R(2) indicate that both methods are able to track the torque profile well but have some trouble (particularly intramuscular EMG) in estimating the exact amplitude. The possible cause for the difference, thus the low performance of intramuscular EMG, may be attributed to the very high selectivity of the recordings used in this study.

A criterion for signal-based selection of wavelets for denoising intrafascicular nerve recordings.

In this paper we propose a novel method for denoising intrafascicular nerve signals with the aim of improving action potential (AP) detection. The method is based on the stationary wavelet transform and thresholding of the wavelet coefficients. Since the choice of the mother wavelet substantially impact the performance, a criterion is proposed for selecting the optimal wavelet. The criterion for selection was based on the root mean square of the average of the output signal triggered by the detected APs. The mother wavelet was parameterized through the scaling filter, which allowed optimization through the proposed criterion. The method was tested on simulated signals and on experimental neural recordings. Experimental signals were recorded from the tibial branch of the sciatic nerve of three anaesthetized New Zealand white rabbits during controlled muscle stretches. The simulation results showed that the proposed method had an equivalent effect on AP detection performance (percentage of correct detection at 6 dB signal-to-noise ratio, mean+/-SD, 95.3+/-5.2%) to the a-posteriori choice of the best wavelet (96.1+/-3.6). Moreover, the AP detection after the proposed denoising method resulted in a correlation of 0.94+/-0.02 between the estimated spike rate and the muscle length. Therefore, the study proposes an effective method for selecting the optimal mother wavelet for denoising neural signals with the aim of improving AP detection.

Comparison of features for movement prediction from single-trial movement-related cortical potentials in healthy subjects and stroke patients

Detection of movement intention from the movement-related cortical potential (MRCP) derived from the electroencephalogram (EEG) signals has shown to be important in combination with assistive devices for effective neurofeedback in rehabilitation. In this study, we compare time and frequency domain features to detect movement intention from EEG signals prior to movement execution. Data were recoded from 24 able-bodied subjects, 12 performing real movements, and 12 performing imaginary movements. Furthermore, six stroke patients with lower limb paresis were included. Temporal and spectral features were investigated in combination with linear discriminant analysis and compared with template matching. The results showed that spectral features were best suited for differentiating between movement intention and noise across different tasks. The ensemble average across tasks when using spectral features was (error = 3.4 ± 0.8%, sensitivity = 97.2 ± 0.9%, and specificity = 97 ± 1%) significantly better (P < 0.01) than temporal features (error = 15 ± 1.4%, sensitivity: 85 ± 1.3%, and specificity: 84 ± 2%). The proposed approach also (error = 3.4 ± 0.8%) outperformed template matching (error = 26.9 ± 2.3%) significantly (P > 0.001). Results imply that frequency information is important for detecting movement intention, which is promising for the application of this approach to provide patient-driven real-time neurofeedback.

Velocity recovery function of the compound muscle action potential assessed with doublet and triplet stimulation

Normative values of muscle fiber conduction velocity depend on the conditions in which conduction velocity is measured due to the velocity recovery function (VRF) of muscle fibers. In this study the VRF of the compound muscle action potential (CMAP) was assessed following doublet and triplet stimulation in order to investigate the effect of repetitive muscle activation on muscle fiber conduction velocity. The VRF from doublet and triplet activation showed a peak of 4.6%–15.0% and 6.4%–25.9%, respectively, which is not significantly different. The VRF of the CMAP with doublet stimulation had a plateau between 25–75 ms, similar to that reported for single muscle fibers, and changed as a consequence of previous activation. The VRFs with doublet and triplet stimulation were different for interstimulus intervals in the range of 12–250 ms, where the triplet resulted in a plateau of supernormal conduction velocity. The VRF of the triplet could be explained by linear summation of the effects from doublet stimulations only for small distances between the two conditioning stimuli. These results provide new information on the adaptation of membrane properties of muscle fibers to repetitive activation. Changes in CMAP properties due to repeated activation may influence the accuracy of techniques based on CMAP recordings, such as collision methods. Muscle Nerve, 2007