Decomposition of surface EMG signals from cyclic dynamic contractions.

Over the past 3 decades, various algorithms used to decompose the electromyographic (EMG) signal into its constituent motor unit action potentials (MUAPs) have been reported. All are limited to decomposing EMG signals from isometric contraction. In this report, we describe a successful approach to decomposing the surface EMG (sEMG) signal collected from cyclic (repeated concentric and eccentric) dynamic contractions during flexion/extension of the elbow and during gait. The increased signal complexity introduced by the changing shapes of the MUAPs due to relative movement of the electrodes and the lengthening/shortening of muscle fibers was managed by an incremental approach to enhancing our established algorithm for decomposing sEMG signals obtained from isometric contractions. We used machine-learning algorithms and time-varying MUAP shape discrimination to decompose the sEMG signal from an increasingly challenging sequence of pseudostatic and dynamic contractions. The accuracy of the decomposition results was assessed by two verification methods that have been independently evaluated. The firing instances of the motor units had an accuracy of ∼90% with a MUAP train yield as high as 25. Preliminary observations from the performance of motor units during cyclic contractions indicate that during repetitive dynamic contractions, the control of motor units is governed by the same rules as those evidenced during isometric contractions. Modifications in the control properties of motoneuron firings reported by previous studies were not confirmed. Instead, our data demonstrate that the common drive and hierarchical recruitment of motor units are preserved during concentric and eccentric contractions.

Dynamical learning and tracking of tremor and dyskinesia from wearable sensors.

We have developed and evaluated several dynamical machine-learning algorithms that were designed to track the presence and severity of tremor and dyskinesia with 1-s resolution by analyzing signals collected from Parkinson's disease (PD) patients wearing small numbers of hybrid sensors with both 3-D accelerometeric and surface-electromyographic modalities. We tested the algorithms on a 44-h signal database built from hybrid sensors worn by eight PD patients and four healthy subjects who carried out unscripted and unconstrained activities of daily living in an apartment-like environment. Comparison of the performance of our machine-learning algorithms against independent clinical annotations of disorder presence and severity demonstrates that, despite their differing approaches to dynamic pattern classification, dynamic neural networks, dynamic support vector machines, and hidden Markov models were equally effective in keeping error rates of the dynamic tracking well below 10%. A common set of experimentally derived signal features were used to train the algorithm without the need for subject-specific learning. We also found that error rates below 10% are achievable even when our algorithms are tested on data from a sensor location that is different from those used in algorithm training.

High-resolution tracking of motor disorders in Parkinson's disease during unconstrained activity.

Parkinson's disease (PD) can present with a variety of motor disorders that fluctuate throughout the day, making assessment a challenging task. Paper-based measurement tools can be burdensome to the patient and clinician and lack the temporal resolution needed to accurately and objectively track changes in motor symptom severity throughout the day. Wearable sensor-based systems that continuously monitor PD motor disorders may help to solve this problem, although critical shortcomings persist in identifying multiple disorders at high temporal resolution during unconstrained activity. The purpose of this study was to advance the current state of the art by (1) introducing hybrid sensor technology to concurrently acquire surface electromyographic (sEMG) and accelerometer data during unconstrained activity and (2) analyzing the data using dynamic neural network algorithms to capture the evolving temporal characteristics of the sensor data and improve motor disorder recognition of tremor and dyskinesia. Algorithms were trained (n=11 patients) and tested (n=8 patients; n=4 controls) to recognize tremor and dyskinesia at 1-second resolution based on sensor data features and expert annotation of video recording during 4-hour monitoring periods of unconstrained daily activity. The algorithms were able to make accurate distinctions between tremor, dyskinesia, and normal movement despite the presence of diverse voluntary activity. Motor disorder severity classifications averaged 94.9% sensitivity and 97.1% specificity based on 1 sensor per symptomatic limb. These initial findings indicate that new sensor technology and software algorithms can be effective in enhancing wearable sensor-based system performance for monitoring PD motor disorders during unconstrained activities.

Dynamic SVM detection of tremor and dyskinesia during unscripted and unconstrained activities.

In this paper, we report an experimental comparison of dynamic support vector machines (SVMs) to dynamic neural networks (DNNs) in the context of a system for detecting dyskinesia and tremor in Parkinson's disease (PD) patients wearing accelerometer (ACC) and surface electromyographic (sEMG) sensors while performing unscripted and unconstrained activities of daily living. These results indicate that SVMs and DNNs of comparable computational complexities yield approximately identical performance levels when using an identical set of input features.

Resolving signal complexities for ambulatory monitoring of motor function in Parkinson's disease.

Automatic tracking of movement disorders in patients with Parkinson's disease (PD) is dependent on the ability of machine learning algorithms to resolve the complex and unpredictable characteristics of wearable sensor data. The challenge reflects the variety of movement disorders that fluctuate throughout the day which can be confounded by voluntary activities of daily life. Our approach is the development of multiple dynamic neural network (DNN) classifiers whose application are governed by a rule-based controller within the Integrated Processing and Understanding of Signals (IPUS) framework. Solutions are described for time-varying occurrences of tremor and dyskinesia, classified at 1 s resolution from surface electromyographic (sEMG) and tri-axial accelerometer (ACC) data acquired from patients with PD. The networks were trained and tested on separate datasets, respectively, while subjects performed unscripted and unconstrained activities in a home-like setting. Performance of the classifiers achieved an overall global error rate of less than 10%.

Resolving signal complexities for ambulatory monitoring of motor function in Parkinson's disease.

Automatic tracking of movement disorders in patients with Parkinson's disease (PD) is dependent on the ability of machine learning algorithms to resolve the complex and unpredictable characteristics of wearable sensor data. The challenge reflects the variety of movement disorders that fluctuate throughout the day which can be confounded by voluntary activities of daily life. Our approach is the development of multiple dynamic neural network (DNN) classifiers whose application are governed by a rule-based controller within the Integrated Processing and Understanding of Signals (IPUS) framework. Solutions are described for time-varying occurrences of tremor and dyskinesia, classified at 1 s resolution from surface electromyographic (sEMG) and tri-axial accelerometer (ACC) data acquired from patients with PD. The networks were trained and tested on separate datasets, respectively, while subjects performed unscripted and unconstrained activities in a home-like setting. Performance of the classifiers achieved an overall global error rate of less than 10%.

What is IPUS and how does it help resolve biosignal complexity?

Integrated Processing and Understanding of Signals (IPUS) combines signal processing and artificial intelligence approaches to develop algorithms for resolving signal complexity. It has also led to development over the last decade and a half of software tools for supporting the algorithm design process. The signals to be analyzed are the superposition of temporally localized and temporally overlapping signal components from broadly defined signal classes pertinent to the given application. Resolving a signal's complexity thus amounts to "decoding" it to reveal details of the specific signal components that are present at each point of a dense temporal grid defined on the signal. IPUS uses artificial intelligence techniques such as rule-based inference in conjunction with parameterized signal processing transformations to combat the combinatorial explosion encountered in any exhaustive search among the possible decoding answers for a given signal. Originally developed in the mid 1990's for auditory scene analysis, the IPUS approach has since been refined and extended in the context of various applications. In this paper, we present an overview of IPUS and discuss why its latest developments significantly impact biosignal analysis in diverse rehabilitation applications.

Detecting freezing-of-gait during unscripted and unconstrained activity.

We present a dynamic neural network (DNN) solution for detecting instances of freezing-of-gait (FoG) in Parkinson's disease (PD) patients while they perform unconstrained and unscripted activities. The input features to the DNN are derived from the outputs of three triaxial accelerometer (ACC) sensors and one surface electromyographic (EMG) sensor worn by the PD patient. The ACC sensors are placed on the shin and thigh of one leg and on one of the forearms while the EMG sensor is placed on the shin. Our FoG solution is architecturally distinct from the DNN solutions we have previously designed for detecting dyskinesia or tremor. However, all our DNN solutions utilize the same set of input features from each EMG or ACC sensor worn by the patient. When tested on experimental data from PD patients performing unconstrained and unscripted activities, our FoG detector exhibited 83% sensitivity and 97% specificity on a per-second basis.

High-yield decomposition of surface EMG signals.

OBJECTIVE: Automatic decomposition of surface electromyographic (sEMG) signals into their constituent motor unit action potential trains (MUAPTs). METHODS: A small five-pin sensor provides four channels of sEMG signals that are in turn processed by an enhanced artificial intelligence algorithm evolved from a previous proof-of-principle. We tested the technology on sEMG signals from five muscles contracting isometrically at force levels ranging up to 100% of their maximal level, including those that were covered with more than 1.5cm of adipose tissue. Decomposition accuracy was measured by a new method wherein a signal is first decomposed and then reconstructed and the accuracy is measured by comparison. Results were confirmed by the more established two-source method. RESULTS: The number of MUAPTs decomposed varied among muscles and force levels and mostly ranged from 20 to 30, and occasionally up to 40. The accuracy of all the firings of the MUAPTs was on average 92.5%, at times reaching 97%. CONCLUSIONS: Reported technology can reliably perform high-yield decomposition of sEMG signals for isometric contractions up to maximal force levels. SIGNIFICANCE: The small sensor size and the high yield and accuracy of the decomposition should render this technology useful for motor control studies and clinical investigations.

Surface EMG signal decomposition using empirically sustainable biosignal separation principles.

We introduce the concept of empirically sustainable principles for biosignal separation as a means of addressing the complexities that are practically encountered in the decomposition of surface electromyographic (sEMG) signals. Recently, we have identified two new principles of this type. The first principle places upper bounds on the inter-firing intervals and residual signal energies of the separated components. The second principle seeks a local minimum in the coefficient of variation of inter-firing intervals of each separated component. Upon incorporation of these principles into our latest Precision Decomposition system for sEMG signals, 20 to 30 motor unit action potential trains (MUAPTs) were decomposed per experimental sEMG signal from isometric contractions with trapezoidal force profiles. Our new system performs well even as the force generated by a muscle approaches maximum voluntary levels.

A combined sEMG and accelerometer system for monitoring functional activity in stroke.

Remote monitoring of physical activity using body-worn sensors provides an alternative to assessment of functional independence by subjective, paper-based questionnaires. This study investigated the classification accuracy of a combined surface electromyographic (sEMG) and accelerometer (ACC) sensor system for monitoring activities of daily living in patients with stroke. sEMG and ACC data (eight channels each) were recorded from 10 hemiparetic patients while they carried out a sequence of 11 activities of daily living (identification tasks), and 10 activities used to evaluate misclassification errors (nonidentification tasks). The sEMG and ACC sensor data were analyzed using a multilayered neural network and an adaptive neuro-fuzzy inference system to identify the minimal sensor configuration needed to accurately classify the identification tasks, with a minimal number of misclassifications from the nonidentification tasks. The results demonstrated that the highest sensitivity and specificity for the identification tasks was achieved using a subset of four ACC sensors and adjacent sEMG sensors located on both upper arms, one forearm, and one thigh, respectively. This configuration resulted in a mean sensitivity of 95.0%, and a mean specificity of 99.7% for the identification tasks, and a mean misclassification error of < 10% for the nonidentification tasks. The findings support the feasibility of a hybrid sEMG and ACC wearable sensor system for automatic recognition of motor tasks used to assess functional independence in patients with stroke.

Decomposition of indwelling EMG signals.

Decomposition of indwelling electromyographic (EMG) signals is challenging in view of the complex and often unpredictable behaviors and interactions of the action potential trains of different motor units that constitute the indwelling EMG signal. These phenomena create a myriad of problem situations that a decomposition technique needs to address to attain completeness and accuracy levels required for various scientific and clinical applications. Starting with the maximum a posteriori probability classifier adapted from the original precision decomposition system (PD I) of LeFever and De Luca (25, 26), an artificial intelligence approach has been used to develop a multiclassifier system (PD II) for addressing some of the experimentally identified problem situations. On a database of indwelling EMG signals reflecting such conditions, the fully automatic PD II system is found to achieve a decomposition accuracy of 86.0% despite the fact that its results include low-amplitude action potential trains that are not decomposable at all via systems such as PD I. Accuracy was established by comparing the decompositions of indwelling EMG signals obtained from two sensors. At the end of the automatic PD II decomposition procedure, the accuracy may be enhanced to nearly 100% via an interactive editor, a particularly significant fact for the previously indecomposable trains.

Decomposition of surface EMG signals.

This report describes an early version of a technique for decomposing surface electromyographic (sEMG) signals into the constituent motor unit (MU) action potential trains. A surface sensor array is used to collect four channels of differentially amplified EMG signals. The decomposition is achieved by a set of algorithms that uses a specially developed knowledge-based Artificial Intelligence framework. In the automatic mode the accuracy ranges from 75 to 91%. An Interactive Editor is used to increase the accuracy to > 97% in signal epochs of about 30-s duration. The accuracy was verified by comparing the firings of action potentials from the EMG signals detected simultaneously by the surface sensor array and by a needle sensor. We have decomposed up to six MU action potential trains from the sEMG signal detected from the orbicularis oculi, platysma, and tibialis anterior muscles. However, the yield is generally low, with typically < or = 5 MUs per contraction. Both the accuracy and the yield should increase as the algorithms are developed further. With this technique it is possible to investigate the behavior of MUs in muscles that are not easily studied by needle sensors. We found that the inverse relationship between the recruitment threshold and the firing rate previously reported for muscles innervated by spinal nerves is also present in the orbicularis oculi and the platysma, which are innervated by cranial nerves. However, these two muscles were found to have greater and more widespread values of firing rates than those of large limb muscles.

The electromyographic signal as a presymptomatic indicator of organophosphates in the body.

Organophosphate (OP) compounds are present in household and agricultural pesticides as well as in nerve agents. The toxic effects of these chemicals result from their anticholinesterase activity, which disrupts nerve junctions and parasympathetic effector sites, leading to a variety of symptoms and possible death. When the anticholinesterase agents in OP compounds reach the neuromuscular junction, they cause a disruption in the firing of muscle fiber action potentials. This effect has the potential of altering the time course of the electromyographic (EMG) signal detected by surface electrodes. We investigated the association between OP compound dose, surface EMG changes, and overt signs of OP toxicity. Daily doses of 10-15 microg/kg of diisopropylfluorophosphate (DFP) were injected into the calf muscle of four rhesus monkeys while surface EMG signals were recorded from two thigh muscles bilaterally. With increasing number of doses, the EMG signal presented an increasing number of time gaps. The presence of the gaps was evident prior to any overt symptoms of cholinesterase toxicity. These findings can lead to the development of noninvasive technology for indicating the presence of OP compounds in muscle tissue prior to clinical abnormalities.

Multi-receiver precision decomposition of intramuscular EMG signals.

The precision decomposition technique can accurately identify a significant number of action potential trains within intramuscular electromyographic (EMG) signals. The original version of this technique (PD I) often requires extensive user-interactive editing to improve upon the results from a maximum a-posteriori probability receiver (MAPR). We have used the integrated processing and understanding of signals methodology from artificial intelligence to formulate and implement a new multi-receiver solution that augments MAPR with two other receivers to gain greater accuracy. Specifically, each new receiver utilizes an interleaving of signal and symbol processing stages to address MAPR inadequacies in resolving cases of acute superposition and shape instability among motor unit trains. Prior to any user-interactive editing, our multi-receiver system achieves a classification accuracy of 85.1%, a significant improvement over the 66.0% accuracy of PD I on the same database of challenging EMG signals.

Functional activity monitoring from wearable sensor data.

A novel approach is presented for the interpretation and use of EMG and accelerometer data to monitor, identify, and categorize functional motor activities in individuals whose movements are unscripted, unrestrained, and take place in the "real world". Our proposed solution provides a novel and practical way of conceptualizing physical activities that facilitates the deployment of modern signal processing and interpretation techniques to carry out activity monitoring. A hierarchical approach is adopted that is based upon: 1) blackboard and rule-based technology from artificial intelligence to support a process in which coarse-grained activity partitioning forms the context for finer-grained activity partitioning; 2) neural network technology to support initial activity classification; and 3) integrated processing and understanding of signals (IPUS) technology for revising the initial classifications to account for the high degrees of anticipated signal variability and overlap during freeform activity.