Christian Doncarli

An online kernel change detection algorithm

A number of abrupt change detection methods have been proposed in the past, among which are efficient model-based techniques such as the Generalized Likelihood Ratio (GLR) test. We consider the case where no accurate nor tractable model can be found, using a model-free approach, called Kernel change detection (KCD). KCD compares two sets of descriptors extracted online from the signal at each time instant: The immediate past set and the immediate future set. Based on the soft margin single-class Support Vector Machine (SVM), we build a dissimilarity measure in feature space between those sets, without estimating densities as an intermediary step. This dissimilarity measure is shown to be asymptotically equivalent to the Fisher ratio in the Gaussian case. Implementation issues are addressed; in particular, the dissimilarity measure can be computed online in input space. Simulation results on both synthetic signals and real music signals show the efficiency of KCD.

Compression of Biomedical Signals With Mother Wavelet Optimization and Best-Basis Wavelet Packet Selection

We propose a novel scheme for signal compression based on the discrete wavelet packet transform (DWPT) decompositon. The mother wavelet and the basis of wavelet packets were optimized and the wavelet coefficients were encoded with a modified version of the embedded zerotree algorithm. This signal dependant compression scheme was designed by a two-step process. The first (internal optimization) was the best basis selection that was performed for a given mother wavelet. For this purpose, three additive cost functions were applied and compared. The second (external optimization) was the selection of the mother wavelet based on the minimal distortion of the decoded signal given a fixed compression ratio. The mother wavelet was parameterized in the multiresolution analysis framework by the scaling filter, which is sufficient to define the entire decomposition in the orthogonal case. The method was tested on two sets of ten electromyographic (EMG) and ten electrocardiographic (ECG) signals that were compressed with compression ratios in the range of 50%-90%. For 90% compression ratio of EMG (ECG) signals, the percent residual difference after compression decreased from (mean ) 48.69.9% (21.58.4%) with discrete wavelet transform (DWT) using the wavelet leading to poorest performance to 28.43.0% (6.71.9%) with DWPT, with optimal basis selection and wavelet optimization. In conclusion, best basis selection and optimization of the mother wavelet through parameterization led to substantial improvement of performance in signal compression with respect to DWT and randon selection of the mother wavelet. The method provides an adaptive approach for optimal signal representation for compression and can thus be applied to any type of biomedical signal.

An online support vector machine for abnormal events detection

The ability to detect online abnormal events in signals is essential in many real-world signal processing applications. Previous algorithms require an explicit signal statistical model, and interpret abnormal events as statistical model abrupt changes. Corresponding implementation relies on maximum likelihood or on Bayes estimation theory with generally excellent performance. However, there are numerous cases where a robust and tractable model cannot be obtained, and model-free approaches need to be considered. In this paper, we investigate a machine learning, descriptor-based approach that does not require an explicit descriptors statistical model, based on support vector novelty detection. A sequential optimization algorithm is introduced. Theoretical considerations as well as simulations on real signals demonstrate its practical efficiency.

Optimization of wavelets for classification of movement-related cortical potentials generated by variation of force-related parameters

The paper presents a novel pattern recognition approach for the classification of single-trial movement-related cortical potentials (MRCPs) generated by variations of force-related parameters during voluntary tasks. The feature space was built from the coefficients of a discrete dyadic wavelet transformation. Mother wavelet parameterization allowed the tuning of basis functions to project the signals. The mother wavelet was optimized to minimize the classification error estimated from the training set. Classification was performed with a support vector machine (SVM) approach with optimization of the width of a Gaussian kernel and of the regularization parameter. The efficacy of the optimization procedures was representatively shown on electroencephalographic recordings from two subjects who performed unilateral isometric plantar flexions at two target torques and two rates of torque development. The proposed classification method was tested on four pairs of classes corresponding to the change in only one of the two parameters of the task. Misclassification rate (test set) in the classification of 1-s EEG activity immediately before the onset of the tasks was reduced from 50.8+/-2.9% with worst wavelet and nearest representative classifier, to 40.2+/-7.3% with optimal wavelet and nearest representative classifier, and to 15.8+/-3.4% with optimal wavelet and SVM with optimization of the kernel and regularization parameter. The proposed pattern recognition method is promising for classification of MRCPs modulated by variations of force-related parameters.

Blind separation of linear instantaneous mixtures of nonstationary surface myoelectric signals

Electromyographic (EMG) recordings detected over the skin may be mixtures of signals generated by different active muscles due to the phenomena related to volume conduction. Separation of the sources is necessary when single muscle activity has to be detected. Signals generated by different muscles may be considered uncorrelated but in general overlap in time and frequency. Under certain assumptions, mixtures of surface EMG signals can be considered as linear instantaneous but no a priori information about the mixing matrix is available when different muscles are active. In this study, we applied blind source separation (BSS) methods to separate the signals generated by two active muscles during a force-varying task. As the signals are non stationary, an algorithm based on spatial time-frequency distributions was applied on simulated and experimental EMG signals. The experimental signals were collected from the flexor carpi radialis and the pronator teres muscles which could be activated selectively for wrist flexion and rotation, respectively. From the simulations, correlation coefficients between the reference and reconstructed sources were higher than 0.85 for signals largely overlapping both in time and frequency and for signal-to-noise ratios as low as 5 dB. The Choi-Williams and Bessel kernels, in this case, performed better than the Wigner-Ville one. Moreover, the selection of time-frequency points for the procedure of joint diagonalization used in the BSS algorithm significantly influenced the results. For the experimental signals, the interference of the other source in each reconstructed source was significantly attenuated by the application of the BSS method. The ratio between root-mean-square values of the signals from the two sources detected over one of the muscles increased from (mean /spl plusmn/ standard deviation) 2.33/spl plusmn/1.04 to 4.51/spl plusmn/1.37 and from 1.55/spl plusmn/0.46 to 2.72/spl plusmn/0.65 for wrist flexion and rotation, respectively. This increment was statistically significant. It was concluded that the BSS approach applied is promising for the separation of surface EMG signals, with applications ranging from muscle assessment to detection of muscle activation intervals, and to the control of myoelectric prostheses.

Optimized Wavelets for Blind Separation of Nonstationary Surface Myoelectric Signals

Surface electromyography (EMG) signals detected over the skin surface may be mixtures of signals generated by many active muscles due to poor spatial selectivity of the recording. In this paper, we propose a new method for blind source separation (BSS) of nonstationary signals modeled as linear instantaneous mixtures. The method is based on whitening of the observations and rotation of the whitened observations. The rotation is performed by joint diagonalization of a set of spatial wavelet distributions (SWDs). The SWDs depend on the selection of the mother wavelet which can be defined by unconstrained parameters via the lattice parameterization within the multiresolution analysis framework. As the sources are classically supposed to be mutually uncorrelated, the design parameters of the mother wavelet can be blindly optimized by minimizing the average (over time lags) cross correlation between the estimated sources. The method was tested on simulated and experimental surface EMG signals and results were compared with those obtained with spatial time-frequency distributions and with second-order statistics (only spectral information). On a set of simulated signals, for 10-dB signal-to-noise ratio (SNR), the cross-correlation coefficient between original and estimated sources was 0.92plusmn0.07 with wavelet optimization, 0.74plusmn0.09 with the wavelet leading to the poorest performance, 0.85plusmn0.07 with Wigner-Ville distribution, 0.86plusmn0.07 with Choi-Williams distribution, and 0.73plusmn0.05 with second-order statistics. In experimental conditions, when the flexor carpi radialis and pronator teres were concomitantly active for 50% of the time, crosstalk was 55.2plusmn10.0% before BSS and was reduced to 15.2plusmn6.3% with wavelet optimization, 30.1plusmn15.0% with the worst wavelet, 28.3plusmn12.3% with Wigner-Ville distribution, 26.2plusmn12.0% with Choi-Williams distribution, and 35.1plusmn15.5% with second-order statistics. In conclusion, the proposed approach resulted in better performance than previous methods for the separation of nonstationary myoelectric signals.

Signal-dependent wavelets for electromyogram classification

In the study, an efficient method to perform supervised classification of surface electromyogram (EMG) signals is proposed. The method is based on the choice of a relevant representation space and its optimisation with respect to a training set. As EMG signals are the summation of compact-support waveforms (the motor unit action potentials), a natural tool for their representation is the discrete dyadic wavelet transform. The feature space was thus built from the marginals of a discrete wavelet decomposition. The mother wavelet was designed to minimise the probability of classification error estimated on the learning set (supervised classification). As a representative example, the method was applied to simulate surface EMG signals generated by motor units with different degrees of short-term synchronisation. The proposed approach was able to distinguish surface EMG signals with degrees of synchronisation that differed by 10%, with a misclassification rate of 8%. The performance of a spectral-based classification (error rate approximately 33%) and of the classification with Daubechies wavelet (21%) was significantly poorer than with the proposed wavelet optimisation. The method can be used for a number of different application fields of surface EMG classification, as the feature space is adapted to the characteristics of the signal that discriminate between classes.

Classification of chirp signals using hierarchical Bayesian learning and MCMC methods

This paper addresses the problem of classifying chirp signals using hierarchical Bayesian learning together with Markov chain Monte Carlo (MCMC) methods. Bayesian learning consists of estimating the distribution of the observed data conditional on each class from a set of training samples. Unfortunately, this estimation requires to evaluate intractable multidimensional integrals. This paper studies an original implementation of hierarchical Bayesian learning that estimates the class conditional probability densities using MCMC methods. The performance of this implementation is first studied via an academic example for which the class conditional densities are known. The problem of classifying chirp signals is then addressed by using a similar hierarchical Bayesian learning implementation based on a Metropolis-within-Gibbs algorithm.

Supervised classification using MCMC methods

This paper addresses the problem of supervised classification using general Bayesian learning. General Bayesian learning consists of estimating the unknown class-conditional densities from a set of labelled samples. However, the estimation requires to evaluate intractable multidimensional integrals. This paper studies an implementation of general Bayesian learning based on Markov chain Monte Carlo (MCMC) methods.

The European project ‘Neuromuscular assessment in the elderly worker’ (NEW): Achievements in electromyogram signal acquisition, modelling and processing

ll_aboratory for Engineering of the Neuromuscular System, Politecnico di Torino, Italy 2National Institute for the Elderly, Firenze, Italy 3Ecole Centrale de Nantes, France 41-1elmholtz Institute for Biomedical Engineering, Aachen, Germany SSirio Automazione, Torino, Italy 6Roessingh Research and Development, Enschede, The Netherlands 7National Institute for Working Life, Goteborg, Sweden alnstitute of Hygiene and Applied Physiology, ETI-I, Zurich, Switzerland 9University of Brescia, Brescia, Italy 1°National Institute for Occupational Health, Copenhagen, Denmark ~Faculty of Computer Science, University of Maribor, Slovenia