Advances for Dysphagia in the Digital Age: Integrative Imaging and Wearable Technologies Free

Swallowing is one of the most complex neurophysiological functions in the body, involving both sensory and motor components and all levels of the nervous system for its effective implementation. Because of this, disorders and diseases affecting any of these levels or areas of control may result in swallowing disorders (a.k.a. dysphagia). Indeed, dysphagia is common in many neurological and structural diseases affecting the head and neck, such as stroke, cerebral palsy, Parkinson’s disease, head and neck cancer, trauma, and more [1, 2]. Critically, untreated dysphagia interferes with a patient’s ability to eat and drink, and can lead to devastating health consequences such as malnutrition, dehydration, and respiratory infections [3, 4]. In addition to its impact on health, the social, cultural, and psychological ramifications of dysphagia in our highly food-centric society can also not be overstated [5‒7]. These consequences highlight the critical importance of diagnosing and treating dysphagia timely and effectively.

In the past ∼40–50 years, the field of swallowing science has made significant strides in increasing our understanding of this complex neurophysiological act, as well as how to evaluate and treat swallowing disorders. Regarding assessment, improvements have been seen through the development and validation of standardized screening and assessment protocols (e.g., [8‒11]), standardization of materials used in our assessments/treatments (e.g., Bracco Varibar barium contrast, International Dysphagia Diet Standardization Initiative [IDDSI]), and the addition of key technologies (e.g., endoscopy and manometry) in the evaluation of dysphagia [12‒15]. In treatment, we have observed increases in our understanding of the use and effectiveness of compensatory strategies and dietary modifications [16‒18], the development of devices to facilitate management efforts [19‒23], and a critically important increased focus on rehabilitation [24‒27], largely due to our enhanced understanding of the supramedullary contributions to swallowing [28‒30]. Despite these promising steps, optimal evaluation and management approaches for dysphagia have yet to be established, and current treatment paradigms remain with limited efficacy [19, 23, 31, 32].

Though many reasons exist to explain this paucity of knowledge, several authors agree that a major hindrance stems from our relatively poor understanding of the complex underlying swallowing mechanisms across all levels of control [26, 33‒35]. This lack of detailed knowledge limits our ability to further examine how these mechanisms may be altered in patients with different types of dysphagia and, more importantly, how to optimally target these mechanisms/areas for acquisition or re-acquisition of swallowing skills. To overcome part of this important hindrance, it is critical that we increase our understanding of the complex nature of swallowing control by developing and validating sensitive new tools and methods that will allow for (a) the precise and personalized examination of this control and any physiological changes that patients develop, and (b) the targeting of these changes to improve treatment effectiveness. We argue that now is the ideal time to act in this direction.

Living in what appears to be the peak of the digital age, our field has begun leveraging some of the enormous technological advancements in fields such as imaging, electrophysiology, wearables, and machine learning, which can act as catalysts for the future of medicine and dysphagia management. Herein, two examples of such technological advances developed by the author’s multidisciplinary team are described, as well as their current state and the role they will play in the future of personalized dysphagia care.

Our early understanding of swallowing control stemmed largely from seminal basic animal research in the 1950s which provided evidence for the role of a complex brainstem nuclei network in the swallowing sequence [36, 37]. This was further validated by early work in humans using peripheral electrophysiological measures (electromyography [EMG]) and specifying elements of neuromuscular swallowing control [38‒40]. However, in later years a substantial body of clinical and neuroimaging research showcased additional supramedullary contributions in the neural control of swallowing [29, 41‒46] highlighting the elaborate nature of this vital biological act. Specifically, this research identified a rather large bilateral network of cortical and subcortical areas involved in healthy swallowing [29, 30, 41, 44‒46] and provided initial support for the shift in swallowing management efforts toward neuroplasticity-driven paradigms that could engage these areas and theoretically lead to more long-lasting treatment outcomes [26, 47]. More recently, studies involving patients with neurogenic dysphagia, such as stroke, cerebral palsy, and Parkinson's disease, have attempted to map brain lesions or pathologies and correlate them to dysphagia symptomatology [48‒53] suggesting how affected brain areas may be related to biomechanical or behavioral swallowing components.

This prior work has offered important initial insights into the control of healthy swallowing and suspected mechanisms of dysphagia in neurogenic populations. However, the findings thus far have been mostly correlational, leaving several unanswered questions about the specificity of these mechanisms. For example, we still do not know the direct neurophysiological drivers of specific swallowing events, how these drivers/areas connect/communicate with each other, and what neuromuscular and neurophysiological changes are developed in patient populations. This gap is significant because it prohibits us from developing physiology-based interventions based on these changes that will ultimately be more effective. Acquiring direct and precise information on these mechanisms has been further recognized as a necessary next step in the development of personalized therapeutic approaches for dysphagia and in promoting improvements in prognoses and treatment outcome research [33‒35].

To begin addressing this gap and help advance the management of dysphagia, it will be critical to examine the advantages and limitations of current approaches and multidisciplinary collaborative efforts to develop novel tools, while also taking advantage of emerging digital innovations. The efforts detailed below represent two examples of the author’s multidisciplinary team work in this direction.

In order to effectively treat dysphagia, it is critical to determine what aspects of the sensorimotor swallowing control have been disrupted, the pathophysiology underlying this disruption, and the neurophysiological changes that have been developed as a result. This information can then be used in developing treatment paradigms specific to these changes and their unique pathophysiological profile.

To identify disruptions and changes at the biomechanical/movement level in clinical practice, single-modality imaging techniques such as videofluoroscopy and flexible endoscopy are commonly used. These well-established techniques provide dynamic visualization of the swallow and allow for some level of detailed biomechanical measurements that can be completed by the end user/clinician. However, such measurements require substantial training and effort and thus, unfortunately, are not widely used in clinical practice. Some groups have attempted to use dynamic fast magnetic resonance imaging (MRI) to visualize oropharyngeal swallowing as well [54‒58]. Dynamic or real-time MRI has certain advantages over videofluoroscopy and endoscopy. First, it allows for better visualization of soft tissues (e.g., muscles), and second, unlike VFSS, it does not require the use of ionizing radiation or barium for the visualization and evaluation of certain biomechanical swallowing events [57]. Main technical challenges of this technique relate to its relatively low imaging frame rate compared to VFSS which limits image quality, and to its susceptibility to artifacts due to magnetic differences that exist at air-tissue edges which are numerous in the oropharynx [59]. Importantly, none of the current single-modality imaging approaches offer direct or objective measurements of swallowing physiology or neurophysiology.

To identify disruptions/changes at the central level of swallowing control, two primary types of research paradigms have been utilized through the years. One includes the use of functional neuroimaging to identify brain activations during swallowing and related tasks [29, 41, 44, 45], and the other involves mapping of brain lesions/pathologies on structural images and correlating them with dysphagia symptoms [48‒50]. Both paradigms can offer valuable insights on the central swallowing control, but also present challenges primarily related to the analysis and interpretation of their findings.

For functional neuroimaging of swallowing, the use of functional MRI (fMRI) has been rather prevalent. Task-based fMRI is a noninvasive technique that can identify brain activations through changes in blood oxygenation related to the execution of a task [28]. For accurate analysis and interpretation of fMRI-based activations, however, it is critical that the exact timing and form of the task are known to the experimenter. For swallowing fMRI experiments, this means that swallow events need to be performed in conjunction with a stimulus (visual, auditory, taste/sensory) and to be confirmed through monitoring devices, such as accelerometers or pneumographic bellows placed around the neck [46, 60‒62]. Critically, for patients with dysphagia it is not always possible to elicit a swallow in a timely fashion or upon command [63, 64], and the neck placement of monitoring devices can be uncomfortable (especially while swallowing in the supine position) and can result in sensory feedback affecting task completion and brain activity [59]. For the research studies involving mapping of brain lesions/pathologies and correlating them with dysphagia symptoms [48‒50], an important limitation is that they do not provide information on the direct drivers of specific swallowing events but only suggest potential correlations.

Each of the aforementioned imaging modalities provides valuable clinical knowledge on either the biomechanical or central levels of swallowing function. However, these modalities also have challenges that limit their ability independently to offer a comprehensive understanding of the underlying mechanisms at all levels of control. Finding a way to combine imaging modalities could be the answer to this clinical conundrum. A description of a novel, integrative imaging approach is presented next.

In 2011, the development of a new multimodal imaging technique, called SimulScan, was introduced by our team of engineers, neuroscientists, and speech-language pathologists/dysphagia experts [59]. This technology enabled the first simultaneous interleaved dynamic MRI of the oropharyngeal area and brain fMRI sequence to be possible. Doing so, it allowed us – for the first time – to simultaneously image oropharyngeal swallowing and full brain activation including cortical and subcortical components, providing unique opportunities to examine all levels of swallowing control (from peripheral structures and muscles to the brain) using one technique (MRI).

The initial SimulScan technology – described in detail in [59] – was developed to provide this unique integration. It comprised of interleaved dynamic and fMRI acquisition blocks based on fast low-angle-shot spiral sequences [65] (Fig. 1). For the dynamic imaging of the swallows, a 6-shot spiral-out fast low-angle-shot acquisition was used to image a 6-mm-thick midsagittal slice of the head and neck (240 × 240 mm FOV, 96 × 96 matrix, 2.5 × 2.5 mm resolution, TE 1.1 ms, flip angle 10°, TR 68.8 ms), while the dynamic frame rate was 14.5 frames per second [59]. The fMRI sequence was a single-shot spiral-in acquisition of 24 oblique axial slices (slice thickness 4 mm, 240 × 240 mm FOV, 64 × 64 matrix, 3.75 × 3.75 mm resolution, TE 25 ms, flip angle 80°, TR 1.6512 s). In a pilot validation experiment [59], three healthy young adults completed three 15-min SimulScan acquisitions while watching a movie on a 3T Siemens Magnetom Allegra MRI scanner. The participants were instructed to relax and watch the film, while the purpose of the study was not revealed until after experiment completion. This allowed examination of the oropharyngeal dynamic events and brain activations during spontaneous un-cued swallows. There were two key findings from this pilot development and validation study. First, swallow events and onsets were reliably identified through both the time series of the dynamic imaging signal-to-noise ratio and visual inspection of the images. Second, brain activations of the natural swallows of the three subjects (Fig. 2) paralleled findings from several prior task-based swallowing fMRI studies [44, 45, 61, 62], providing additional preliminary validation for the fMRI portion of the sequence. A critical limitation of this first version of the SimulScan technology is that it required a high number of shots to reduce magnetic susceptibility artifacts, thus restricting the frame rates and subsequently the dynamic image quality. As such, although the swallows were easily identifiable at the 14.5 frames per second frame rate, accurate kinematic and temporal quantification of specific structures or events was not feasible with this first iteration.

In later years, Sutton and colleagues continued to work on advancing high-speed 3D dynamic MRI to evaluate oromotor behaviors during speech [66‒68] using an innovative image acquisition and estimation method known as the partial separability (PS) model. In this work, the researchers used the PS model and achieved high-quality dynamic images at up to 166 frames per second and 3D coverage of the vocal tract at 2 × 2 × 5 mm³ spatial resolution [66, 67]. For SimulScan, the PS model allows preservation of the high number of shots required for the simultaneous dynamic and fMRI imaging while also achieving a high spatiotemporal image with excellent dynamic imaging quality. Recently, our team showed that this new advanced SimulScan sequence (∼10-min duration) significantly improved the quality and speed of the dynamic portion of the sequence in a subject performing oral motor tasks [69].

In addition to improving the dynamic image quality, our team is currently working on incorporating morphometric and biomechanical measures [70, 71] and developing an automatic analysis platform, leveraging the multivariate partial least squares correlation to integrate the dynamic biomechanical measures and their correlated neural signals during spontaneous swallows. These additional improvements will be essential in ensuring precision in identifying the underlying swallowing mechanisms in a patient-centered and personalized manner. Upon optimization, SimulScan will be validated against current gold standard single-level modalities (fMRI and VFSS) in a group of healthy young and older adults, and in a group of adult patients with neurogenic dysphagia. At the end of the validation study, this novel technology would be made available in other (similar vendor’s) MRI facilities and could be easily added to standard MRI protocols – such as those typically completed during a diagnostic workup of stroke patients or other neurological conditions.

Refinement of this technology – and similar multimodal digital age imaging tools – would offer unparalleled precision in identifying the underlying swallowing mechanisms of a patient’s dysphagia, would complement current imaging technologies, and would further guide our efforts to improve specificity in targeting appropriate swallowing pathways in a personalized manner. High cost of the technology and swallowing in the supine position can be viewed as potential limiting factors. However, cost issues could be alleviated (or remain low) if the scan can be performed as part of a standard MRI diagnostic workup. Furthermore, swallowing spontaneously (saliva) in the supine position is also a naturally occurring act and would be the only type of swallow required for SimulScan completion. This would eliminate the need for patients to swallow food and liquid boluses, which could be unsafe for some populations, especially in the acute stages of disease.

Developments, such as the SimulScan, can provide critical insights on the biomechanics of the swallow and its direct underlying central neural control. Though some muscle tissue is visible in dynamic and structural MRI scans, muscle physiology can only be fully assessed using advanced electrophysiological methods [72]. Understanding the peripheral or neuromuscular underpinnings of swallowing and swallowing disorders is also critically important in identifying neuromuscular pathways that may be affected and need to be targeted in treatment. Recent developments in the area of wearable electromyography (EMG) technologies and machine learning can further offer electrophysiological data that can help us understand these pathways which could have both high diagnostic and therapeutic values.

EMG measures the electrical signals (action potentials) within a motor unit or across many motor units of a contracted muscle [72‒74]. These signals are typically pre-amplified and bandpass filtered, and then undergo further pre-processing and post-processing steps (including filtering, demeaning, full-wave rectification, smoothing, and normalization) in order to be clinically interpretable [72, 75]. Two types of EMG sensors (electrodes) are typically utilized to capture muscle activity signals, either intramuscular (i.e., wire or needle) or surface sensors [73]. Intramuscular sensors are inserted into the muscle body through the skin and can measure even single muscle activity (though this is hard to achieve in the small and often overlapping head and neck area muscles [76]). Therefore, intramuscular sensors are precise and can offer valuable diagnostic information especially for large muscles, but are considered invasive and could result in some side effects, such as pain or bleeding [77]. Surface EMG (sEMG) sensors are more frequently used, especially in clinical swallowing practice, because they only require simple placement on the skin, are not invasive, and allow for muscle activity recording from groups of muscles [75]. The information they provide is less precise, but they are more user-friendly and are thus becoming increasingly more popular as a tool to provide therapeutic biofeedback in dysphagia management. Specifically, the main muscle group typically targeted in dysphagia therapy through the use of sEMG biofeedback is the suprahyoid or submental muscle group (i.e., mylohyoid, geniohyoid, and anterior belly of the digastric muscles) which is relatively easily accessible and known to contribute to the critical act of hyolaryngeal excursion during the pharyngeal swallow [39]. The diagnostic value of sEMG for dysphagia is currently limited, but could be significantly enhanced in the future, particularly if large datasets can be combined with machine learning approaches.

Importantly, using appropriate sEMG signal analysis methods (see [73] for a detailed review), sEMG signals can provide valuable outcome variables (Table 1) that can be highly useful in dysphagia management. Specifically, upon post-processing, information can be obtained on amount/level of muscle contraction during a task (i.e., normalized amplitude or area under the curve) [78, 79], on duration of muscle activation (i.e., burst duration) [79], as well as on speed of activation/reaction time [78], and on measures of coordinative synchrony between muscle groups or pairs [80]. All these neuromuscular components may be affected differentially in patient populations depending on the level and type of motor involvement, and thus, identifying these patterns could have diagnostic and prognostic significance, while also offering physiology-based therapeutic targets.

Currently, most commercially available sEMG systems allow for measurement or visualization of amplitude and burst duration of swallow events and are therefore mostly used as crude biofeedback systems during swallow therapy. Further, most remain large, and/or expensive, and thus are primarily available in large urban clinical centers. In recent years, wireless portable sEMG sensors have started to emerge, including the Trigno™ Avanto Platform (Delsys) and the MobiliT (True Angle Medical Technologies). However, these sensors have a mostly rigid form (which is not ideal for the curvilinear surface of the submental region), allow measurement on one muscle location per sensor (e.g., measurements from one side of the neck) or require multiple sensors to be used at once to cover the orofacial or submental areas, and overall remain costly. Some of the newer highly flexible wearable sensors that have emerged are designed for large areas of the body (e.g., chest or legs) [81] or are multi-sensors that remain experimental [82]. Further, despite commercialization efforts of several of these technologies, their efficacy in diagnosing or treating swallowing function remains understudied. This highlights the need to (a) develop swallowing-specific sEMG systems that will allow for accurate measurement and therapeutic targeting of all relevant neuromuscular outcome variables outlined in Table 1 and (b) systematically examine their efficacy in improving physiological, functional, and neurophysiological swallowing outcomes.

To start addressing this need, our team has developed a novel prototype wearable sEMG system (i-Phagia), comprised of a thin, flexible, self-contained, non-stretchable sensor patch (Fig. 3) designed to simultaneously record high-quality sEMG recordings from both sides of the submental area. i-Phagia includes a handheld wireless unit and custom-built software that is downloaded on a personal device and which provides immediate feedback to the patient as well as adherence data to the clinician [83]. This technology is designed to be a small, cost-effective, self-contained, and portable swallowing-specific sEMG device ultimately optimized for both therapeutic and diagnostic purposes.

To initiate this work, in 2019 we presented a first-generation sEMG sensor patch technology which consisted of an ultrathin flexible and stretchable sEMG multi-sensor patch that used a honeycomb design ([22], patent pending). In a randomized clinical trial (RCT) of 40 healthy older adults (age range: 53–85), this first-generation sensor technology was shown to be equivalent to commercially available wired sEMG sensors in signal quality and technical performance [79]. However, it had a short lifetime (maximum 5 uses) and required the use of an additional external ground electrode for signal integrity, and accurate submental placement was time-consuming (∼20 min), making it non-clinically feasible [79]. These limitations motivated the development of a more robust second-generation sensor patch that is the current major component of i-Phagia (Fig. 3a). The new sensor patch (Fig. 3; patent pending) is fabricated on a slightly thicker, flexible but non-stretchable double-sided thin circuit board and includes 5 electrodes/pads on one layer and traces transferring the electrical signal from the pads to the wireless unit on the back/second layer of the device [83]. This separation of the components eliminates opportunities for signal contamination, as evidenced by our subsequent RCT of 60 healthy adults (30 young and 30 older) showcasing (a) improved signal performance (average S/N achieved with new sensor patch 25.3 [Malandraki et al. [83], in preparation] compared to 20.4 in prior RCT [79] with the first-generation patch) and (b) substantial increases in patch durability (>20 uses) critical for clinical translation. The i-Phagia sensor patch further maintains optimal layout; i.e., the interelectrode distance for the active pads/electrodes (Fig. 3a, white circles) is 1.5 cm from edge to edge, so that the pads are aligned with the submental muscle fibers for optimal electrode placement [73, 74, 79]. It also includes a ground electrode within the patch (Fig. 3a, blue circle) designed to come into direct contact with the middle line of the mandible (just below the mental protuberance). This addition has key advantages: it helps maintain the apparatus self-contained and eliminates the need for a separate ground electrode, and it provides a point of reference for placement and facilitates consistent self-placement of the sensor in the submental area.

Currently, the submental patch is connected to a commercial wireless unit via a flexible ACF cable allowing for remote data transmission. We have also developed a prototype software (copyright pending) that includes all necessary pre-processing and normalization steps [73, 79], allows for the recording of submental muscle activity during swallows and swallow maneuvers, measurement of amplitude and timing parameters, and incorporates a user-friendly interface. Current development of the software is ongoing to incorporate more refined measurements of reaction time and coordinative synchrony to allow for comprehensive swallowing neuromuscular profiling of patients’ swallows.

Optimization of this technology – and similar wearable technologies – could offer several critical improvements in current dysphagia and orofacial disorder management. First, it would offer a way to evaluate and target more granular neuromuscular components than is currently possible. Importantly, such tools could be adopted for both in-person and telehealth service provision expanding rehabilitation options and increasing treatment access. Wearable technologies further offer the opportunity for data to be collected for long periods of time and even throughout the day. Such large datasets combined with advanced machine learning and artificial intelligence analysis methods [84, 85] could provide essential insights into diagnostic or even prognostic patient profiles with far-reaching clinical implications in the future. Considerations that need to be addressed include issues related to remaining technical challenges specific to wearables, such as limited post-processing abilities, storage or battery life; clinical parameters, such as efficacy and clinical translation; as well as cost and patient and clinician acceptability [86]. New developments in on-chip machine learning hardware technology are expected to address many of the current technical challenges [87]. In regard to clinical challenges, efficacy testing and clinical translation can be accelerated through enhanced research-industry relations and will be essential in securing reductions in cost through reimbursement avenues. For improved acceptability and overall clinical adoption, inventors and researchers are highly encouraged to include patients and clinicians as critical stakeholders in the development and validation of such technologies.

Shifting clinical practice from symptom-based to physiology-based therapeutic targets and interventions has been recognized as a critical priority for the field of dysphagia research and practice [26, 34]. In order to make this shift, a deeper understanding of the underlying swallowing mechanism at all levels of control is essential. Upon optimization and validation, technological advancements such as those showcased in this manuscript, and many others under development, can offer unprecedented opportunities to gain direct and precise insights on this control. Information gained from these new technologies can act as a catalyst for the future development of personalized and optimized dysphagia care. By leveraging advances in current methods, multidisciplinary collaborations, and new digital age technologies and methods, the field of dysphagia can take the next giant leap forward in improving clinical care and patient lives.

The author wishes to thank Jaime Bauer Malandraki for her help with editing the final version of the manuscript.

Financial relationships: Georgia A. Malandraki is employed by Purdue University. Part of the work described in this review paper (sEMG sensor technology) has been supported by Purdue University, the Purdue Research Foundation, and the National Institute of Biomedical Imaging and Bioengineering R21 Trailblazer Grant (Grant 1R21EB026099-01A1, PIs: Malandraki and Lee). Also, she is one of the main inventors of the wearable sEMG technology (patents pending) described in the current review paper. Nonfinancial relationships: Georgia A. Malandraki is the Immediate Past President (Board Member) of the Dysphagia Research Society.

Part of the work described in this review paper (sEMG sensor technology) has been supported by Purdue University, the Purdue Research Foundation, and the National Institute of Biomedical Imaging and Bioengineering R21 Trailblazer Grant (Grant 1R21EB026099-01A1, PIs: Malandraki and Lee).

G.A.M. conceptualized, synthesized, and wrote the manuscript.