Introduction to this Special Issue

Medical Engineering & Physics 21 (1999) 379–388 www.elsevier.com/locate/medengphy

Introduction to this Special Issue Intelligent data analysis in electromyography and electroneurography Constantinos S. Pattichis b

a,*

, Ian Schofield b, Roberto Merletti c, Philip A. Parker d, Lefkos T. Middleton e

a Department of Computer Science, University of Cyprus, P.O. Box 20537, Nicosia 1678, Cyprus Department of Clinical Neurophysiology, Regional Neurosciences Centre, Newcastle-upon-Tyne, Tyne & Wear NE4 6BE, UK c Centro di Bioingegneria, Dipartimento di Elettronica, Politechnico di Torino, Turin, Italy d Department of Electrical Engineering, University of New Brunswick, New Brunswick, Canada e The Cyprus Institute of Neurology and Genetics, 1683 Nicosia, Cyprus

1. Introduction Computer-aided electromyography (EMG) and electroneurography (ENG) have become indispensable tools in the daily activities of neurophysiology laboratories in facilitating quantitative analysis and decision making in clinical neurophysiology, rehabilitation, sports medicine, and studies of human physiology. These tools form the basis of a new era in the practice of neurophysiology facilitating the: (i) Standardization. Diagnoses obtained with similar criteria in different laboratories can be verified. (ii) Sensitivity. Neurophysiological findings in a particular subject under investigation may be compared with a database of normal values to determine whether abnormality exists or not. (iii) Specificity. Findings may be compared with databases derived from patients with known diseases, to evaluate whether they fit a specific diagnosis. (iv) Equivalence. Results from serial examinations on the same patient may be compared to decide whether there is evidence of disease progression or of response to treatment. Also, findings obtained from different quantitative methods may be contrasted to determine which are most sensitive and specific. Different methodologies have been developed in computer-aided EMG and ENG analysis ranging from simple quantitative measures of the recorded potentials, to more complex knowledge-based and neural network systems that enable the automated assessment of neuromuscular disorders. However, the need still exists for the further

* Corresponding author. Tel.: +357-2-892230; fax: +357-2-339062. E-mail address: [email protected] (C.S. Pattichis)

advancement and standardization of these methodologies, especially nowadays with the emerging health telematics technologies which will enable their wider application in the neurophysiological laboratory. The main objective of this Special Issue of Medical Engineering & Physics is to provide a snapshot of current activities and methodologies in intelligent data analysis in peripheral neurophysiology. A total of 12 papers are published in this Special Issue under the following topics: Motor Unit Action Potential (MUAP) Analysis, Surface EMG (SEMG) Analysis, Electroneurography, and Decision Systems. In this introduction, the papers are briefly introduced, following a brief review of the major achievements in quantitative electromyography and electroneuropathy.

2. Major milestones in electromyography and electroneurography 2.1. Early developments The first extensive study on electromyography was carried out by Piper [1], in Germany in 1912. Piper was able to record the human voluntary muscle activity utilizing the Einthoven’s string galvanometer. For the following years physiologists were trying to correlate laboratory findings with ‘normal’ human muscle potentials. Proebster [2] in 1928, described spontaneous irregular action potentials in the denervated muscle of a boy with a traumatic plexus birth lesion and in another patient with long-standing poliomyelitis. Proebster was given the credit for opening the field of clinical electromyogra-

1350-4533/99/$ - see front matter  1999 IPEM. Published by Elsevier Science Ltd. All rights reserved. PII: S 1 3 5 0 - 4 5 3 3 ( 9 9 ) 0 0 0 8 4 - 3

380

C.S. Pattichis et al. / Medical Engineering & Physics 21 (1999) 379–388

phy. In his measurements, he used a recording galvanometer, and tested muscles in voluntary as well as in electrically induced contractions. In 1929, Adrian and Bronk [3] devised the concentric needle electrode and applied it to the study of the motor unit action potential (MUAP). They were able to amplify the muscle action potential activity, and utilize the amplified signal to drive a loudspeaker. The information carried by sound proved to be an important aid in interpreting the EMG signal. Adrian and Bronk found that in completely relaxed normal muscle, there was no spontaneous electrical activity. Also they noted the tiring rates and values of action potentials in voluntary contraction. The use of differential amplification [4] in electromyographic equipment allowed neurophysiologists to record motor unit potentials of small amplitude. Although Gasser and Erlanger [5] first used the cathode ray oscilloscope in place of the galvanometer in 1924, it was not but until the 1940s that the cathode ray oscilloscope was more widely used by electrophysiologists. The cathode ray oscilloscope helped neurophysiologists to study the undistorted shape of muscle activity, not limited by the string galvanometer. In the years that followed, the development of sensitive electronic equipment and the invention of new techniques in studying the muscle, contributed significantly to the analysis of EMG and ENG. 2.2. Electromyography 2.2.1. Needle EMG Traditionally, in clinical electromyography, neurophysiologists assess MUAP shape by watching the oscilloscope and by their audio characteristics. Using these simple means, an experienced electrophysiologist can detect abnormalities with reasonable accuracy. These techniques were mastered in the laboratories of Kugelberg [6,7] and Buchthal [8–10] in the 1940s and 1950s who defined the characteristics of EMG signals derived from normal and diseased muscle. Subjective MUAP assessment, although satisfactory for the detection of unequivocal abnormalities, may not be sufficient to delineate less obvious deviations or mixed patterns of abnormalities. These ambiguous cases call for quantitative MUAP analysis [11]. Buchthal and coworkers introduced manual methods for quantitative MUAP analysis in the 1950s [8–10]. MUAPs recorded photographically were selected for analysis. Manual measurement of the parameters duration, amplitude and number of phases was carried out. Twenty different potentials were collected, and their data were then compared to the values obtained from the corresponding muscle in age-matched normal subjects. As expected, although this method provides useful information, it is very time consuming, and limits itself to subjective measurement of the MUAP parameters of interest. The wider availability of

microcomputers in the 1970s and 1980s greatly facilitated the development of computer aided quantitative EMG methods. (For a brief introduction of MUAP computer-aided methods see Section 2.4). In order to investigate motor unit activity recruited at higher effort it is necessary to examine the EMG when it has reached the level of interference. The individual motor unit potentials can no longer be identified, and new quantitative methods for analysis have to be investigated. Willison [12] introduced turns and amplitude analysis in 1964. This method was further developed to be of more clinical usefulness by Fulgsang-Frederiksen and Mansson [13] in 1975. EMG decomposition is another way of analyzing the EMG interference pattern. Decomposition is the term commonly used to describe the process whereby individual MUAPs are identified and uniquely classified from a set of concurrently active MUAP trains. Pioneering work on EMG signal decomposition has been carried out by LeFever and DeLuca [14,15], Guiheneue et al. [16], and Dorfman and McGill [17]. New information about the organization and function of the lower motor neuron in health and disease have been revealed by new recording techniques like singlefiber EMG, scanning EMG and macro EMG. Singlefiber EMG is the most selective EMG method and its recordings include one or two fibers in the normal motor unit. This method was introduced by Ekstedt [18] at the beginning of 1960, and further developed by Stalberg [19], and Stalberg and Trontelj [20]. Stalberg [21] has also devised a new recording technique, called macro EMG, which is able to pick up activity from most of the muscle fibers within the motor unit. The scanning EMG technique introduced by Stalberg and Antoni [22] in 1980 investigates the motor unit organization by moving an EMG electrode using a stepper motor at 10 mm intervals through the motor unit. 2.2.2. Surface EMG Surface EMG is a technique that has emerged in the last 10–15 yr. In contrast to the EMG signal derived with needle electrodes where the activity of a very small number of motor units is detected, surface EMG is composed of the summation of hundreds of contributions from the active motor units. Each contribution is heavily filtered by the tissue interposed between the sources and the electrodes and most of the morphological information is lost. Because of this limitation the technique is applicable exclusively to superficial muscles and, at the present time, does not allow the investigation of a muscle when another above it is active. Crosstalk between adjacent muscles may be an additional problem. However the global nature of the signal is useful in providing information about muscle fatigue as well as activation time and level. If electrode arrays are used, anatomical and physiological information about individual

C.S. Pattichis et al. / Medical Engineering & Physics 21 (1999) 379–388

motor units may be recovered concerning the location of innervation zones, fiber length and conduction velocity. Applications of surface EMG are different or complementary with respect to those that require needles and consist of: 1. Analysis of muscle fatigue. The changes that take place in the EMG signal during sustained voluntary or electrically elicited contractions provide information on the physiological and histological properties of the muscle [23–25]. 2. Analysis of muscle activation patterns. The timing and intensity of muscle activation provide information about motor events as well as the strategies adopted by the central nervous system during dynamic performance. Particular care must be applied in these evaluations because of the artifacts introduced by the relative movement between muscle and electrodes [26]. 3. Analysis of individual motor units using electrode arrays. This approach is promising but still in its infancy and might partially replace the use of needles in a few cases when only superficial muscles are of interest [27–30]. Applications concern the diagnosis and follow up of slowly changing neuromuscular disorders [31], assessment of muscle conditions in low back pain [32] and sport medicine [33], evaluation of rehabilitation treatments and, ergonomic studies in occupational health [34]. An important activity in the field is provided by the European Concerted Action on Surface Electromyography for non invasive assessment of muscles (SENIAM) which will be concluded at the end of 1999. This action coordinates 16 European partners with the goal of defining methodological approaches and standards in the research and clinical applications of surface EMG. 2.3. Electroneurography Electroneurography is now a well-established technique for the diagnosis of peripheral nerve disorders. The basis of the procedure is to initiate an action potential in a peripheral nerve by means of an electrical stimulus and to record a response from either the nerve itself or from a muscle innervated by the nerve under study. The first report of the contraction of a muscle initiated by an electrical stimulus was presented by Kratzenstein in 1745. Subsequently, in 1794, Galvani demonstrated that the stimulation of a nerve resulted in contraction of the innervated muscle. The conduction velocity of a nerve was first measured by Von Helmholz in 1850. Since this time there have been major contributions from many workers in the field. However, it was not until 1948 that Hodes et al. [35] described the procedure that is in use

381

today. The technique was then applied in the clinical context by a number of workers notably, Buchtal, Gilliatt and Lambert who demonstrated that conduction in the peripheral nervous system is abnormal in a number of neurological disorders. This procedure now forms the core of the neurophysiological examination and contributes significantly to the key requirements which are, the identification of the underlying pathophysiology and it’s distribution. The two major parameters that may be derived from a study are the Nerve Conduction Velocity (NCV) and the amplitude of the response. The characteristics of the recorded responses vary greatly according to the type of recording electrodes. Two basic types are in regular use. Surface electrodes are the easiest to use but there are a number of different electrode designs available, which require separate sets of normal ranges. Needle electrodes may also be used to record from both muscle and nerve. These electrodes provide a much better signal to noise ratio particularly when recording sensory potentials due to the fact that the recording may be made from a site very close to or, with very fine electrodes, from within the nerve. Recordings of this type allow the conduction velocity of individual axons to be calculated in contrast to surface electrodes where the calculated velocity reflects the response from either the fastest conducting axons or a dominant sub-population. The recent development of techniques to assess the distribution of conduction velocities in peripheral nerve promises to considerably improve the sensitivity of the technique to minor abnormalities of myelin and disturbances of the function of ion channels. Data of this kind can be obtained from collision studies, first described by Hopf [36] and a number of modeling procedures have been investigated beginning with the work of Buchtal and Rosenfalck [37]. The clinical application of these techniques has been limited to short studies intended to validate the methods. The most important development in the area of Electroneurography has been the identification of a requirement for the standardization of techniques with the intention of creating databases of normative values and pathological case studies. A number of initiatives are in progress in this area. 2.4. Motor unit action potential (MUAP) analysis Quantitative analysis of EMG findings is desirable because it minimizes observer bias, allows for a more precise interpretation of the results, and facilitates comparison of results across individuals and different methodologies. Quantitative analysis is mainly applied in patients with known or suspected neuromuscular disorders. These include the muscular dystrophies, peripheral polyneuropathies, and anterior horn cell disorders. In such disorders, quantitative analysis may prove to

382

C.S. Pattichis et al. / Medical Engineering & Physics 21 (1999) 379–388

provide useful information to help the physician in reaching a more accurate diagnosis. It should be emphasized that if the disease process is advanced, and the electrophysiologic abnormalities are many and obvious, although automated EMG analyses might be of less importance diagnostically, it may be useful for detecting and characterizing changes on serial examinations [38]. However, the application of computer-aided EMG analysis is very important in detecting pathology in those early or mild cases of a disease, where the electrophysiological abnormalities are relatively slight and may escape accurate diagnosis. Although a number of computer-aided quantitative MUAP analysis algorithms have been developed, some of them commercially available, practically, none of them have gained wide acceptance for extensive routine clinical use. Most importantly, there are no uniform international criteria neither for pattern recognition of similar MUAPs nor for MUAP feature extractions [38,39]. A brief survey of quantitative EMG studies carried out during the last three decades in chronological order is now presented. Kopec and Hausmanowa-Petrusewicz in 1969 [40] developed the first dedicated microcomputer named ANOPS to facilitate the automatic MUAP analysis. The main criticism against their method was that similar MUAPs were considered repeatedly for analysis, as well as superimposed potentials were accepted. The template matching method for the recognition of similar MUAPs was introduced in the 1970s by Bergmans [41], Tanzi et al. [42], and Hynninen et al. [43]. LeFever and DeLuca [14,15] used a special threechannel recording electrode and a hybrid visual-computer decomposition scheme based on template matching and tiring statistics identification. Stalberg et al., in their original system, used waveform template matching [44], whereas more recently in their system called Multi-MUP (Multi-Motor Unit Potentials), they used different shape parameters as input to a template matching technique [45]. Guiheneuc et al. [16] classified MUAPs at low levels of contraction through comparison of shape parameters, Coatrieux et al. [46] used both hierarchical and non-hierarchical clustering techniques for classification, McGill et al. [17] developed the ADEMG (Automatic Decomposition Electromyography) system that used template matching and a specific alignment for classification. Andreassen [47] followed as closely as possible the manual method developed by Buchthal [10] also using template matching with four templates for the recognition of MUAPs recorded at threshold contraction. Stashuk and De Bruin [48] used a single-fiber EMG needle electrode for signal acquisition and a template matching technique similar to that of LeFever and DeLuca [14,15], based on power spectrum features and firing statistics. Their system required some operator interventions. Haas and Meyer [49], in their system

called ARTMUT (Automatic Recognition and Tracking of Motor Unit Potentials) used a hierarchical clustering method, followed by a two-stage decomposition. Pattichis et al. [50] used MUAP parameters input to a sequential parametric pattern recognition classifier. Loudon et al. [51] used eight MUAP features as input to a statistical pattern recognition technique for classification. Hassoun et al. [52] in their system called NNERVE (Neural Network Extraction of Repetitive Vectors for Electromyography) used the time domain waveform as an input to a three-layer artificial neural network (ANN) with a ‘pseudo-unsupervised’ learning algorithm for classification. More recently, an unsupervised neural network pattern recognition system based on the selforganizing feature maps algorithm for the classification of MUAP waveforms has been developed [53]. This system performed slightly better than template matching techniques, especially in the recognition of MUAPs recorded from subjects with neuromuscular disorders. 2.5. Surface EMG analysis The complex nature of the signal and the large amount of information contained in it makes the interpretation of the surface EMG very challenging. Sophisticated approaches for extraction of the relevant information are being developed, tested and published continuously. Classical approaches are based on the analysis of EMG amplitude and spectral variables and ‘average’ muscle fiber conduction velocity [54]. They have been applied for monitoring myoelectric manifestations of muscle fatigue during sustained contractions. It should be underlined that amplitude and spectral variables do not reflect exclusively muscle features but are affected by a variety of anatomical parameters such as the thickness of the subcutaneous layers, the size and location of the electrodes and the statistical distribution of muscle fiber conduction velocity. During sustained medium to high level isometric contractions the signal is not stationary and during dynamic contractions artifacts are introduced. The role of models as tools for understanding the weight of different factors is of paramount importance in surface EMG. The estimate of ‘average’ muscle fiber conduction velocity is actually a weighted average of the conduction velocity values where the weights are unknown and related to the size and depth of the motor units. Electrode arrays provide some kind of ‘lens’ that may be ‘focused’ on specific regions of space to provide the spatial selectivity required to reduce the complexity of the signal and to identify and study individual surface MUAPs. The analysis and processing of surface EMG are central to the success of controllers for powered artificial limbs in rehabilitation. Prostheses with one to three functions have conventionally been controlled with EMG amplitude and require fairly simple processing and

C.S. Pattichis et al. / Medical Engineering & Physics 21 (1999) 379–388

decision algorithms. However the development of multifunction prostheses with six or more functions is expected. In this case sophisticated EMG processors which more efficiently utilize the information present in the EMG are required. Pattern classification algorithms and decision strategies using feature vectors derived from the transient EMG, electrode arrays and time–frequency transformations are of considerable interest in this regard [55,56]. 2.6. Decision systems To further the development of quantitative EMG techniques, the need has emerged for adding automated decision making support to these techniques so that all data is processed in an integrated environment. Towards this goal, Blinowska [58] proposed the use of discriminate analysis for the evaluation of MUAP findings, Coatrieux and associates [46] applied cluster analysis techniques, whereas Pattichis and Schizas [59] applied genetics-based machine learning for the automatic diagnosis of myopathology based on MUAP records. Christodoulou et al. [60] also developed a MUAP multi feature–multi classifier decision making system for the assessment of myopathology. Andreassen and co-workers [61] developed the MUNIN (Muscle and Nerve Interface Network) which employs a causal probabilistic network for the interpretation of EMG findings, FuglsangFrederiksen and his group [62] developed a rule-based EMG expert system named KANDID, and Jamieson [63] developed an EMG processing system based on augmented transition networks. NEUROMYOSIS [64] is a more complex system using a variety of knowledge representation schemes and information processing procedures, HINT (High Level Inferencing Tool) [65] was developed by Schofield based on an anatomical network for the interpretation of neurophysiological studies of the peripheral nervous system. In most of these systems, the generation of the input pattern assumes a probabilistic model, with the matching score representing the likelihood that the input pattern was generated from the underlying class. In addition, assumptions are typically made concerning the probability density function of the input data. HINT however, does not contain a set of potential diagnostic classes. Rather, the study is described in terms of a minimal set of lesions that would account for the findings. This method considerably simplifies the problem of multiple diagnoses. Furthermore, Schizas et al. [66] introduced a hybrid neural network system for the assessment of neuromuscular disorders where clinical data in the form of signs and symptoms, neurophysiological findings, muscle imaging data, and biochemical genetic findings are used. The motivation behind the use of neural networks is that no assumptions about the underlying probability density function are made.

383

3. Overview of this Special Issue 3.1. MUAP analysis Three papers in this Special Issue relate to motor unit analysis in needle EMG. The first paper introduces a new technique for EMG decomposition, whereas the other two papers focus on quantitative MUAP analysis in neuromuscular disorders. The paper by Stashuk presents a new decomposition technique recording simultaneously selective or micro, and non-selective or macro EMG signals. The decomposition algorithm consists of MUAP detection, clustering, and supervised classification based on shape and tiring pattern information. This procedure is applied interactively to resolve a composite micro EMG signal into its constituent motor unit action potential trains. The system was evaluated on a set of 10 representative 20–30 s, concentric needle and surface EMG signals. The system’s maximum and average error rate were 2.5 and 0.7%, respectively, and on average assigned 88.7% of the detected MUAPs in 4–8 s. The results displayed to the user include the average of the micro and macro MUAP waveshapes, the variability of micro MUAP shapes, and their motor unit tiring patterns, and their parameters. The paper by Pattichi and Elia investigates the usefulness of autoregressive modeling and cepstral analysis for the diagnostic assessment of MUAPs recorded from normal (NOR) subjects and subjects suffering with motor neuron disease (MND) and myopathy (MYO). The following feature sets were extracted from the MUAP signal: (i) time domain measures, (ii) AR spectral measures, (iii) AR coefficients, and (iv) cepstral coefficients. Discriminative analysis of the individual features was carried out using the univariate and multiple covariance analysis methods. Both methods showed that: (i) the duration measure is the best discriminative feature among the time domain parameters, and (ii) the median frequency is the best discriminative feature among the AR spectral measures. Furthermore, the classification performance of the above four feature sets was investigated for three classes (NOR, MND and MYO) using artificial neural networks. Results showed that the highest diagnostic yield was obtained with the time domain measures followed by the cepstral coefficients, the AR spectral parameters, and the AR coefficients. The paper by Zalewska and Hausmanowa-Petrusewicz investigates the relationship between the global and detailed features of MUAPs in various neuromuscular disorders. The MUAP global feature is characterized by amplitude and duration and may be quantified by the following indexes: area/amplitude or size index, which are interpreted as the potential ‘thickness’. The detailed MUAP feature is characterized by the number of phases and turns. A new coefficient is introduced named irregu-

384

C.S. Pattichis et al. / Medical Engineering & Physics 21 (1999) 379–388

larity coefficient that combines the number of phases and turns with their amplitudes. This coefficient is related to the ‘length’ of the MUAP waveshape. It was found that in neurogenic cases, with the exception of the correlation between the duration and number of turns (r=0.58), no correlation between any other parameters related to global and detailed features were significant (rⱕ0.4). In myogenic cases, a significant correlation between the irregularity coefficient and duration (r=0.71), irregularity coefficient and area/amplitude (r=0.54) rather than that between the irregularity coefficient and area (r=0.32) was found. The correlation coefficient between parameters were more significant in more chronic processes. 3.2. Surface EMG analysis Some of the topics mentioned above in Section 2.5 are addressed by the papers on surface EMG included in this Special Issue. The paper by Englehati et al. addresses the issues of feature set choice and dimension in the classification of transient myoelectric patterns. Transient myoelectric signal patterns show considerable promise in myoelectric control systems based on pattern classification strategies. The performance of the classification algorithm depends critically on the feature set chosen to represent the patterns, both in the type of features and the set dimension. The paper considers time domain and time–frequency domain features which include those obtained from the short-term Fourier transform, wavelet transform, and wavelet packet transform. The feature set is subjected to two different dimension reduction techniques—feature selection and feature projection. Feature selection is performed using an Euclidian distance separability criterion, and feature projection by principal component analysis. The results demonstrate that time–frequency features are superior to time–domain features. It is also shown that wavelet-packet features, reduced in dimension by principal component analysis, provide the best classification performance of the sets studied. An alternative interesting approach to the analysis of surface EMG is provided by Filligoi and Felici who present the application of the method of Recurrence Quantification Analysis, originally proposed by Webber, in the study of isometric variable force contractions. The method seems to have some interesting advantages with respect to the traditional spectral approach and it is worthy of further investigation and preliminary clinical application. The work of Staude and Wolf addresses the issue of detecting onset of voluntary muscle contractions. An analysis of the existing methods is presented and improvements obtained by using a model-based approach are discussed. The paper by Cesarelli et al. studies the activation pattern of the femoral quadriceps muscles of individuals

with anterior knee pain. A quantitative analysis of the activation pattern is accomplished through the use of EMG signals from the vastus lateralis, rectus femoris, and vastus medialis. The EMG signals were acquired by bipolar electrode configurations during extension of the knee and subjected to mean absolute value linear envelope analysis. The envelope amplitudes were normalized to the maximum amplitude over the extension. A group of patients with anterior knee pain were compared with a control group of healthy individuals on the basis of the normalized linear envelopes, and features extracted from the envelopes. The features include a Gaussian pulse decomposition, and an asymmetry coefficient. The results demonstrate a delay in muscle activation compared to healthy subjects, and a asymmetry coefficient distribution which shows behaviour in the vastus medialis which differs from healthy subjects. The paper by Dimitrova et al. investigates MUAPs detected at a large distance (typical of surface recording) on the basis of mathematical modeling. MUAPs were calculated as a single convolution of the first temporal derivative of the intracellular action potential and the motor unit impulse response. MUAP spatial averaging over the rectangular plate electrodes was performed through analytical integration of the motor unit impulse response over the electrode area. It was found that the effects of longitudinal dimension of the electrode were stronger than those of the transversal one. The longitudinal dimension of the electrode influenced the MUAP main phases (that reflected the excitation origin and propagation) more than the MUAP terminal phases (that reflected the excitation extinction at the muscle fibers’ ends). It was shown that the relative weight of the individual MUP phases could be stressed or suppressed by a proper choice of the electrode dimensions, position and orientation. One of the main difficulties in surface EMG is the extraction of information about single motor units. The problem of ‘focusing’ on single MUAPs has been addressed by G. Rau and C. Disselhorst-Klug in a number of papers [29] and is discussed in this issue by Farina and Rainoldi who designed a bidimensional array to implement a spatial filter that compensates for the blurring introduced by the subcutaneous and coetaneous layers of conductive tissue. The simulations presented show that ‘deblurring’ could be implemented if the thickness and conductivities of the layers are known. This finding shows the need for methods suitable to estimate the properties of the skin and fat layers. 3.3. Electroneurography Dorfmann et al. reviewed the development of techniques for the analysis of conduction velocity distribution [57]. These computationally intensive techniques are becoming more applicable now that significant pro-

C.S. Pattichis et al. / Medical Engineering & Physics 21 (1999) 379–388

cessing power is relatively cheap. These techniques considerably extend the basic nerve conduction study procedures. Two papers in this Special Issue are presented in the area of electroneurography. The first paper introduces a new technique for the estimation of the distribution of conduction velocities, and the second paper investigates the effect of stimulus current pulse with on nerve fiber size recruitment patterns. Papdopoulou and Panas introduce a novel technique based on the bispectrum for the estimation of the distribution of conduction velocities (DCV) in peripheral nerve. The compound action potential (CAP) is usually expressed in the discrete domain as the circular convolution of a delay sequence and the sampled single-fiber action potentials (SFAPs). In the proposed method, averaging of low signal to noise ratio of CAPs is carried out in the third order spectrum domain, so that time alignment is not required. The averaged bispectra are computed based on a modified Hirose’s method, that allows to estimate the delay sequence for conduction distance l1. The last linear phase is recovered by using the delay phase cepstrum. Finally, their DCV can be calculated from the estimated delay sequence, according to the formulation of the forward problem. The proposed algorithm was tested on simulated data, where it was found that better performance was obtained with bispectrum averaging compared to time averaging, especially in the case of noisy electroneurograms. The paper by Szlavik and de Bruin investigates the selective recruitment of nerve fibers based on stimulus pulse duration. The paper describes both a simulation study and experimented study of the effect of pulse width on fiber recruitment. In the simulation study a resistive tissue model is used incorporating tissue anisotropy. For a given stimulus current the potential distribution in the tissue is calculated, and compared to the fiber threshold potential of 25 mV. A population of 60 fibers with diameters ranging from 2 to 20 µm and depths ranging from 1 to 3 mm and current pulse widths of 10–1000 µs are used. An experimental study is also carried out on the motor nerve of the thenar muscle for five subjects. Motor unit potentials and their onset delay times were estimated from the -wave, and their fiber diameters estimated from the delay time and distance traveled. The simulation study results show a 20% decrease in fiber diameter when stimulus duration is increased from 10 to 1000 µs. The experimental study results were consistent with those of the simulation study.

385

prevalence of a given disease in general population and in the population seen in the EMG lab. It can be argued that both prevalences are the correct choice as prior probabilities for the diseases. However, this issue can be resolved by recognizing that the EMG-diagnosis is only based on the information provided by the EMG-examination and it thus represents a partial view of the patient. The paper proposes a solution summarizing the set of findings, signs, symptoms, and lab results, that let to the referral of the patient for an EMG examination. This information can be described by stochastic variables called FIDL factors (Found In Doctor’s Lab). The proposed methodology was tested on the EMG expert system MUNIN with 30 cases. It was shown that this method improved the specificity of the diagnosis, without affecting the sensitivity.

4. Concluding remarks

3.4. Decision systems

The 12 papers presented in this Special Issue represent a snapshot of current activities in the area of MUAP, surface EMG, electroneurography, and decision systems in the clinical neurophysiology research laboratory. These systems contribute to the opening of a new era in the practice of neurophysiology targeting in the better delivery of healthcare. However, more work and effort are needed to enable further the advancement and standardization of these methodologies in peripheral neurophysiology. The EU funded project ESTEEM (European Standardized Telematic tool to Evaluate EMG knowledgebased systems and Methods) provides a very promising framework that facilitates the quality development in electromyography among a large number of neurophysiological laboratories in Europe. Furthermore, the EU Concerted Action SENIAM is providing an important step forward in reaching a general consensus and in defining a set of standards in surface EMG detection, processing, modeling and applications. This European project provides a common background and a starting point for a number of more specific projects which may now be pursued by different labs with the purpose of applying surface EMG in the clinical environment using a common knowledge base and methodology so that results may be compared and clinical standard protocols may be defined. The field is now sufficiently consolidated for a widespread clinical research application and for teaching. It is proposed that further work is needed in the following areas:

In this Special Issue, the paper by Suojanen et al. presents a system for the interpretation of EMG findings based on partial information. The system is based on the argument that there is a large difference between the

1. Quantitative analysis. There is a need for the documentation of standardized algorithms and the formation of biosignal digital libraries for the analysis, interpretation, and further decision making of single-

386

C.S. Pattichis et al. / Medical Engineering & Physics 21 (1999) 379–388

fiber action potentials, MUAPs, interference pattern, decomposition and surface EMG. A first set of computer generated and experimental signals is available from SENIAM. (SENIAM material is available from H. Hermens, Roessingh Research and Development, Enschede, The Netherlands, e-mail: [email protected]). 2. Decision systems. There is a need for the creation of a library of cases to facilitate the assessment and monitoring of neuromuscular disorders and function through a decision support system. In addition to neurophysiological findings, the data to be collected should include clinical assessment, muscle and, or nerve biopsy data, biochemical findings, and genetic and molecular genetic findings. These data should be based on procedures applied in the early and accurate diagnosis of the clinical types and stages of neuromuscular disorders, their clinical subtypes and variants. The library will facilitate the evaluation, as well as the development of decision making systems. 3. Telematics applications. The development of the above technologies will enable their integration to therapy planning systems, department and hospital information systems, archiving and communication systems, and internet-based cooperative working environments for teleconsultation, telediagnosis, and telemonitoring (for example patient home monitoring). 4. Teaching, training and continuing education. The field of EMG, in particular surface EMG, is part of the education and training of very few physicians, most of whom are neurologists or neurophysiologists. Many other clinicians and health professionals may benefit from the use of surface EMG but do not receive a proper education and training in the field. Among these are rehabilitation doctors and therapists, experts in sport and occupational medicine, ergonomists and, to some degree, psychologists working in the field of occupational health and stress. There is a strongly felt need to transfer the consolidated knowledge concerning EMG to the university and post university training and continuing education programs that the Health Delivery Systems of some countries provide for their physicians and health personnel. This education may be provided through traditional channels (medical schools), summer schools and courses, multimedia teaching tools, and internet based training. Although the purpose of this Special Issue is to present the state of the art and the most recent advances in the fields of electromyography and electroneurography we feel that is also our responsibility and duty to insure that our research does not only produce scientific papers but is transferred to the clinical community through the most appropriate channels and results in a real improvement of the diagnostic, assessment and monitoring procedures

for the ultimate benefit of the patients and of the community in general. One of the main goals of the Fifth Framework Program of the European Union and the program Technologies for the 21st century by the United States is the improvement of the quality of life and health of the citizens. A contribution to the achievement of this goal is the clinical application of the available tools and methods that have been developed for quantitative and objective assessment of the neuromuscular system not only for the care of the pathological subject but for the elderly and the disabled as well as for the workers that may be subject to stress, cumulative trauma disorders and fatigue.

Acknowledgements The guest editors would like to express their sincere thanks to the numerous reviewers for their valuable help, constructive criticism, and suggestions, during the review process of this Special Issue. Many thanks are extended to the authors whose work and scientific initiative over the last years have made possible the work presented in this Special Issue. Furthermore, the guest editors would like to extend their gratitude to C. Roberts, Former Editor-in-Chief of MEP for his valuable support and guidance in materializing this Special Issue. Thanks also are extended to R. Allen, Editor-in-Chief of MEP for his support. Many thanks to E. Polycarpou for her efficient handling of the submitted manuscripts.

References [1] Piper H. Elektrophysiologie menschlicher Musleln. Berlin: Julius Springer, 1912. [2] Proebster R. Uber Muskelaktionsstrome am gesunder und Kranken Menschem. Zeitschr Orthop Chir 1928;50:1. [3] Andrian ED, Bronk DW. The discharge of impulses in motor nerve fibres II: the frequency of discharge in reflex and voluntary contractions. J Physiol 1929;67:119–51. [4] Matthews BHC. A special purpose amplifier. J Physiol (London) 1934;81:28–9. [5] Gasser HS, Erlanger J. The compound nature of the action current of nerve as disclosed by the cathode ray oscillograph. Am J Physiol 1924;70:624–66. [6] Kugelberg E. Electromyogram in muscular disorders. J Neurol Neurosur Psychiat 1947;60:140. [7] Kugelberg E, Petersen I. Insertion activity in electromyography. J Neurol Neurosur Psychiat 1949;12:268. [8] Buchthal F, Pinelli P, Rosefalck P. Action Potential parameters in normal human muscle and their physiological determinations. Acta Physiol Stand 1954;32:219–29. [9] Buchthal F, Guld C, Rosenfalck P. Action potential parameters in normal human muscle and their dependence on physical variables. Acta Physiol Scand 1954;32:200–15. [10] Buchthal F. An introduction to electromyography. Copenhagen: Gyldendal, 1957. [11] Buchthal F, Kamieniecka Z. The diagnostic yield of quantified

C.S. Pattichis et al. / Medical Engineering & Physics 21 (1999) 379–388

[12]

[13]

[14]

[15]

[16]

[17]

[18] [19] [20] [21] [22] [23]

[24]

[25] [26]

[27]

[28]

[29]

[30]

[31]

[32]

electromyography and quantified muscle biopsy in neuromuscular disorders. Musc Nerve 1982;5:265–80. Willison RG. Analysis of electrical activity in healthy and dystrophic muscle in man. J Neurol Neurosurg Psychiat 1964;27:386–94. Fuglsang-Frederiksen, Mansson A. A pattern of electrical activity during moderate effort and properties of motor unit potentials sampled during weak effort in progressive muscular dystrophy. Electroenceph Clin Neurophysiol 1974;36:576. LeFever RS, DeLuca CJ. A procedure for decomposing the myoelectric signal into its constituent action potentials: I. technique, theory and implementation. IEEE Trans Biom Eng 1982;29(3):149–57. LeFever RS, DeLuca CJ. A procedure for decomposing the myoelectric signal into its constituent action potentials: II. execution and test for accuracy. IEEE Trans Biom Eng 1982;29(3):158–64. Guiheneuc P, Calamel J, Doncarli C, Gitton D, Michel C. Automatic detection and pattern recognition of single motor unit potentials in needle EMG. In: Desmedt JE, editor. Computeraided electromyography. Karger: Basel Press, 1983;10:73–127. McGill KC, Cummins KL, Dorfman LJ. Automatic decomposition of the clinical electromyogram. IEEE Trans Biom Eng 1985;32(7):470–7. Ekstedt J. Human single muscle fibre action potentials. Acta Physiol Scand 1964;61 (Suppl)(226):1. Stalberg E. Propagation velocity in human muscle fiber in situ. Acta Physiol Scand 1966;70 (Suppl)(287):1–109. Stalberg E, Trontelj JV. Single fibre electromyography. Old Working, UK: Mirvalle Press, 1979. Stalberg E. Macro EMG, a new recording technique. J Neurol Neurosurg Psychiat 1980;43:475–83. Stalberg E, Antoni L. Electrophysiological cross section of motor unit. J Neurol Neurosurg Psychiat 1980;43:469–74. Duchene J, Goubel F. Surface electromyogram during voluntary contraction: processing tools and relation to physiological events. CRC Crit Rev Biom Eng 1993;21:313–97. Kupa E, Roy S, Kandarian S, De Luca C. Effect of muscle fiber type and size on EMG median frequency and conduction velocity. J Appl Physiol 1995;78:23–32. Merletti R. Surface electromyography: possibilities and limitations. J Rehab Sci 1994;7:24–34. Chen J, Shiavi R, Zhang Q. A quantitative and qualitative description of EMG linear envelopes for synergy analysis. IEEE Trans Biom Eng 1993;39:9–18. Masuda T, Miyano H, Sadoyama T. A surface electrode array for detecting action potential trains of single motor units. Electroenc Clin Neurophys 1985;60:435–43. Merletti R, Farina D, Granata A. Non-invasive assessment of motor unit properties with linear electrode arrays. In: Comi G, Locking C, Kimura J, Rossini PM, editors. Clinical neurophysiology: from receptors to perception. Amsterdam: Elsevier, 1999. Rau G, Disselhorst-Klug C. Principles of high spatial resolution surface EMG: single motor unit detection and application in the diagnosis of neuromuscular disorders. J Electrom Kinesiol 1997;7:233–9. Roeleveld K, Stegeman D, Vingerhoets H, van Oosterom HA. The motor unit potential distribution over the skin surface and its use in estimating the motor unit location. Acta Physiol Scand 1997;161:465–72. Meyer M, Hilfiker P, Gygi A. Surface EMG for diagnosis of neuromuscular diseases. In: Desmedt JE, editor. Computer-aided electromyography and expert systems. Amsterdam: Elsevier, 1989:181–8. Roy SH. The role of muscle fatigue in low back pain. In: D’Orazio P, editor. Back pain rehabilitation. Andever Medical Publisher, 1993.

387

[33] Clarys JP, Cabri J. Electromyography and the study of sport movements: a review. J Sports Sci 1993;11:379–448. [34] Veirsted KB, Westgaard R, Andersen P. Electromyographic evaluation of muscle work pattern as a predictor of trapezius myalgia. Scand J Work Env Health 1993;19:284–90. [35] Hodes R, Larrabee MF, German W. The human electromyogram in response to nerve stimulation and the conduction of motor axons. Arch Neurol Psychait 1948;60:340–52. [36] Hopf HC. Untersuchungen uber die unterschiede in der leitgeschwindigkeit motorische nervenfasern beim menschen. Deutsch Zeitschr Nervenheil 1962;183:579–88. [37] Buchtal F, Rosenfalck A. Evoked action potentials and conduction velocity in human sensory nerves. Brain Res 1966;3:1–122. [38] Dorfman LJ, McGill KC. Automatic quantitative EMG. Musc Nerve 1988;11:804–27. [39] Stalberg E, Andreassen S, Falck B, Lang H, Rosenfalck A, Trojaborg W. Quantitative analysis of individual motor unit potentials: a proposition for standardised terminology and criteria for measurement. J Clin Neurophysiol 1986;3(4):313–48. [40] Kopec J, Hausmanowa-Petrusewicz I. Histogram of muscle potentials recorded automatically with the aid of the averaging computer ‘ANOPS’. Electromyogr Clin Neurophysiol 1969;9:36–45. [41] Bergmans J. Computer-assisted measurement of the parameters of single motor unit potentials in human EMG. In: Desmedt JE, editor. New developments in electromyography and clinical neurophysiology. Karger: Basel Press, 1973;2:482–8. [42] Tanzi F, Taglietti V, Zucca G, Arrigo A, Moglia A, Sandrini G, Cinquini G. Computerized EMG analysis. Electromyogr Clin Neurophysiol 1979;19:495–503. [43] Hynninen P, Philipson L, Manderson B, Elmqvist D. A computer based method for automatic motor unit potential analysis. Acta Neural Scand 1979;60(73):300. [44] Stalberg E, Antoni L. Computer-aided EMG analysis. In: Desmedt JE, editor. Progress in clinical neurophysiology. Karger: Basel Press, 1983;10:186–234. [45] Stalberg E, Falck B, Sonoo M, Stalberg S, Astrom M. MultiMUP EMG analysis—a two year experience in daily clinical work. Electroenc Clin Neurophysiol 1995;145–54. [46] Coatrieux JL, Toulouse P, Rouvrais B, Le Bars R. Automatic classification of electromyographic signals. Electroenc Clin Neurophysiol 1983;55:333–41. [47] Andreassen S. Methods for computer-aided measurement of motor unit parameters. In: Ellington RJ et al, editors. The London Symposium (EEG suppl. 39). Amsterdam: Elsevier, 1987:13–20. [48] Stashuk D, de Bruin H. Automatic decomposition of selective needle-detected myoelectric signals. IEEE Trans Biom Eng 1988;35:1–10. [49] Hass W, Meyer M. An automatic EMG decomposition for routine clinical examinations and clinical research. In: Desmedt JE, editor. Computer-aided electromyography and expert systems. Amsterdam: Elsevier, 1989:67–81. [50] Pattichis CS, Schizas CN, Middleton L. Neural network models in classifying motor unit potential parameters. IEEE Trans Biom Eng 1995;42(5):486–96. [51] Loudon GH, Jones NB, Sehmi AS. New signal processing techniques for the decomposition of EMG signals. Med Biol Eng Comp 1992;591–9. [52] Hassoun MH, Wang C, Spitzer AR. NNERVE: neural network extraction of repetitive vectors for electromyography—part I: algorithm. IEEE Trans Biom Eng 1994;41(11):1039–52. [53] Christodoulou C, Pattichis CS. Unsupervised pattern recognition for the classification of EMG signals. IEEE Trans Biom Eng 1999;46:169–78. [54] Merletti R, Conte LL. Advances in processing of surface myoelectric signals, part 1 and 2. Med Biol Eng Comp 1995;33:362–84.

388

C.S. Pattichis et al. / Medical Engineering & Physics 21 (1999) 379–388

[55] Dickhaus H, Heinrich H. Classifying biosignals with wavelet networks. IEEE Trans Eng Med Biol 1996;103–13. [56] Hudgins B, Parker PA, Scott RN. A new strategy for multifunction myoelectric control. IEEE Trans Biom Eng 1993;40:82–94. [57] Dorfman LJ, Cummins KL, Leifer LJ, Alan R, editors. Conduction velocity distributions, “A population approach to electrophysiology”, Progress in clinical and biological research.: Liss Inc, 1980;52. [58] Blinowska KJ, Hausmanowa-Petrusewiczi, Miller-Larsson A, Zachra J. The analysis of single EMG potentials by mean of multivariate methods. Electrom Clin Neurophysiol 1980;20:105–23. [59] Pattichis CS, Schizas CN. Genetics-based machine learning models for the diagnosis of certain neuromuscular disorders. IEEE Trans Neur Net 1996;7(2):427–39. [60] Christodoulou CI, Pattichis CS, Fincham WF. A modular neural network decision support system in EMG diagnosis. J Intelligent Systems 1998;8(1–2):99–143.

[61] Andreassen S, Jensen F, Andersen SK, Falck B, Kjaerulff U, Woldbye M, Sorensen AR, Rosenfalck A, Jensen F. MUNIN— an expert EMG assistant. In: Desmedt JE, editor. Computer-aided electromyography and expert systems. Amsterdam: Elsevier, 1989:255–77. [62] Fuglsang-Frederiksen A, Jeppesen SM. A rule based EMG expert system for diagnosing neuromuscular disorders. In: Desmedt JE, editor. Computer-aided electromyography and expert systems. Amsterdam: Elsevier, 1989:289–96. [63] Jemieson PW. A model for diagnosing and explaining multiple disorders. Comput Biomed Res 1991;24:307–20. [64] Vila A, Ziebelin D, Reymond F. Experimental EMG expert systems as an aid to diagnosis. Electroen Clin Neuroph 1985;61:240. [65] Schofield IS. HINT — high level inference tool. J Intelligent Sys 1998;8(1–2):85–98. [66] Schizas CN, Pattichis CS, Bonsett CA. Medical diagnostic systems: a case for neural networks. Technol Health Care 1994;2:1–18.