Non-invasive monitoring of functionally distinct muscle activations during swallowing
Introduction
Dysphagia, or difficulty swallowing, is a common complication of stroke, occurring in approximately 30–42% of acute stroke patients requiring hospital admission (Gordon et al., 1987, Odderson et al., 1995). Of those initially affected, approximately 50% spontaneously recover to a normal swallow within 14 days after the onset of the stroke event (Bath et al., 2000). However, those who remain affected show much slower recovery rates. The prevalence of clinically diagnosed dysphagia in patients at 1 month poststroke has been reported to be 2–21% and can remain as high as 7% at 3 months poststroke (Martino et al., 2000). Due to the changes in population demographics, the incidence of dysphagia can be expected to progressively increase over the next few years. Early identification of dysphagia is essential to ensure timely treatment and the prevention of complications (Martino et al., 2000), as dysphagia puts subjects at significant risk for increased morbidity and mortality, including aspiration pneumonia, reduced functional independence and malnutrition.
Aspiration pneumonia is a well-recognized complication of dysphagia (Brin and Younger, 1988, Smithard et al., 1996, Teasell et al., 1996, Daniels et al., 1998). Daniels et al., ascertained that two-thirds of the patients after acute stroke demonstrating aspiration were ‘silent’ aspirators (Daniels et al., 1998), that is, not demonstrating overt clinical signs of aspiration, emphasizing the need for early and accurate evaluation of swallowing function.
Swallowing is normally assessed with a clinical examination by a skilled Speech Language Pathologist (SLP). The SLP typically assesses each of the 4 phases of normal swallowing: the oral preparatory phase, the oral transfer phase, the pharyngeal phase (which includes swallow reflex initiation and pharyngeal function and clearance) and the esophageal phase. Subjects with dysphagia may have varying amounts of abnormality in each of these phases.
The clinical examination is often supplemented with ancillary tests. Videofluorography, a method that has been widely used to illuminate the path taken by the bolus as it completes the 4 phases of swallowing, is useful in demonstrating any evidence of aspiration. The newer fiberoptic endoscopic evaluation of swallowing (FEES) studies enable the clinician to more directly assess swallowing. This procedure involves passing a small, flexible tube connected to a camera and a light source, through the nose to the top of the throat, allowing direct visualization of the swallowing process on a TV monitor.
Direct visualization of human swallowing as well as animal studies make clear that swallowing is inherently a complicated motor movement. Traditional models of swallowing have concentrated on the brainstem integration of sensory and motor inputs (Sessle and Henry, 1989, Bieger, 1993), but there is clear evidence that swallowing is a muscular activity that requires the complex integration of motor, sensory and attentional resources that other motor functions require (Hamdy et al., 1999, Zald and Pardo, 1999). For example, an event-related fMRI study of swallowing revealed consistent activation in caudal sensorimotor cortex, anterior insula, premotor cortex, frontal operculum, anterior cingulate and prefrontal cortex, anterolateral and posterior parietal cortex and precuneus and superiomedial temporal cortex (Hamdy et al., 1999). It is therefore not surprising that recovery of swallowing corresponds to re-organization of the motor cortex (Hamdy et al., 1998).
Perhaps more than other motor acts, swallowing incorporates the use of central pattern generators (CPGs; Dick et al., 1993, Arshavsky et al., 1997, Broussard and Altschuler, 2000), i.e. brainstem or spinal structures that can encode complex sequences of muscle activity. However, theoretical and clinical evidence supports the notion of CPGs in the cortical control of the limb muscles in a variety of vertebrates, including humans (Bussel et al., 1996, Dimitrijevic et al., 1998, Tresch et al., 1999).
The presence of CPGs has implications for methods assessing the intricate firing pattern of coordinated muscles such as is seen with deglutition. If swallowing is under the control of CPGs which distribute cortical commands across several muscles, there is a need to look at several muscles simultaneously as more than one muscle may act as part of a functional unit. However, the standard method of recording muscular activity during swallowing is to record the electrical activity of one, or at most, a few muscles with EMG (Blanton et al., 1970, Shipp et al., 1970, van Overbeek et al., 1985, Trigos et al., 1988, Perlman et al., 1989, Perlman et al., 1999).
EMG is usually performed with a monopolar or bipolar needle electrode, or with surface electrodes. Each method has relative strengths and weaknesses. With needle EMG, especially in the limb muscles, it is possible to examine the morphology of individual motor units, suggesting either myopathic or neuropathic causes for muscle weakness (Kimura, 1989). This requires a partial voluntary contraction, which may be more problematic for swallowing studies. Also, since dysphagia is frequently caused by disruption of upper motor neuron control, examination of the motor unit morphology looking for neuropathic or myopathic motor units adds little further information. The EMG is thus used mostly for relative timing of the musculature.
Needle EMG of the swallowing musculature is more challenging than that for the limb muscles. Although it appears to be a safe procedure (Mu and Yang, 1990), there is some discomfort for the patient, especially if some time is taken to manipulate the needle to ensure that it is in the muscle of interest, which may be deep to other overlapping muscles not directly related to swallowing. In some cases, anesthetic is used, which may alter the sensory and motor cues normally used in swallowing (Ertekin et al., 2000c), and hence affect the interpretation of the results. It is usually practical to have at most a few needle electrodes inserted simultaneously (Ertekin et al., 1998a, Ertekin et al., 1998b, Ertekin et al., 2000a, Ertekin et al., 2000b, Ertekin et al., 2000c, Ertekin et al., 2001a, Ertekin et al., 2001b, Ertekin et al., 2001c).
In contrast, surface EMG (sEMG) electrodes provide little or no discomfort, allowing several muscles to be examined simultaneously. The main disadvantage of sEMG is the problem of ‘cross-talk’, whereby several different muscles may contribute to the recording of a given electrode, making the source of the signal difficult to interpret. This is of particular concern with swallowing sEMG, where several muscles involved with swallowing are in close proximity and deep to the surface. This is probably why the role of sEMG in swallowing has been largely restricted to its therapeutic potential in biofeedback (Sukthankar et al., 1994, Barofsky, 1995, Huckabee and Cannito, 1999). Another serious problem is that the raw sEMG recordings are not necessarily reliable; that is, repeated measures of the same movement may give varying results (Boline et al., 1993, Goodwin et al., 1999).
Recently, McKeown and coworkers have suggested a means to account for cross-talk between electrodes thereby enabling the extraction of reliable and robust activations from sEMG recordings (McKeown, 2000, McKeown and Radtke, 2001). This method, which involves calculating the independent components (ICs; Comon, 1994) of several simultaneous sEMG recordings, appears to produce robust activations that demonstrate less trial-to-trial variability than the raw sEMG recordings.
Here, we give a brief qualitative description of the Independent Component Analysis (ICA) technique; the reader is referred to (Bell and Sejnowski, 1995) for an in-depth description of the computation method, (Jung et al., 2001) for a general review of its application to biomedical signals, and (McKeown, 2000, McKeown and Radtke, 2001) for a detailed discussion of the application of ICA to sEMG.
ICA was originally developed to solve the ‘cocktail party problem’. Consider for example, 4 individuals sitting around a table, speaking independently and that there are 4 microphones on the table recording the different voices. Each microphone will record a mixture of the 4 different voices, but the relative contribution of each voice to the microphones will differ. The goal of ICA is, given only the recordings from the microphones, but not knowing their positions or what each person was saying, is it possible to separate out the individual voices? ICA solves this by assuming that the voices were independent and trying to carefully combine the voice-mixture recordings in just such a way that all voices but one cancel each other out. Stated more rigorously, ICA attempts to find linear combinations of the recordings so that the combinations are as statistically independent as possible.
This simple example is illuminating in that it emphasizes many of the underlying properties and assumptions of the ICA analysis technique. Note that extracted ICs (the voices in the example above) from an ICA analysis are invariant to the exact placement of the microphones. Once the individual voices are determined, however, it is straightforward to infer where the microphones must have been placed to give the mixed-voice recordings. Also note that ordering of the separated voices is somewhat arbitrary, although it may be possible to rank the components in order of the amplitude they contribute to the original recordings, for example. The above scenario demonstrates that a key ICA assumption is that the number of microphones equals the number of speakers. Also note that in the example above, the ICA algorithm assumes that the speakers are not moving. If a subject were to walk around the table and say, ‘See Jane run’, ICA could interpret this as one subject who said, ‘See’, another subject sitting at another part of the table, saying, ‘Jane’ and a third at another position at a later time saying ‘run’.
Considering this example as an analogy of the ICA analysis of sEMG of swallowing, while being imperfect, is illustrative. Like the above microphones, the sEMG electrodes will, due to cross-talk and common cortical influence, record a mixture of the electrical activity from different muscles contributing to the swallowing. Note that each ‘voice’ separated out may not necessarily correspond to the activity of anatomically defined muscles. Rather, each ‘voice’ will correspond to functional combinations of muscles that tend to act independently of the other functional units, and their ordering will be arbitrary. With the ICA approach, the contribution of individual anatomically defined muscles to the overall swallowing act, the emphasis of other published reports (e.g. Palmer et al., 1999), is not stressed. Rather, the relationships between channels are highlighted.
Just as the extracted voices in the example were invariant to the position of the microphones, the ICA components of the sEMGs will tend to be robust to the exact placement of the sEMG electrodes, important for serial monitoring of subjects. In fact, in the extreme case, if the positions of electrodes are accidentally switched, the temporally ICs will theoretically remain unchanged.
The assumption of ICA that the spatial relationship between the sEMG electrodes is constant has implications for the recording of the sEMG of swallowing, as laryngeal and tongue movement is an integral part of the swallowing act.
Estimating the actual number of components in the swallowing EMG, analogous to the number of voices in the scenario above, could potentially be difficult. In practice, the number of components that can be extracted relates to the underlying signal-to-noise ratio of the data. We have often found it reasonable to first reduce the dimensionality of the data by principal component analysis (PCA) to capture an arbitrary percentage (e.g. >95%) of the variance of the data (McKeown et al., 1998, McKeown and Radtke, 2001). By varying the number of principal components retained, any number of components, up to a maximum of the number of electrodes placed, can be extracted.
In this paper, we apply ICA to the sEMG of normal subjects repeatedly swallowing liquids of differing consistency. We demonstrate that the ICs of the sEMG are robust, demonstrate less trial-to-trial variability than the raw sEMGs and appear to correspond to known physiological processes of swallowing.
Section snippets
General
All research was approved by Duke University's Institutional Review Board. Surface EMG recordings were obtained from 7 normal subjects (6 females, 1 male, ages 22–37). After signing informed consent, an alcohol swab was used to clean the skin before 15 standard EKG electrodes with ‘snap-on’ connectors were applied to the face and throat (Fig. 1). Electrodes were placed submentally, at the corners of the mouth, over the center of the buccal muscle bellies, 2 cm superior to the suprasternal notch
Results
The number of swallows recorded from subjects ranged from 30 to 56. Dimension reduction took about 5 s, and the calculations of the ICs took approximately 50 s on a Windows NT machine with a Pentium II Xerion processor.
The integrated and rectified sEMG demonstrated considerable swallow-to-swallow variability (Fig. 4, left). In contrast, the ICs demonstrated less variability when their unaveraged values were overplotted (Fig. 4, right).
The components isolated were specific to that individual,
Discussion
We have demonstrated that it is possible to derive individually specific, reproducible, functionally independent components from non-invasive monitoring of the sEMG during swallowing. The temporal profile of the components appeared to reflect known activity of some of the muscles normally obtained only through invasive needle EMG. The robustness of the components and intersubject variability of the components (Fig. 6) is compatible with previous descriptions using needle EMG, although we note
Acknowledgements
The authors are grateful to Leslie Collins for her help with collecting much of the data, and Milisa Batten, Juli Trautman, Maura English and Bonita Adams, all SLPs at the Durham Rehabilitation Institute, for their encouragement and helpful comments. The cartoon of the face used in the figures is part of Corel Draw's clip art collection (CorelDRAW®).
References (57)
- et al.
Botulinum toxin injection of the cricopharyngeus muscle for the treatment of dysphagia
Otolaryngol Head Neck Surg
(2000) - et al.
Pattern generation
Curr Opin Neurobiol
(1997) - et al.
Neurologic disorders and aspiration
Otolaryngol Clin North Am
(1988) - et al.
Aspiration in patients with acute stroke
Arch Phys Med Rehabil
(1998) - et al.
Electrodiagnostic methods for neurogenic dysphagia
Electroenceph clin Neurophysiol
(1998) - et al.
Cricopharyngeal sphincter muscle responses to transcranial magnetic stimulation in normal subjects and in patients with dysphagia
Clin Neurophysiol
(2001) - et al.
Recovery of swallowing after dysphagic stroke relates to functional reorganization in the intact motor cortex
Gastroenterology
(1998) - et al.
Laryngeal electromyography for the diagnosis and management of swallowing disorders
Otolaryngol Head Neck Surg
(1997) - et al.
Swallow management in patients on an acute stroke pathway: quality is cost effective
Arch Phys Med Rehabil
(1995) - et al.
Design and development of portable biofeedback systems for use in oral dysphagia rehabilitation
Med Eng Phys
(1994)
A fuzzy logic diagnosis system for classification of pharyngeal dysphagia
Int J Bio-Med Comput
Pneumonia associated with aspiration following stroke
Arch Phys Med Rehabil
Surface electromyographic biofeedback and the patient with dysphagia: clinical opportunities and research questions
Dysphagia
An information-maximization approach to blind separation and blind deconvolution
Neural Comput
Central nervous system control mechanisms of swallowing: a neuropharmacological perspective
Dysphagia
Electromyographic analysis of the buccinator muscle
J Dent Res
Interexaminer reliability of eight evaluative dimensions of lumbar segmental abnormality: part II
J Manipulative Physiol Ther
Central integration of swallow and airway-protective reflexes
Am J Med
Evidence for a spinal stepping generator in man. Electrophysiological study
Acta Neurobiol Exp
Independent component analysis: a new concept?
Signal Process
Cricopharyngeal function and dysfunction
Dysphagia
Interaction between central pattern generators for breathing and swallowing in the cat
J Physiol
Evidence for a spinal central pattern generator in humans
Ann N Y Acad Sci
An electrophysiological investigation of deglutition in man
Muscle Nerve
Clinical and electrophysiological evaluation of dysphagia in myasthenia gravis
J Neurol Neurosurg Psychiatry
Mechanisms of dysphagia in suprabulbar palsy with lacunar infarct
Stroke
The electromyographic behavior of the thyroarytenoid muscle during swallowing
J Clin Gastroenterol
Cited by (57)
Oropharynx
2018, Dysphagia Evaluation and Management in OtolaryngologyNeural and cortical analysis of swallowing and detection of motor imagery of swallow for dysphagia rehabilitation—A review
2016, Progress in Brain ResearchCitation Excerpt :Videofluoroscopy is an X-ray technique that provides a detailed anatomical assessment by illuminating the path taken by the bolus (mixed with barium) as it completes the four phases of swallowing (Hamdy et al., 2000; McKeown et al., 2002). The newer fiberoptic endoscopic evaluation of swallowing (FEES) technique enables more direct assessment (McKeown et al., 2002). FEES examines motor and sensory functions of swallowing so that proper treatment can be given to patients with swallowing difficulties.
The normal swallow: Muscular and neurophysiological control
2013, Otolaryngologic Clinics of North AmericaCitation Excerpt :In the following section, unique features of the muscular control for each phase are reviewed. The oral preparatory phase is the first phase of swallowing, during which food is broken down (via mastication) and a cohesive bolus is formed.14,15,18–20 Many muscles are active during this phase and help to contain liquid and/or food within the oral cavity.
Toward a robust swallowing detection for an implantable active artificial larynx: a survey
2023, Medical and Biological Engineering and Computing