1340 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 50, NO. 12, DECEMBER 2003 A Novel Approach for Estimating Muscle Fiber Conduction Velocity by Spatial and Temporal Filtering of Surface EMG Signals Dario Farina*, Member, IEEE, and Roberto Merletti, Member, IEEE Abstract—We describe a new method for the estimation of muscle fiber conduction velocity (CV) from surface electromyo- graphy (EMG) signals. The method is based on the detection of two surface EMG signals with different spatial filters and on the compensation of the spatial filtering operations by two temporal filters (with CV as unknown parameter) applied to the signals. The transfer functions of the two spatial filters may have different magnitudes and phases, thus the detected signals have not nec- essarily the same shape. The two signals are first spatially and then temporally filtered and are ideally equal when the CV value selected as a parameter in the temporal filters corresponds to the velocity of propagation of the detected action potentials. This approach is the generalization of the classic spectral matching technique. A theoretical derivation of the method is provided together with its fast implementation by an iterative method based on the Newton’s method. Moreover, the lowest CV estimate among those obtained by a number of filter pairs is selected to reduce the CV bias due to nonpropagating signal components. Simulation results indicate that the method described is less sensitive than the classic spectral matching approach to the presence of nonpropagating signals and that the two methods have similar standard deviation of estimation in the presence of additive, white, Gaussian noise. Finally, experimental signals have been collected from the biceps brachii muscle of ten healthy male subjects with an adhesive linear array of eight electrodes. The CV estimates depended on the electrode location with positive bias for the estimates from electrodes close to the innervation or tendon regions, as expected. The proposed method led to significantly lower bias than the spectral matching method in the experimental conditions, confirming the simulation results. Index Terms—Conduction velocity estimation, end-of-fiber com- ponents, spatial filtering, surface electromyography. I. INTRODUCTION M USCLE fiber conduction velocity (CV) indirectly re- flects the contractile properties of motor units (MUs) through the size principle [1] and can be indicative of muscle fatigue [2], [25], [26]. CV has also been shown to change as a consequence of pathological conditions [19], [35] and to be in- Manuscript received October 12, 2002; revised April 3, 2003. This work was supported by the Fondazione “Cassa di Risparmio di Torino” and Compagnia di San Paolo, Torino, Italy under the European Shared-Cost Project Neuromus- cular assessment in the Elderly Worker (NEW) QLRT-2000-00 139. Asterisk indicates corresponding author. *D. Farina is with the Dipartimento di Elettronica, Politecnico di Torino, Corso Duca degli Abruzzi 24, Torino 10129, Italy (e-mail: dario.fa- rina@polito.it). R. Merletti is with the Centro di Bioingegneria, Dip. di Elettronica, Politec- nico di Torino, Torino 10129, Italy. Digital Object Identifier 10.1109/TBME.2003.819847 dicative of muscle fiber type constituency and particular muscle training [30]. As alternative to intramuscular recordings [33], [34], surface electromyography (EMG) signals constitute a useful means for estimating muscle fiber CV in a noninvasive way. CV, as well as characteristic spectral frequency, rate of change have been used in a number of studies as indicative of muscle fatigue (e.g., [2] and [26]). Although characteristic spectral frequencies and CV show similar trends during fatiguing isometric constant force contractions, they may provide different information in less standardized conditions [15]. In these cases, CV has a direct physiological meaning, as opposed to spectral frequencies which are indirect variables. Many methods for CV estimation from surface EMG have been proposed in the past. Both global and MU CV distribution have been subject of study [5], [6], [9], [10], [12], [18]. Many of the approaches proposed in the past imply detection of sig- nals in different locations in space with the same shape and a temporal delay [3]. Along this line, peak methods [17], [31], ap- proaches based on spectral dips [20], [21], maximum-likelihood two-channel and multichannel techniques [12], [23], phase [18], or distribution function [29] approaches have been proposed and applied. Many of these techniques can be seen as the application of one or more spatial and temporal filters to the surface EMG signal, to be discussed later. Although the techniques for CV estimation have recently been considerably improved, especially with respect to the standard deviation of estimation [12], [14], some important issues about CV estimate still remain. They are particularly related to the bias of the estimation in the case of nonideal conditions. Surface EMG potentials are not waves which propagate without shape change along an infinite line. From the mathe- matical point of view, the delay between signals with different shapes is not trivially defined and different definitions of delay lead to different results, even in the case of noise free signals. Thus, a delay estimator becomes implicitly a definition of delay. This estimator should be based on some reasonable assumptions. Factors introducing shape changes between signals detected along the fiber direction are, for example, the end-plate and end-of-fiber components which arise at the generation and ex- tinction of the intracellular action potentials. These nontraveling signals bias the CV estimates toward higher values. The relative weight of nontraveling potentials on surface EMG depends on anatomical, physical, and detection-system parameters [7], [13], 0018-9294/03$17.00 © 2003 IEEE