The Impact of AI on the Development of Multimodal Wearable Devices in Musculoskeletal Medicine

Abstract

As wearables are becoming an increasingly important part of wellness and everyday life for many people, their potential in healthcare is also expanding, particularly in personalized and remote healthcare. However, many wearables lack sophistication, relying on simple sensors such as accelerometers and pulse meters to measure heart rate, body composition, and daily activity. Such basic metrics are insufficient for musculoskeletal disease diagnosis, which requires more detailed, multimodal neuromusculoskeletal monitoring. A major challenge in wearables development is the need for precise electromechanical signal measurements, which are difficult to obtain with low-cost systems. Artificial intelligence (AI) holds promise in addressing these analytical challenges and enabling the creation of affordable, sophisticated wearables. While AI has been used for decades in engineering, its clinical application is still emerging, creating an opportunity for the development of AI-enhanced wearables capable of clinical diagnosis. AI can enhance data generated by various sensor types in wearable devices (such as accelerometers, electrical, optical, and acoustic sensors), enabling clinicians to monitor and diagnose complex conditions that require multiple sensing modalities. This review explores current wearable technologies, ongoing research in AI-enhanced wearables, the potential for AI to advance wearable technologies in healthcare, and the future directions in the development of multimodal wearables.

Plain Language Summary

How AI Is Helping Build Smarter Wearable Devices for Muscle and Joint Health

Wearable devices like fitness trackers and smartwatches are becoming more common in everyday life and show promise for use in healthcare. But right now, most of these devices aren’t advanced enough to help doctors diagnose medical conditions—especially complex ones like muscle, nerve, and joint disorders. These conditions require more detailed data than just steps taken or heart rate. Developing wearables that can collect this kind of information is challenging, especially when trying to keep them affordable. That’s where artificial intelligence (AI) comes in. AI can help make sense of complex data from different types of sensors—like ones that track movement, electrical signals, light, or sound—making it possible to build smarter, more useful devices. This review looks at the current state of wearable technology, how AI is being used to improve it, and what the future might hold for creating affordable, AI-powered wearables that can actually help diagnose and monitor health conditions.

Introduction

The global healthcare wearables market is expected to surpass $324 billion by 2032 due to advances in sensor technology, machine learning (ML), and the Internet of Things, alongside a shift toward personalized, data-driven care [42,50,67,68]. Factors such as rising chronic diseases and the demand for remote monitoring drive this growth. While consumer wearables such as the Apple Watch incorporate multiple sensors (ie, accelerometers, optical heart rate monitors, and electrocardiographs [ECGs]), they fall short of the sophistication needed for clinical assessments and diagnosis [20,29]. This is also recognized in the Food and Drug Administration (FDA) classification of healthcare and general wellness devices and wearables as wellness devices, not healthcare devices.

To enter clinical use, devices must both provide reliable data for accurate diagnosis and screening and meet FDA safety standards. While safety concerns vary based on a device’s size, shape, and method of deployment, and can range from the risks associated with implantable devices to those that simply cause slight discomfort, several ethical concerns are consistently associated with all wearable devices. These include impacts on care relationships, privacy and justice issues, research ethics, marginalization of vulnerable populations, and the challenges of inadequate regulation and uncertain benefits—especially with newer technologies—making it essential to evaluate risks at every stage of design, development, and use [9]. The safety requirement can be addressed by developing safe, noninvasive devices, while the need for reliable data often requires multimodal devices that assess multiple aspects of the physiological system. For example, in the musculoskeletal system, the stretch reflex test, which evaluates neuromuscular interactions, could be significantly enhanced by multimodal wearables that capture both mechanical and electrical tissue behavior, expanding the diagnostic scope and precision beyond the current approach where physicians only observe the final muscle contraction. However, current limitations hinder this development.

Traditionally, analytical approaches in the development of biomedical devices have been limited to modeling relatively simple physiological behaviors, often requiring low-noise, high-fidelity signals to produce reliable outputs [36]. As a result, biomedical devices designed using such methods typically demand the acquisition of highly precise and clean signals to ensure compatibility with analytical models [27]. This limitation becomes even more pronounced in the context of multimodal wearable devices, which integrate diverse signal types such as acoustic and electrical data. In such systems, understanding the interactions and combined effects of these heterogeneous signals through purely analytical means becomes increasingly complex and often intractable [55].

With the advent of AI, there is now an opportunity to develop biomedical devices for interpreting more complex physiological patterns, even while using low-cost sensors and noisier signals. This shift opens the possibility for broader, more accessible, and cost-effective diagnostic and monitoring solutions that are less constrained by signal quality. This may involve integrating AI into the device development process, applying AI to enhance the interpretation of device-generated signals, or leveraging AI for a more comprehensive and holistic understanding of patient data, including demographic information, physiological signals, medical imaging, and clinical history. Our recent works on the development of AI-enhanced acoustic and electrical wearables for implant health monitoring show examples of this concept [17,23,57].

In this review, we discuss the differences between wellness and healthcare devices and wearables, the current technologies, the integration of AI and devices, and the gaps that need to be addressed to enable the entrance of biomedical devices, especially multimodal wearables, into the clinic.

Wearables in Wellness versus Healthcare

The distinction between medical and wellness devices lies in their purpose and regulatory oversight. Wellness devices are commonly designed for general health and fitness tracking rather than for diagnosing or treating medical conditions. These devices are considered low-risk products designed to promote a healthy lifestyle without direct involvement in the prevention, diagnosis, or treatment of a disease or medical condition [19]. This distinction was presented in 2016 when the U.S. Food, Drug, and Cosmetic Act was amended to exclude the definition of wellness devices from Section 201(h) for medical devices, thereby formally separating the two [66]. Consequently, the U.S. FDA does not enforce regulations on wellness devices, instead offering nonbinding policy suggestions for low-risk wellness products.

In contrast, the FDA explicitly defines medical devices, which can be summarized as any instrument, machine, or related article with the intention of diagnosing, treating, or preventing disease, or affecting the structure or function of the body without relying on chemical action or metabolism [65]. Devices meeting these requirements must provide detailed information on their intended use, design, materials, and manufacturing processes for risk classification [64]. Categorization is split into low, moderate, and high-risk classes with associated increases in regulatory control. While many Class I and some Class II devices, such as bandages and blood pressure cuffs, may be exempt from regulatory submission, most nonexempt Class II devices require a 510(k) clearance. Class III devices, for example, implantable defibrillators, necessitate premarket approval and often involve clinical trials before being placed on the market.

There is a growing interest in transitioning from wellness wearables to healthcare wearables, driven by several factors: (1) technological advancements in AI (multimodal AI algorithms), software (cloud computing and Internet of Things), and hardware (miniature sensors, fast microcontrollers); (2) the need for continuous monitoring due to growing chronic disease and aging population; (3) growing demand for remote monitoring driven by a shift to value-based care; (4) telemedicine and remote patient monitoring, and hospital-at-home models; (5) possibility of integrating wearables with electronic health record (EHR) and AI; and (6) reimbursement and regulatory support that enables reimbursements of remote patient monitoring devices. For these reasons, there have been large efforts in both the development of new healthcare wearables and validating applications of wellness wearables in the clinic. Table 1 shows a few examples of the current wearables in the market. Almost all of these wearable devices are running clinical studies to show their clinical relevance in the diagnosis, prognosis, and progress assessment of different diseases.

Table 1.

Some of the commercially available medical and wellness wearable devices, as of May 2025.

Device Name	Manufacturer	Domain	Price Range	Application
Apple Watch	Apple	Wellness	$199–$749	Monitors heart rate, activity levels, and sleep patterns.
Dexcom G6	Dexcom	Medical	$300–$1,000	Real-time blood glucose monitoring for diabetes management.
Fitbit Charge 5	Fitbit	Wellness	$150–$180	Tracks steps, caloric expenditure, heart rate, and sleep patterns.
Garmin Venu 2	Garmin	Wellness	$350–$400	Tracks steps, heart rate, sleep, and other fitness metrics.
KardiaMobile	AliveCor	Medical	$100–$150	Records six-lead ECG for cardiac monitoring and arrhythmia detection.
Medtronic InPen	Medtronic	Medical	$200–$400	Provides precise insulin dosing for diabetes management.
MyoWare Muscle Sensor	Advancer Technologies	Medical	$50–$100	Monitors muscle activity for rehabilitation and research.
Nonin WristOx2	Nonin	Medical	$250–$400	Measures oxygen saturation levels in patients with respiratory conditions.
Oura Ring	Oura	Wellness	$299–$399	Monitors sleep, activity, and provides personalized coaching.
Samsung Galaxy Watch 4	Samsung	Wellness	$250–$350	Monitors heart rate, activity, and sleep patterns.
WHOOP Strap 3.0	WHOOP	Wellness	$30/month (subscription)	Monitors heart rate, sleep, and provides recovery insights.
Zio AT	iRhythm Technologies	Medical	$200–$300	Provides continuous ECG monitoring and end-of-wear reports for cardiac care.
Withings Steel HR	Withings	Wellness	$180–$250	Monitors activity, heart rate, and sleep.

Challenges of Developing Multimodal Wearables in Healthcare

Developing wearable devices involves conceptual and technical challenges and requires a focus on the right clinical questions, emphasizing the need for integrating the clinical perspective alongside engineering principles. Building and maintaining a lasting clinician-engineer partnership is rare and requires compromise and open-mindedness due to their differing scientific languages. This is even more prominent when developing multimodal wearables. Usually, the technical and clinical backgrounds required for developing multimodal devices cannot be found in a single engineer or clinician.

The technical shortcomings in both hardware and analytical understanding also pose great challenges in developing multimodal wearables. These challenges, though, are mainly due to the fact that most research has been done on single-mode wearables, and existing multimodal wearables are still in their infancy [31]. A major limitation in developing electroacoustic wearables is the lack of electrodes that can effectively handle both electrical and acoustic signals, as most research has focused solely on electrical electrodes [6,25,56]. Another example is that the physical models of biological tissue, which are used in the development of wearables and medical devices, typically link the physiology to specific physical domains such as electrical, optical, magnetic, or acoustic models. However, these models often fail to integrate these different modes with one another. While AI can help address some of the limitations, particularly in terms of analytical understanding, significant advancements will require a deeper, more comprehensive understanding of the underlying physical principles. Once these become widely understood, multimodal wearables will come to play an essential role in the monitoring of musculoskeletal diseases.

Applications of AI in Musculoskeletal Medicine and Wearables

ML-based AI algorithms are believed to have an integral role in the future of healthcare. ML algorithms can identify complex patterns in data without relying on detailed decision rules created by medical professionals, but they require large amounts of data. The vast medical data needed to train AI models only became available after the 2009 Health Information Technology for Economic and Clinical Health Act, which facilitated the adoption of electronic health records [59]. Traditional AI algorithms have also been heavily used for prediction, screening, and diagnosis [21,22,26,39,43,45,46,49,52]. However, automated medical image diagnosis is believed to be the most promising domain of AI application in musculoskeletal medicine. The development of appropriate AI models is strongly affected by the availability of AI-specific data registries [4,41]. This is why deep learning has also been used to create image-based data registries as opposed to procedural billing codes currently used for data registries [24]. In orthopedic surgery, ML has most commonly been applied in two areas: predictive algorithms for medical management and survival. Many models focused on medical management aim to predict readmissions and discharge placement, which are significant cost drivers in healthcare. For survival prediction, ML algorithms are used to estimate life expectancy, an important factor in orthopedic oncology decision-making.

AI and biomedical devices have a long history, particularly in engineering, where traditional AI algorithms such as neural networks and support vector machines have been used for signal processing for over four decades. However, the current focus in research is on combining multimodal clinical data—such as patient history, notes, images, and medications—with signals from devices and wearables. AI not only aids in signal processing and integrating diverse data types but also plays a crucial role throughout the biomedical device design process, contributing significantly to the development of concepts, prototypes, and final products. Our group’s recent work has demonstrated this concept for acoustic [57], electrical [16,17,23], and even a combination of wearables with extended reality in musculoskeletal medicine [15,61,62].

There are various processes involved in the development of biomedical devices, from initial conception to their application in healthcare (Fig. 1). The process begins as an iterative cycle of identifying the right questions, developing ideas, refining initial concepts, and reviewing existing literature (Fig. 1a). During this phase, the application of large language models (LLMs) holds considerable promise. These models can assist in various stages, from analyzing literature to supporting idea generation, and even contributing to the early stages of design development. Next, prototyping begins, involving a multidisciplinary approach across bioengineering, electronics, software, analytics, and mechanics (Fig. 1b). This phase includes initial prototypes, in vivo testing, and validation of hardware and models. AI models, particularly LLMs, can greatly assist in circuit design, microcontroller programming, selecting components and materials, writing code, and performing statistical analysis on data. The process then moves to initial human subject tests, followed by signal processing (Fig. 1c). AI can greatly aid in identifying patterns within the signals that traditional analytical models may overlook (Fig. 1d). Finally, multimodal AI, which is one of the main active research topics in healthcare, can be used to integrate the wearable-generated data with patient information (Fig. 1d-f).

Fig. 1.

Development and validation of wearable devices from conception to application. (a) Conception, (b) Initial prototyping and validations, (c) Initial human subject tests, (d) processing different signal types, (e) creating multimodal databases, (f) Integrating wearable signals with patient information, that is, EHR, imaging, patient-reported outcomes, lab results, etc.

Wearables in Musculoskeletal Health

Several recent advancements in musculoskeletal wearables are powered mainly by AI for motion tracking based on accelerometers, and electrical, optical, and acoustic sensors.

Motion Tracking and Accelerometers

Wearables using accelerometers are likely the most used wearable type in both wellness and healthcare, especially in orthopedics. Accelerometers measure acceleration by detecting changes in motion or orientation through the displacement of an internal mass, which generates an electrical signal proportional to the applied force. They enable continuous monitoring of patient movements, providing real-time data on parameters such as range of motion, gait patterns, and activity levels. This continuous monitoring allows for personalized care plans and timely interventions, enhancing overall patient outcomes.

For total knee arthroplasty patients, accelerometers have been used to accurately measure knee range of motion, facilitate at-home rehabilitation, and reduce the need for frequent clinical visits [8]. Accelerometer-equipped wearables have been used for the early detection of movement abnormalities associated with various musculoskeletal disorders. For example, in patients with adhesive capsulitis (frozen shoulder), inertial measurement units incorporating accelerometers can quantitatively assess shoulder mobility deficits, which helps with objective assessment of severity and progress assessment [1]. Post-spine surgery, wearable accelerometers have been used for objective progress assessment by measuring metrics such as daily step count, walking speed, and step length, providing insights into recovery and facilitating personalized rehabilitation plans [35]. In sports medicine, wearable accelerometers assess movement patterns and training intensity, aiding in optimizing performance and preventing injuries [40]. Furthermore, incorporating wearable technology into rehabilitation programs has been shown to improve patient outcomes by facilitating remote monitoring and enabling timely adjustments to therapy plans [44].

One of the main problems with using accelerometers in healthcare is that they measure kinematics as opposed to real joint forces or real muscle monitoring, therefore, the data is an indirect measurement and can result in inaccuracies. Modern motion sensors typically integrate 3-axis accelerometers, and sometimes gyroscopes as well, providing up to six degrees of freedom in motion tracking, which can be integrated with AI to analyze motion patterns, enabling precise classification of activities, range of motion, gait patterns, and more. For example, ML models have been used to predict trip-related falls in older adults based on gait characteristics observed during regular walking. These models, which were trained on 20 selected gait features derived from a gait lab, achieved up to 94% prediction accuracy [54]. Regardless of the shortcomings because of their cost, low weight, size, and easy integration with other devices, they can be extremely useful in healthcare, especially for progress assessment following treatment.

Electrical Wearables and Devices

Electrical impedance spectroscopy (EIS) [61], electrical impedance tomography (EIT) [62], electrical impedance myography (EIM) [34], and electromyography (EMG) are prominent electric-based sensing modalities that provide critical insights into tissue health, muscle activity, and body movement [53]. EIT and EIS originate from a broader field of technologies with wide-ranging applications, including non-biomedical areas such as material composition analysis and biomedical fields like pulmonary monitoring and oncology [28,30]. Bioelectrical and bioimpedance measurements have long been central to biomedical applications, with the musculoskeletal system being a key area driving the advancement of wearable technologies [16]. These techniques are used in various areas such as the diagnosis of neuromuscular disorders, muscle fatigue monitoring, and gesture recognition [7,53].

While some wearable devices that are employing these technologies, such as X-trode’s Smart Skin, have achieved regulatory approval and market presence, many prototypes remain in experimental stages due to challenges in data interpretation, signal variability, and regulatory compliance [3,47]. The extension devices featuring EIT and EIS sensing modalities into wearable technologies for musculoskeletal assessment introduce unique challenges in addition to the general limitations inherent to these techniques. Both EIS and EIT techniques are challenged by factors such as sensitivity to electrode placement and contact impedance [30], high signal-to-noise ratio sensitivity, real-time data acquisition and processing demands, and limitations in data interpretation and physical modeling—particularly in complex materials compositions and geometries (such as that of biological tissues) which exhibit frequency-dependent behavior [14,51,61,63]. These challenges are amplified in dynamic settings such as musculoskeletal monitoring, where motion artifacts and fluctuating electrode-tissue interfaces can generate additional complexity.

AI-driven techniques can be transformative in overcoming these types of challenges by offering robust tools for data processing, noise reduction, pattern recognition, and real-time reconstruction [12,13,37,58]. In addition, AI-driven signal processing can aid in decoding and modeling complex relationships between tissue impedance and underlying physiological states and compositional characteristics, enabling applications beyond lung/pulmonary imaging—such as detecting localized edema, tracking bone healing progress, bone quality, monitoring muscular hydration, and dynamic muscle monitoring [14,33]. For example, in the EIT technique, conductivity-based tomographic maps of electrical properties are reconstructed from multi-electrode readings (typically 8–32 electrodes) by solving an inherently ill-posed problem. Deep learning methods have been shown to significantly enhance spatial resolution, reduce noise, and enable the integration of EIT maps with other imaging modalities, such as CT and ultrasound [59].

One example of combining AI with EIT wearables, MuscleRehab, a virtual reality rehabilitation system, employs optical tracking and EIT to visualize users’ muscle engagement on a virtual avatar, aiming to improve exercise accuracy during unsupervised rehabilitation [61]. When enhanced with ML, the system can both clarify conductivity images and recognize specific movements; in one example, a model applied to EIT data from a forearm cuff achieved 97.5% accuracy in identifying six distinct hand gestures [62]. This integration further supports accurate movement execution and improves therapist evaluations post-rehabilitation.

EMG has also greatly benefited from ML advancements. Raw EMG signals are inherently noisy and vary based on electrode placement, skin conditions, and subject-specific anatomy [32]. ML techniques such as Support Vectors, Random Forest, and Long Short-Term Memory (LSTMs) have been used to enhance the classification capabilities of EMG signals for prosthetic control, movement recognition, and diagnosing neuromuscular disorders. AI-powered wearable EMG systems now enable real-time, highly accurate hand gestures, such as rock, paper, scissors, left, right, and thumbs-up, recognition with classification accuracies exceeding 90% in some cases [62]. These enhanced EMG systems are also finding applications in rehabilitation robotics, athletic performance monitoring, and human-machine interaction systems [60]. AI is significantly changing the landscape of electrical wearables, and we can expect AI-powered electrical wearables to very soon enter the daily practice of musculoskeletal medicine.

Optical Wearables and Devices

Optical sensors transform light into electrical signals, capturing measurements by detecting changes in light patterns. Photoplethysmography (PPG) is the most widely used optical sensing technique in wearable health monitoring due to its simplicity and noninvasive assessment of vascular function. While this review focuses on PPG, a wide range of alternatives exist, including fiber-Bragg grating, Fabry-Perot, silica fiber, side-polished fiber, and micro-nano fiber sensors, as well as interferometers, radiation sensors, and more. These technologies are often integrated into more specialized applications, such as flexible substrates for smart textiles, contact lenses, and even tattoos [12,37,58]. PPG devices typically consist of a light source and photodetector, with light absorbed by tissues—primarily blood, melanin, fat, and water—and reflected or transmitted light measured at the surface. Multi-wavelength setups distinguish blood-oxygen saturation (SpO₂) and vascular function by exploiting differences in absorption spectra of hemoglobin and deoxyhemoglobin [13]. Their high accuracy, low sensitivity drift, immunity to electromagnetic interference, and potential for miniaturization make them particularly well-suited for wearable applications, enabling sleek product designs without compromising performance [51]. Despite challenges such as tissue absorption, light scattering, and motion artifacts, PPG’s noninvasive design and ease of manufacturing have made it popular in wearables, including smartwatches and fingertip oximeters, offering real-time SpO₂, heart rate, and pulse rate variability data.

The development of “hearables” creatively combats wearable limitations by applying PPG to the ear, where higher vascular density and proximity to the carotid artery improve signal quality. Notable in-ear products include the Cosinuss One, designed exclusively for fitness tracking, and the Amazfit PowerBuds Pro, which combines exercise HR monitoring with standard earbud functionality. Interwoven optical fiber probes for receiving and emitting light additionally apply PPG techniques in textile sleeves at accuracies comparable to clinically approved pulse oximeters. Smart garments such as Hexoskin’s Astroskin shirt use these techniques combined with AI to provide quality-of-life and diagnostic assessments.

Moreover, advancements in AI have opened new avenues for addressing the limitations of PPG without necessitating hardware modifications. A significant challenge in PPG signal acquisition is motion artifacts (MAs) caused by body movements, which are the primary source of waveform noise. Traditional approaches to mitigating MAs often rely on additional sensors, such as accelerometers, to detect and compensate for noise, showing up to 94.7% reduction in MA removal [5,18,32]. These approaches can introduce drawbacks from increased power consumption, device complexity, and cost. AI-driven solutions have demonstrated the potential to address these challenges more efficiently. Beyond signal cleaning, AI has helped open new avenues of research using PPG. New analyses as a result of AI involvement in PPG signal processing have been used to assess vascular age, arterial stiffness, pulse wave velocity, respiratory rate, mental stress, blood glucose, and atrial fibrillation with growing accuracy [2,10,11,48]. Because of their versatility and compatibility with AI and other sensor types, PPGs can be extremely useful in the future development of biomedical devices, especially within the context of multimodal devices.

Acoustic Wearables and Devices

Acoustic wearables can be classified into two categories: therapeutic and diagnostic/monitoring. The therapeutic effects of acoustic waves have been extensively studied, with methods such as Shockwave Therapy, Low-Intensity Pulsed Ultrasound (LIPUS), and High-Intensity Focused Ultrasound (HIFU) commonly used for various treatments [33,34]. While acoustic waves have been researched for bone healing [53], tendon repair [7], tissue regeneration [47], or damaging undesired tissues [3], there are few wearable devices designed for these purposes, apart from some notable studies. For example, Lewis and colleagues developed a self-applied, battery-operated LIPUS wearable for chronic myofascial pain reduction. This device, featuring a radiofrequency (RF) ultrasound driver and an ultralow electrical impedance ultrasound transducer, was worn by 30 patients, with the active group showing pain reduction [28]. In another study, Best and colleagues evaluated the Sam 2.0 (ZetrOZ, Inc.), a commercially available FDA-approved LIPUS wearable, which demonstrated pain reduction and strength improvement in 25 patients with elbow or Achilles tendinopathy [7]. In addition, Liu and colleagues showed improved fracture healing in mice [30], and Zou and the team demonstrated that focused acoustic waves can reduce tumor size by generating toxic reactive oxygen species [63].

Diagnostic acoustics can be categorized into imaging and non-imaging devices, with imaging devices seldom utilized as wearables; consequently, this section primarily addresses non-imaging acoustic devices. The most basic form of non-imaging acoustic monitoring involves listening to sounds produced by the body, known as acoustic emission (AE). These sounds provide valuable information about the tissue or organ generating the acoustic waves and the structures through which the waves pass. Sensors designed to detect these phenomena are called acoustic physiological signal sensors, and many are suitable for wearable development. In musculoskeletal medicine, AE has been explored for applications such as monitoring implant health and assessing osteoarthritis (OA) and joint health. A notable example of utilizing AE for diagnostic purposes is the study by Nevalainen et al. [37], which examined the relationship between AE and kinematic instability (KI) in OA knee joints. AE and KI data were recorded using a custom-designed prototype, resembling a knee orthosis with integrated acoustic sensors. The study, which involved 99 human subjects, demonstrated the potential of combining both acoustic and kinematic modalities for detecting knee OA [38]. FitzPatrick et al used AE for the assessment of implant wear and stability in total hip replacement patients [38]. The device consists of four passive ultrasonic sensors arranged in a flexible array and placed against the hip region. The device was validated in vivo on 90 patients and showed promise as a potential diagnostic [14].

One of the biggest challenges in AE is the low signal-to-noise ratio and the difficulty in interpreting signals, as the inverse problem can have multiple solutions, with different conditions producing similar AE signals. This presents a significant barrier to making accurate and reliable decisions based on AE data. Active devices, another type of non-imaging acoustic device, offer a potential solution to this challenge by applying controlled acoustic signals to the body and analyzing the tissue’s response. Our group was one of the pioneers in introducing this concept and demonstrating its potential in musculoskeletal medicine. In these wearables, controlled acoustic signals are applied to the body, and the tissue’s response to the signal is analyzed. In 2 recent studies, we showed that this method can (1) detect internal defects in cortical bone, such as interfacial implant failures, and (2) monitor tendon health. We also demonstrated how AI can significantly reduce the measurement and computation time [57]. Devices employing this approach assess the structure, function, and dynamics of tissues by analyzing the relationship between acoustic wave propagation patterns and the mechanical, geometrical, and biological properties of the tissue or organ. By leveraging ultrasound’s established diagnostic value and the possibilities of AI integration for enhanced real-time health monitoring, wearable medical devices with acoustic sensors have strong potential in the wearables market.

Future Directions and Conclusion

Wearables are becoming an integral part of healthcare, yet most studies on healthcare wearables have focused on single-mode devices. We believe that the multifaceted nature of disease diagnosis requires the development of multimodal wearables. Multimodal wearables can be categorized into three types: (1) different modes (eg, electrical or acoustic) operate independently, with their outcomes investigated separately; (2) modes do not interfere with each other’s measurements but inform each other, such as an EIT system guiding an acoustic stimulation device to focus on a specific area; and (3) modes influence each other’s measurements, ie, a wearable where acoustic stimulation alters the tissue’s electrical or optical properties. Most current work on multimodal wearables falls into the first category due to several fundamental and technical gaps in their development:

• Physics-based models: The main hurdle is the lack of physics-based tissue models that define tissue behavior under different excitation modes, requiring collaboration between clinicians, biologists, and physicists.

• Hardware: Advancements in microcontrollers, embedded AI, and cloud computing have largely addressed computation challenges at small scales, not hindering new device development. However, for multimodal wearables, signal delivery and sensing are critical. For instance, designing an electroacoustic wearable ideally requires using similar electrodes for both electrical and acoustic modes, yet this has not been explored. In addition, continuous monitoring during activity introduces noise across different excitation modes, presenting an optimization challenge that requires collaboration among wearable, biomedical, mechanical, and electrical engineers.

• Physiology: The connection between disease physiology and tissue behavior under different excitation modes needs study, requiring collaboration between clinicians and applied physicists or engineers.

• Connection with EHR: To enable multimodal analysis, the data from wearables needs to be recorded in the EHR to enable large studies.

Finally, we believe healthcare wearables, particularly multimodal ones, offer immense potential for the scientific and clinical community. With advancements in AI and hardware, numerous opportunities exist in this field to significantly impact healthcare. We envision low-cost multimodal wearables becoming an integral part of healthcare, enabling objective telemedicine and the quantification of patient conditions, including functional exams.