Advances in Wearable Bioimaging

ABSTRACT

The development of wearable bioimaging platforms has accelerated in the era of precision medicine, addressing the limitations of conventional, facility‐based systems, which can be costly and difficult to access. In this review, we first outline the operating principles of major imaging modalities, with an emphasis on target tissues and factors that contribute to image quality. Next, we survey advances in miniaturizing these modalities into wearable devices, from electrical impedance tomography belts to flexible ultrasound and photoacoustic patches, highlighting materials selection and fabrication strategies that enhance image quality, skin adhesion, and functional integration. Lastly, we discuss outstanding technical and translational challenges that must be addressed to realize the full potential of these platforms, including safety and long‐term use. With sustained research and responsible commercialization, these wearable devices could reshape continuous physiological monitoring and enable more proactive, preventive health management.

Wearable bioimaging technologies are reshaping medical diagnostics by enabling continuous, noninvasive monitoring outside traditional clinical settings. This review summarizes recent advances in wearable imaging platforms, including electrical impedance tomography, ultrasound, functional near‐infrared spectroscopy, and photoacoustic imaging. It also highlights materials strategies that enable flexible, skin‐conforming devices and discusses key challenges related to power, stability, safety, and clinical translation.

1. Introduction

Wearable bioelectronics have emerged as a transformative platform capable of shifting the current reactive, disease‐centric healthcare paradigm toward a personalized and proactive healthcare model emphasizing continuous physiological monitoring, early disease detection, preventive intervention, and long‐term health promotion [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]. The development of affordable and wearable platforms for continuous deep‐tissue imaging remains a critical challenge in personalized healthcare. Standard‐of‐care imaging systems are often costly and available only in specialized clinical settings, reducing access for low‐resource communities [12]. Many of these systems also require invasive procedures and pose risks of immunological responses. For instance, X‐ray and computed tomography (CT) are ionizing radiation techniques that often require contrast agents to improve angiographic visualization [12, 13, 14]. Intravascular ultrasound (IVUS) is one of the few methods available for detecting atherosclerosis in deep blood vessels, but it requires a lengthy accompanying procedure [15]. Magnetic resonance imaging (MRI) is considered the gold standard for comprehensive imaging of major organ systems, including the brain, spinal cord, and myocardium [16, 17]. However, it provides limited tissue specificity and requires costly maintenance [18, 19, 20].

To address these limitations, advances in materials engineering have enabled the miniaturization and integration of imaging systems into low‐cost wearable platforms, while preserving the fundamental operating principles of conventional imaging modalities [21]. In addition to their accessibility, wearable imagers are often considered safe, primarily relying on low‐risk, noninvasive modalities such as near‐infrared (NIR) spectroscopy, ultrasound, and photoacoustic tomography (PAT) [22]. These platforms can be integrated into daily life, enabling continuous tissue monitoring, facilitating early disease detection, and promoting timely interventions to improve patient outcomes [23, 24]. Thus, wearable imagers may facilitate the shift from a traditional treatment‐centered healthcare system to one that is focused on disease prevention [25, 26, 27, 28, 29, 30, 31, 32, 33, 34].

In this review, we discuss the development and working principles of conventional imaging modalities, considering their trade‐offs in use cases and image quality. We then examine recent efforts in downscaling these modalities for use on wearable devices that can be administered outside the clinic. These devices span a wide range of imaging depths, from superficial measurements to visualization of deep tissues several centimeters beneath the skin, including major blood vessels, myocardium, lungs, and abdominal organs [22]. We conclude this review by identifying key challenges that need to be addressed to realize the full potential of wearable imaging.

2. Survey of Biomedical Imaging Modalities for Wearable Systems

Compared with conventional clinical imaging systems, wearable biomedical imaging platforms are subject to substantially stricter design constraints [22]. They must be noninvasive, safe for repeated or continuous use, and portable, which collectively limits the range of imaging modalities that can realistically be adapted for wearable use. Accordingly, this section focuses on low‐risk modalities, such as electrical impedance tomography (EIT), ultrasound, optical imaging, and related hybrid approaches that underpin many current wearable platforms. Figure 1 places these modalities in a historical context, illustrating the progression from foundational work largely in the 20th century to modern wearables, including imaging patches and self‐powered wearable belts [35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46]. Building on this perspective, Table 1 provides a comparative overview of these modalities based on their typical performance in conventional systems, serving as contextual reference points for wearable implementations. The operating principles of these systems are discussed in detail in the following subsections.

FIGURE 1.

A timeline of key milestones in noninvasive biomedical imaging modalities that are compatible with wearable implementation. Early developments, including ultrasound (1950s), electrical impedance tomography (1980s), near‐infrared (NIR) spectroscopy (1980s), and photoacoustic tomography (1990s), established the physical principles for safe, functional tissue imaging. These foundations have enabled the recent transition toward wearable and conformable systems, including fNIRS headcaps, ultrasound patches, photoacoustic imaging patches, and EIT‐based electronic skin. These advances illustrate the evolution from benchtop imaging systems to continuous, wearable platforms for physiological monitoring and preventative healthcare. Reproduced with permission from Kazazian et al. (2024) [43], Schmidt et al. (2024) [39], Gao et al. (2022) [44], Lin et al. (2024) [45], and Kim et al. (2024) [36]. Copyright Proceedings of the National Academy of Sciences, Springer Nature, American Association for the Advancement of Science.

TABLE 1.

Comparison of biomedical imaging modalities.

	Electrical impedance tomography a)	High‐resolution optical imaging	Diffuse optical imaging	Ultrasound	Photoacoustic tomography
Basic principle	Electrical impedance	Light	Light	Acoustics	Light‐induced acoustics
Anatomical targets	Large, electrically‐conductive tissues (i.e., lungs, muscle, liver, breast)	Superficial tissues (i.e., skin, retina, cornea) Intravascular imaging also possible with minimally invasive procedures	Large tissues with hemodynamic contrast (i.e., cerebral cortex, muscle, breast)	Soft tissues with clear organ boundaries (i.e., myocardium, large vessels)	Optical absorbers (i.e., blood vessels, pigmented structures)
Resolution b)	1–10 mm	1–20 µm	1–10 mm	50–500 µm	25–300 µm
Clinical information	Functional	Anatomical and functional	Functional	Anatomical and functional	Anatomical and functional
Maximum penetration depth	∼30 cm (Ideal for deep‐tissue imaging)	∼2 mm	∼5 cm	∼20 cm (Ideal for deep‐tissue imaging)	∼5 cm

^a) Values shown represent the typical capabilities of conventional systems and serve as a reference for what may be achievable with wearable imaging platforms. In practice, constraints such as power, geometry, and target organ can limit wearable performance.

^b) Tunable based on design parameters, such as electrode geometry, reconstruction methods, and operating frequency.

^*)The information in this table was compiled from representative values reported in refs. [43, 47, 48, 62, 63, 72, 102, 274, 275, 276].

2.1. Electrical Impedance Tomography

EIT is a medical imaging technique that reconstructs tissue based on variations in electrical conductivity [47]. A wearable device, such as a belt or bracelet, measures impedance by passing low‐amplitude alternating current (AC) through a range of frequencies and electrode combinations placed around the target tissue (Figure 2a) [48]. The data is fitted to a physiologically equivalent circuit, such as the Randles circuit (Figure 2b), which models the tissue resistance (R_t ) and the ion diffusion at the electrode‐tissue interface, represented by the charge transfer resistance (R_ct ), Warburg element (W), and constant phase element (CPE) [49].

FIGURE 2.

Working mechanisms and properties of conventional imaging modalities. (a) EIT consists of a series of electrodes, placed circumferentially around a heterogeneous tissue, in which a low‐amplitude current is injected across pairs to measure the differences in conductivities. Reproduced with permission from Lin et al. (2022) [48]. Copyright Springer Nature. (b) The equivalent circuit model consists of a three‐element mesh to model the parasitic impedance at the electrode‐tissue interface (R_ct , W, CPE), as well as the true tissue resistance (R_t ). (c) Scanning electron microscopy shows a significant increase in surface area following the oxidation of CNT. Scale bars: 200 nm. Reproduced with permission from Hao et al. (2021) [62]. Copyright American Chemical Society. (d) A representative EIT‐rendered image of the lungs. Reproduced with permission from Schullcke et al. (2016) [64]. Copyright Springer Nature. (e) A standard ultrasound probe transmitting and recording high‐frequency sound waves as they produce echoes after encountering heterogeneous tissues. Reproduced with permission from Lin et al. (2022) [48]. Copyright Springer Nature. (f) The ultrasound transducer contains piezoelectric crystals, such as PZT, which produce sound waves upon electrical stimulation from the central processing unit. The same piezoelectric crystals generate a reverse current when reflected sound waves induce small mechanical deformations. (g) A three‐dimensional arrangement of piezoelectric composites contained within a polymer matrix. Reproduced with permission from Cyprien Rusu. (h) An ultrasound‐rendered image of the myocardium. Reproduced with permission from Price et al. (2017) [73]. Copyright Springer Nature. (i) A photoacoustic tomography system consists of a laser source and an ultrasound transducer. Reproduced with permission from Kratkiewicz et al. (2019) [99]. Copyright Multidisciplinary Digital Publishing Institute. (j) A schematic of a typical photoacoustic setup. A laser is converted to sound waves upon absorption by the sample. Reproduced with permission from Lin and Wang (2022) [100]. Copyright Springer Nature. (k) Tissue penetration depth of photoacoustic tomography, as compared to traditional photonic modalities. (l) A photoacoustic tomography‐rendered image of the hand, showing active blood vessels. Scale bar: 5 mm. Reproduced with permission from Li and Wang (2021) [103]. Copyright American Association for the Advancement of Science.

A potentiostat measures EIT signals, which contain both phase (ϕ) and magnitude (Z) information [47]. To reconstruct the spatial conductivity map and improve resolution, several algorithms have been developed, including the open‐source program EIDORS [50]. These algorithms can be classified as either stochastic or deterministic. Stochastic algorithms use random statistical variables to reduce the relative error when solving the ill‐posed inverse conductivity problem [48]. In EIT, internal conductivity σ must be inferred from boundary voltage measurements V _meas through a nonlinear forward operator of the form F(σ)  [50]. Because different conductivity distributions can produce nearly identical boundary measurements, and because measurement noise can cause large fluctuations in the reconstructed solution, the inverse mapping, σ = F ⁻¹(V _meas), is not stable without additional constraints [51]. This sensitivity to noise and non‐uniqueness renders direct inversion impractical, necessitating regularized optimization strategies. Genetic algorithms approach this problem by iteratively evolving a population of candidate conductivity distributions to minimize an objective function of the form,

$$\underset{\sigma }{\min }\sum _i{\left[V_{\mathrm{meas},i}-F{\left(\sigma \right)}_i\right]}^2$$ (1)

where candidate solutions are refined over successive generations through selection, crossover, and mutation until a predefined convergence criterion is satisfied [52]. On the contrary, deterministic algorithms solve the same optimization problem using fixed update rules derived from the local sensitivity of the measurements to changes in conductivity [50]. For example, the Gauss‐Newton and Levenberg‐Marquardt methods linearize the forward model around the current conductivity by computing the Jacobian matrix J = ∂F/∂σ, which quantifies how small changes in conductivity affect the boundary voltages [51, 53]. The conductivity distribution is then iteratively updated according to

$${\sigma }_{k+1}={\sigma }_k+\mathrm{\Delta }\sigma$$ (2)

where Δσ is obtained by solving a regularized linear system involving J^TJ and the voltage residual vector [54]. Because these updates follow a deterministic sequence with no random initialization or mutation steps, the reconstruction is fully reproducible for a given dataset and starting condition. Nonlinear deterministic algorithms follow a similar framework, though they do not rely on a single linearization of the forward model [48, 55]. Instead, the Jacobian and forward solution are recomputed at each iteration, allowing the algorithm to capture the nonlinear relationship between conductivity and boundary voltages more precisely. This improves the spatial resolution of the reconstructed data, but at the expense of increased computation time.

The signal‐to‐noise ratio (SNR) of the EIT rendering is highly dependent on minimizing the interfacial parasitic impedance, which can be approximated as

$$Z\approx \frac{d}{j\omega \epsilon A}$$ (3)

where ω is the frequency, ε is the dielectric permittivity, d is the electrode‐tissue separation, and A is the effective interfacial area [56, 57, 58]. To reduce this value, previous studies have implemented conductive polymers or nanomaterials with high surface roughness as electrode coatings to increase the interfacial capacitance required for charge transfer [59, 60, 61]. These coatings enlarge the effective electrode‐tissue interfacial area, thereby lowering parasitic impedance, improving coupling/SNR, and reducing drive voltage and power. For example, a hydrogel‐based electrode coated with poly(3,4‐ethylenedioxythiophene): polystyrene sulfonate (PEDOT:PSS) for pacemaker leads reduced impedance in the low‐frequency regime by up to two orders of magnitude in porcine heart tests [61]. In another study, a 16‐electrode EIT system using oxidized carbon‐nanotube (CNT) films detected indentations as small as 1 mm in a 150 × 150 mm² composite carbon‐fiber panel, an improvement over the ∼10 mm spatial resolution typical of conventional platforms with a comparable field of view [62, 63]. Scanning electron microscopy (SEM) further confirms the increase in surface area resulting from CNT oxidation (Figure 2c).

EIT achieves tissue penetration depths up to ∼15 cm, among the highest of the wearable imaging modalities surveyed in this review [48]. However, its spatial resolution is restricted by the number of electrodes and the computational expense of reconstruction. For this reason, EIT is typically applied to monitor functional changes, such as lung perfusion, and to infer bulk tissue properties such as atherosclerotic plaque composition, rather than for anatomical imaging. A representative EIT‐reconstructed image from a patient with emphysema is shown in Figure 2d, demonstrating higher conductivities in the ventral lung regions [64].

2.2. Ultrasound Imaging

Ultrasound imaging is one of the earliest and most widely adopted modalities compatible with wearable use. It operates by transmitting acoustic waves into tissue and detecting echoes generated by impedance mismatches between anatomical structures [65]. These waves are generated from a piezoelectric transducer, commonly lead zirconate titanate (PZT), in conjunction with impedance‐matching and backing layers (Figure 2e–g) [31, 66, 67]. A pulser‐receiver applies short, high‐voltage bursts to the piezoelectric element, causing it to deform and radiate sound; returning echoes stress the crystal and are converted back to voltage by the same piezoelectric effect [68]. This process generates sound waves that propagate through heterogeneous tissues with different acoustic impedances Z_i [68]. Acoustic impedance is a physical property that depends on the medium's density and speed of sound. To minimize reflection losses, the acoustic impedance of the matching layer should correspond to the geometric mean between that of the piezoelectric transducer and biological tissue [69]. When an incident wave encounters an interface between two distinct acoustic impedances Z ₁ and Z ₂, it produces a wave that travels back toward the transducer [68]. The relative intensity of this reflected signal is given by

$$R=\frac{{\left(Z_2-Z_1\right)}^2}{{\left(Z_2+Z_1\right)}^2}$$ (4)

The performance of piezoelectric transducers is also governed by the electromechanical coupling coefficient (k), which quantifies the efficiency of energy conversion between electrical and mechanical domains [70]. Higher coupling coefficients improve transmission efficiency and sensitivity, directly enhancing SNR and imaging depth.

One of the key advantages of ultrasound is that the same transducer can be used to both transmit and detect sound waves. Moreover, the number of cycles (n_cyc ) and the transmitting frequency (f_t ) can be tuned to balance penetration depth, spatial resolution, and SNR [71]. Fewer cycles improve axial resolution by increasing acoustic bandwidth, but this is accompanied by a reduction in SNR. Higher frequencies also improve spatial resolution, but they experience greater tissue attenuation, which limits penetration depth [72]. As a result, ultrasound imaging of deep organs, such as the myocardium, is often performed at frequencies below 5 MHz. Figure 2h shows a representative B‐mode echocardiogram acquired with a low‐frequency clinical cardiac transducer, which resolves the cardiac chambers [73]. Ultrasound is also routinely used to image the liver [74], kidneys [75], pancreas [76], and major abdominal vasculatures [77], achieving penetration depths of several centimeters while maintaining clinically useful spatial resolution.

2.3. Optical Imaging

Optical imaging relies on the propagation and detection of light as it interacts with tissue, undergoing reflection, refraction, scattering, and absorption [78]. Compared to ultrasound, it offers unmatched spatial resolutions, often on the order of micrometers, enough to resolve single cells and even certain organelles [72, 79]. A typical system comprises a light source, a photodetector, and several intermediate components, such as lenses, dichroic filters, and diffusers [80]. Established modalities include optical coherence tomography (OCT), confocal microscopy, and two‐photon microscopy [81, 82, 83, 84, 85, 86]. OCT relies on low‐coherence interferometry, a system in which incident light is split into two optical pathways that share a common optical path length [87]. The so‐called “sample arm” contains the biological tissue to be imaged, whereas the “reference arm” is often a set of mirrors to redirect the light back to the detector. Reflected signals from both arms are recombined to generate an interference pattern [88]. This pattern is Fourier‐transformed to generate cross‐sectional images of the tissue structure, known as B‐scans. Confocal microscopy, on the other hand, improves optical sectioning and contrast by placing a pinhole at the conjugate image plane to reject out‐of‐focus light [83]. Lastly, two‐photon microscopy utilizes femtosecond excitation and nonlinear absorption to confine excitation to the focal volume, thereby improving SNR and reducing photodamage [84, 89].

Although the spatial resolution of optical imaging far exceeds that of other imaging modalities, its use is largely confined to superficial tissues, such as the epidermis, or to tissues behind optically transparent media, such as the retina. Fundamentally, these depth limitations arise from the combined effects of diffusion and diffraction in biological tissue [72]. Here, “diffusion” refers to multiple scattering experienced by light as it propagates through tissue. This effect can be described by Beer‐Lambert Law, as shown in Equation (5), where I is the intensity of light, μ is an effective attenuation coefficient combining absorption and scattering, and x is the depth.

$$I\left(x\right)=I_0\exp \left(-\mu x\right)$$ (5)

Diffusion restricts many applications of optical imaging to the ballistic regime, the depth range over which minimally scattered photons dominate and can be reliably back‐projected to their origin. For non‐invasive applications, this depth is typically less than 1–2 mm from the outer epidermal surface [90]. Diffraction, on the other hand, is the principal constraint for cellular‐resolution microscopy, because it sets the minimum resolvable separation between two points. Equation (6) shows that diffraction is a physical limit that depends on the wavelength (λ), refractive index (n), and the observed angle (α) [91].

$$\mathrm{\Delta }x=\frac{\lambda }{2n\sin \alpha }$$ (6)

These constraints contribute to the relative scarcity of optical‐based wearable imaging platforms compared with other modalities, such as ultrasound and photoacoustic tomography. However, deep‐tissue applications remain feasible by operating in the NIR spectral window, which reduces absorption and enables photons to propagate through several centimeters of biological tissue [92]. At this depth, beyond the ballistic regime, spatial resolution is severely degraded by scattering, but the detected signals remain sensitive to endogenous chromophores such as hemoglobin [93]. This principle underpins diffuse optical modalities, such as functional near‐infrared spectroscopy, that probe tissue hemodynamics and metabolism without relying on diffraction‐limited imaging [94].

2.4. Photoacoustic Tomography

Photoacoustic tomography (PAT) is an emerging wearable imaging technique that couples optical excitation with ultrasonic detection, providing ultrasound‐like penetration while preserving optical absorption contrast [41]. PAT arises from the photoacoustic effect: short optical pulses absorbed by tissue chromophores produce a transient temperature rise and thermoelastic expansion, generating broadband ultrasound, a phenomenon first reported by Alexander Graham Bell in the late 19th century (historically in solids and gases, now extended to soft tissues) [95]. Mathematically, this is governed by the photoacoustic wave equation,

$$\left({\nabla }^2-\frac{1}{v_s^2}\frac{{\partial }^2}{\partial t^2}\right)p\left(r,t\right)=-\frac{\beta }{\kappa v_s^2}\frac{{\partial }^{2}T\left(r,t\right)}{\partial t^2}$$ (7)

where p is the induced pressure, v_s is the speed of sound in the medium, κ is the compressibility, T is the temperature function, and β is the isobaric volume expansion coefficient [72]. The amplitude of the generated photoacoustic signal is also determined by the initial pressure rise, which depends on both the tissue's optical absorption properties and the local optical energy delivered [96]. This relationship is commonly expressed as

$$p_0\left(r\right)=\mathrm{\Gamma }{\mu }_a\left(r\right)\mathrm{\Phi }\left(r\right)$$ (8)

where Γ is the Grüneisen parameter, a dimensionless factor that describes the conversion efficiency from thermal energy deposition to acoustic pressure, μ _a is the optical absorption coefficient, and Φ is the optical fluence. Conceptually, optical fluence is the amount of optical energy deposited per unit area within the tissue, accounting for both absorption and multiple scattering [97]. Unlike incident irradiance at the surface, fluence reflects how light is redistributed internally as it propagates through a highly scattering medium. It decays exponentially with depth, following the Beer‐Lambert Law discussed in Section 2.3. Consequently, the effective imaging depth of PAT is constrained to regions where sufficient fluence remains to induce detectable pressure rises, typically on the order of millimeters to a few centimeters, depending on wavelength and tissue optical properties [98].

PAT systems comprise a pulsed optical source to excite tissue chromophores and an ultrasound transducer, often an array, to record the induced acoustic waves (Figure 2i) [99]. A central beam combiner directs the transmitting light toward the sample and the resulting ultrasound waves toward the detector (Figure 2j) [100]. Subsequently, signal‐processing techniques, including bandpass filtering, envelope detection, delay‐and‐sum beamforming, and model‐based iterative reconstruction, are applied to reconstruct the photoacoustic images [101]. Compared to conventional photonic modalities, photoacoustic tomography leverages sound to overcome the optical diffusion limit, which is ∼1–2 mm for most human tissues (Figure 2k) [90]. Moreover, photoacoustic tomography can achieve spatial resolutions of up to 25 µm, roughly half those of ultrasound [72, 102]. A representative photoacoustic tomography‐rendered image of the manus is shown in Figure 2l, where hemoglobin serves as a natural contrast agent to clearly outline the blood vessels [103].

Several portable and wearable photoacoustic imaging systems, including patch‐style prototypes, have been reported [44, 104, 105, 106]. Because acoustic detection relies on wavelengths that are orders of magnitude longer than optical wavelengths, PAT systems are less sensitive to small‐scale optical misalignment and do not require diffraction‐limited optics [98]. In addition, ultrasound propagation in tissue exhibits relatively low scattering compared to light, allowing signals generated at depth to be detected with high spatial resolution using conformable or array‐based transducers [107]. These characteristics reduce the need for bulky optical components and stringent alignment requirements, making PAT well‐suited for miniaturized, mechanically robust systems [44, 72, 108].

3. Wearable Implementations of EIT

Over the past few decades, many variations of EIT have been developed for both clinical settings and everyday use [109]. These devices have applications in assessing lung function, identifying malignant tumors, and detecting strokes, among others [110]. They have also been used as human‐machine interfaces to assist disabled patients with everyday tasks [36]. Furthermore, EIT signals can be automatically mapped to corresponding medical conditions using recent advances in machine learning and neural networks [111, 112, 113]. In the following sections, we present an in‐depth exploration into the clinical applications of wearable EIT platforms, followed by a discussion of the underlying challenges in sensor design.

3.1. Clinical Applications

3.1.1. Abdominal Imaging

One of the most common applications of wearable EIT is to screen for lipid content in abdominal organs [114]. Achieving a timely diagnosis is of the utmost importance for conditions such as nonalcoholic fatty liver disease (NAFLD), which affects more than one‐third of adults in Western countries [115]. The current gold standard for detecting NAFLD is liver biopsy, though this procedure carries risks of bleeding, hypertension, and sampling errors [114, 116]. MRI serves as a noninvasive alternative, but it remains inaccessible to several patient populations, including those with cardiac implants, metallic foreign bodies, and neurostimulation systems [117]. These limitations have motivated the development of EIT belts for assessing hepatic fat at the point of care.

In 2018, Luo et al. developed a theoretical and experimental framework for a 32‐electrode EIT belt capable of quantifying liver fat distribution (Figure 3a) [114]. Before testing, the belt was simulated using finite‐element analysis to assess how liver conductivity changes with lipid accumulation. Imaging was subsequently performed on an ex vivo porcine liver and the fatty liver of a New Zealand white rabbit. A correlation study was also conducted between MRI‐based fat volume fraction (FVF) maps and EIT‐measured conductivities of human volunteers (Figure 3b–d). Opposite trends of comparable magnitude were observed across all three studies: MRI‐based FVF increased by ∼4.6% in low‐ and high‐BMI subjects, and EIT‐measured liver conductivity decreased by ∼4.5%. These results indicate that the EIT belt could serve as an effective tool for detecting NAFLD in patients, offering a more affordable and practical alternative to current gold standards [114].

FIGURE 3.

Wearable EIT and its clinical applications. (a) A 32‐electrode EIT belt positioned around the waist for the diagnosis of NAFLD. (b) An MRI image of a high‐BMI patient (>25 kg/m²) serves as the gold standard for comparison. The darker region represents the liver, which, in this case, is surrounded by fat (white). (c) The EIT‐rendered image shows a conductivity distribution, with regions of low conductivity appearing in red, indicating high fat content. (d) The fat volume fraction is calculated from impedance data, revealing some of the same anatomical features as the MRI‐rendered image. Reproduced with permission from Luo et al. (2018) [114]. Copyright Ivyspring International. (e) A 16‐electrode bracelet was designed to analyze muscular activity, in which the device was placed around a subject's left forearm and connected to a wireless microcontroller. The signals were transmitted to a smartphone for further analysis of hand gestures. (f) An artificial‐reality application comparing the muscular activity within a subject's thighs in a contracted state. (g) The spherical electrodes from EIT‐kit resulted in higher current intake compared to conventional ECG electrodes, given that a conductive gel was applied. Reproduced with permission from Zhu et al. (2021) [118]. Copyright Association for Computing Machinery. (h) A brassiere‐shaped textile consisting of 90 EIT electrodes, arranged across five concentric circles for the screening of breast cancer. (i) Results of a proof‐of‐concept experiment in which a 5 mm tumor phantom was successfully detected using the EIT bra. Reproduced with permission from Hong et al. (2015) [135]. Copyright IEEE. (j) A stretchable sensor was seamlessly printed on a flexible fabric substrate for continuous imaging of the human thorax. (k) This device achieved a noticeable reduction in parasitic contact impedance, compared to conventional electrodes. Reproduced with permission from Jose et al. (2021) [144]. Copyright Institute of Physics.

3.1.2. Musculoskeletal Monitoring

Another application of EIT is to sense musculoskeletal activity in the extremities, such as the hand and forearms. In a recent study, a versatile toolkit comprising an electrode platform, microcontroller, and open‐source software for image reconstruction was developed [118]. This toolkit was used to fabricate a 16‐electrode bracelet, in which the corresponding signals were wirelessly transmitted to a smartphone application for the analysis of hand gestures (Figure 3e) [118, 119, 120, 121, 122]. The EIT band distinguished between six gestures, including a thumbs‐up and a fist, achieving 97.5% accuracy [118]. Another application of this toolkit involved assessing muscle function for physical rehabilitation. Two electrode arrays were fabricated from the toolkit and placed in parallel around a subject's thighs to monitor the quadriceps femoris, a set of four muscle groups that assist with mobility [118, 123]. The signals were processed by a smartphone application, which renders the spatial distribution of muscle strain in a 3D augmented reality environment (Figure 3f). Accordingly, this information can facilitate rapid recovery and prevent further muscle injury.

To evaluate the charge‐transfer efficiency and parasitic contact impedance of the spherical electrodes from the toolkit, a 5 V peak‐to‐peak signal was injected across a pair of electrodes on human skin, with a frequency sweep from 1 kHz to 100 kHz [124]. Amplitude measurements were taken with spherical and commercial ECG electrodes, focusing on the 50 kHz frequency range, which is considered optimal for human skin. At 50 kHz, the spherical electrodes yield the highest measured‐to‐injected amplitude ratio when conductive gel is added, retaining approximately half of the original current (Figure 3g) [118].

3.1.3. Thoracic and Breast Imaging

EIT is also well‐suited to image the thoracic region, where changes in tumor malignancy and tissue composition can produce measurable differences in electrical conductivity [125]. To support long‐term thoracic measurements, EIT systems have increasingly been integrated into textiles, which provide a familiar, conformal interface that maintains stable electrode contact during everyday wear [26, 119, 126, 127, 128, 129, 130, 131, 132, 133, 134]. One example came from the Korea Advanced Institute of Science and Technology, where a set of flexible EIT electrodes was embedded into a brassiere‐shaped textile connected to a smart device for continuous monitoring (Figure 3h) [135]. The impedance contrast between malignant tumors and normal breast tissue has been reported to be several‐fold, making EIT an appealing modality for this application [136]. Using a planar‐fashionable circuit board (F‐PCB), 90 electrodes were patterned across five concentric rings to achieve uniform angular sampling. To reduce parasitic impedance, Ag/AgCl paste was applied to form a conformal, dry adhesive interface that follows the skin contours and improves the electrical contact [135, 137, 138, 139]. A small‐amplitude AC excitation (±8.72 mV, DC bias: 8.99 mV) was applied across electrode pairs with a 100 Hz–100 kHz frequency sweep to acquire data. As a result, this device was shown to detect tumors as small as 5 mm with a sensitivity of 4.9 mΩ. Figure 3i shows the results of a proof‐of‐concept experiment aiming to detect a tumor‐mimicking substrate within an agar phantom.

EIT textiles have also been applied to image pulmonary function. In one study, a 16‐electrode textile belt generated cross‐sectional conductivity maps of the lungs, which were subsequently used to estimate physiological parameters, such as lung volume and respiratory rate [140]. To fabricate the textile belt, the authors used a nylon strap with 16 snap buttons, coupled to silver‐wire cloth electrodes. The increased effective surface area of these electrodes removes the need for conductive gels, which can dry out during extended wear [140, 141]. Moreover, all materials in the system have been tested in previous studies for durability during everyday maintenance, such as washing and ironing [140, 142, 143]. Building on this, a separate study further streamlined fabrication by proposing a fully printed, stretchable thoracic sensor [144]. The electrodes for this textile were printed with a flexible silver ink and over‐coated with polymeric insulation (Figure 3j). A Polydimethylsiloxane (PDMS) top layer was fabricated to reduce motion artifacts and enhance contact with the skin. Overall, the fully printed textile achieved notably lower contact impedances than typical commercial medical electrodes (Figure 3k), with values approximately half an order of magnitude lower across all investigated frequencies.

3.2. Underlying Challenges of Wearable EIT

Although the clinical demonstrations in the previous subsection highlight the versatility of wearable EIT, translating these systems from benchtop settings to everyday use introduces significant engineering and modeling challenges. Unlike benchtop or hospital‐based systems, wearable platforms must operate under continuous motion and tissue deformation. Here, we discuss the challenges inherent in wearable EIT and survey potential solutions implemented in prior studies.

3.2.1. Boundary Uncertainty

The forward operator in EIT explicitly depends on the geometry of the imaging domain [50]. Most reconstruction algorithms assume a fixed boundary shape derived from simplified geometric models, such as circular or elliptical cross‐sections, or from a reference anatomical scan [51]. In wearable systems, however, the boundary is dynamic. Respiratory expansion, abdominal distension, muscle contraction, and postural changes alter the effective shape and volume of the domain over time [145]. Boundary deformation modifies current flow pathways and redistributes equipotential lines, thereby altering the measured boundary voltages independently of changes in internal conductivity. Even small geometric perturbations can produce voltage deviations comparable to those caused by clinically relevant conductivity contrasts. When the assumed computational mesh does not match the true geometry, systematic reconstruction bias may occur, often manifesting as artificial conductivity gradients near the boundary [109].

Several approaches have been proposed to mitigate boundary uncertainty in EIT reconstruction. One class of methods formulates the reconstruction problem to explicitly incorporate shape‐driven difference imaging, in which the inverse problem is expressed in terms of geometric variables that characterize the inclusion boundaries [146]. For example, Liu et al. proposed a Boolean‐operation‐based shape reconstruction framework in which inclusion boundaries are represented explicitly using closed B‐spline curves, and complex geometries are constructed via Boolean unions or intersections of multiple shape primitives [147]. In their approach, the control points of the B‐spline curves serve as the primary design variables, and the conductivity distribution is updated through a Gauss‐Newton scheme that computes the Jacobian via the chain rule with respect to these geometric parameters. This explicit boundary representation reduces the dimensionality of the inverse problem compared to pixel‐based methods and improves robustness to modeling errors, particularly in preserving sharp geometric features under boundary mismatch.

Another strategy integrates auxiliary measurements or dynamic meshes to refine the forward model. In wearable EIT systems, researchers have explored dynamic boundary sensing, such as integrating bend, stretch, or inertial measurement sensors within an electrode belt, to estimate the instantaneous body contour and update the computational mesh during reconstruction, thereby reducing systematic errors arising from geometric mismatch [148]. Finally, data‐driven and hybrid reconstruction methods informed by machine learning have shown emerging promise in improving robustness to geometric uncertainty without full shape measurement. For example, Jeschke et al. trained a conditional generative adversarial network (cGAN) on simulated datasets that incorporated random electrode displacements arising from belt deformation and placement variability [149]. By embedding geometric perturbations directly into the training process, the network learned to maintain reconstruction accuracy under electrode shifts that significantly degraded conventional methods, effectively compensating for model mismatch during inference.

3.2.2. Motion Artifacts

Motion artifacts in wearable EIT arise primarily from electrode displacement and variability in electrode‐tissue contact impedance [11]. Because EIT relies on accurate current injection and voltage measurement at discrete boundary locations, minor shifts in electrode position can significantly alter measured boundary potentials [50]. Mechanical stretching of textile substrates, belt repositioning, and soft‐tissue sliding can modify inter‐electrode spacing and angular orientation, introducing measurement inconsistencies. In addition, fluctuations in contact impedance, driven by changes in pressure distribution, perspiration, hydration, or skin compliance, perturb measured voltages independently of internal conductivity. These effects are particularly pronounced at lower frequencies, where interfacial impedance contributes significantly to the total measured signal [150]. If contact impedance is not explicitly modeled in the forward solution, its variability may be misinterpreted as a change in conductivity.

Mitigation strategies for motion artifacts typically target mechanical stabilization of the electrode interface, correction of faulty channels, and suppression at the signal‐processing level. For example, Hu et al. integrated dry textile electrodes into an elastic belt to maintain conformal contact during breathing and used difference imaging, a process in which conductivity changes are reconstructed relative to a baseline state, to detect poor electrode contact [140]. Upon detection, they introduced two low‐complexity compensation schemes: a voltage‐replace method that substitutes corrupted measurements with adjacent‐electrode data, and a voltage‐shift method that cyclically shifts neighboring voltage sets to reconstruct missing injection patterns [151]. Simulation, phantom, and in vivo lung experiments demonstrated that these methods significantly reduced reconstruction artifacts and restored structural similarity toward baseline images, maintaining shape deformation within ∼5% of well‐contacted conditions and limiting position error to approximately 2 mm, without modifying the underlying reconstruction algorithm.

4. Functional Near‐Infrared Spectroscopy

Functional near‐infrared spectroscopy (fNIRS) has emerged as a leading modality for wearable functional brain monitoring, enabling long‐term measurements that are difficult to obtain with conventional neuroimaging [43]. It assesses functional brain activity by detecting optical contrast induced by changes in hemoglobin oxygenation. In contrast to functional magnetic resonance imaging (fMRI), which is typically restricted to brief, stationary imaging sessions in controlled environments, fNIRS systems can be implemented as lightweight, head‐mounted devices that support continuous measurements during naturalistic behavior [152]. Extensions such as diffuse optical tomography (DOT) further enable three‐dimensional (3D) reconstruction of hemodynamic changes by leveraging overlapping source‐detector measurements [153]. This capability has made wearable fNIRS particularly well‐suited for longitudinal studies of brain function, including monitoring neurodevelopmental trajectories, tracking recovery during neurorehabilitation, and assessing cognitive fatigue over extended periods (Figure 4a) [43]. The ability to acquire functional data across days to weeks outside of dedicated imaging facilities positions fNIRS as a practical complement to conventional neuroimaging for ambulatory applications.

FIGURE 4.

Wearable functional near‐infrared spectroscopy (fNIRS) and applications. (a) Schematic of the fNIRS setup for bedside monitoring, including a whole‐head optode cap, modular electronic architecture, and flexible hexagonal optode units integrating light sources and detectors. Reproduced with permission from Kazazian et al. (2024) [43]. Copyright Proceedings of the National Academy of Sciences. (b) Exploded view of the probe module architecture. Reproduced with permission from Anaya et al. (2023) [154]. Copyright SPIE. (c) Design of a flexible hexagonal high‐density optode unit. (d) Representative cortical hemodynamic measurements, showing spatially resolved channel signals and typical time courses of oxygenated hemoglobin (HbO), deoxygenated hemoglobin (HbR), and total hemoglobin (HbT) concentration changes. Reproduced with permission from Zhao et al. (2021) [164]. Copyright SPIE.

4.1. Continuous‐Wave fNIRS

Continuous‐wave fNIRS (CW‐fNIRS) is the most widely used implementation of fNIRS in wearable brain imaging platforms due to its simplicity, low power consumption, and compatibility with compact hardware. CW‐fNIRS assesses functional brain activity by continuously illuminating tissue with NIR light and detecting intensity changes associated with variations in oxygenated (HbO) and deoxygenated hemoglobin (HbR) [154]. In wearable CW‐fNIRS systems, quantification of functional signals is typically based on the modified Beer‐Lambert law (MBLL), as shown in Equation (9), which relates relative changes in detected light intensity ΔA to changes in hemoglobin concentration Δ[HbO] and Δ[HbR] under the assumption of constant scattering [155].

$$\mathrm{\Delta }A\left(\lambda \right)=L·DPF\left(\lambda \right)\left[{\epsilon }_{HbO}\left(\lambda \right)\mathrm{\Delta }\left[HbO\right]+{\epsilon }_{HbR}\left(\lambda \right)\mathrm{\Delta }\left[HbR\right]\right]$$ (9)

Here, ε _i refers to the molar extinction coefficient of HbO or HbR, L represents the geometric distance between the source and detector, and DPF is the differential pathlength factor, a dimensionless quantity to account for scattering‐induced pathlength elongation. To determine the changes in oxygenated and deoxygenated hemoglobin concentrations, measurements must be taken at two wavelengths (λ₁ and λ₂), yielding a system of equations [156].

The accuracy of MBLL measurements in wearable systems is strongly influenced by source‐detector geometry. Previous studies have shown that approximately 30 mm of separation between the source and detector yields optimal sensitivity for measuring oxygenation in cortical layers [157]. However, in ambulatory settings, head curvature and subject motion can alter optical coupling and sampling depth. To address these challenges, wearable systems have increasingly incorporated conformable interfaces that distribute pressure evenly across the scalp. In one design, a printed circuit board (PCB) was optically isolated with neoprene and fully encased in a silicone layer, with integrated light guides that protruded slightly to maintain consistent optical contact [158]. In another study, Anaya et al. introduced a CW‐fNIRS system, embedded in a conformable silicone substrate with hair‐combing optical ferrules to guide light to the scalp (Figure 4b) [154]. It was shown that improved cap fit, quantified by MRI‐derived optode‐to‐scalp distance, was positively associated with hemodynamic accuracy; poorly fitted regions showed up to 40% reductions in response magnitudes. Another approach is to design a patch to be placed directly on the forehead, rather than as a head cap that covers the entire scalp, where hair can interfere with optical signal transmission. For example, one study developed a dual EEG‐fNIRS platform enclosed in a 3D‐printed housing with integrated optical barrels and long‐pass filters, Ag/AgCl dry electrodes, and a flexible thermoplastic polyurethane fixation belt to ensure close skin contact and mechanical stability in a lightweight wearable format [159]. This packaging strategy enabled low input noise (<0.9 µV_rms), low amplitude distortion (<2%), and effective acquisition of co‐located hemodynamic and electrophysiological responses during cognitive tasks, supporting its suitability for portable, ambulatory monitoring.

4.2. High‐Density fNIRS and DOT

High‐density fNIRS (HD‐fNIRS) is an advanced form of conventional CW‐fNIRS, in which multiple sources and detectors are implemented in dense, overlapping arrays [160]. Rather than relying on a single measurement channel per optode pair, HD‐fNIRS samples the same cortical region through many partially redundant light paths, enabling tomographic reconstruction of spatially resolved hemodynamic changes [161]. This dense sampling improves localization accuracy, reduces sensitivity to superficial physiological signals, and allows depth‐dependent separation of scalp and cortical contributions.

HD‐fNIRS is particularly appealing for wearable applications because it enables improved spatial resolution without altering the fundamental principles of continuous‐wave measurement. However, translating HD‐fNIRS from benchtop systems to fully wearable platforms introduces several design challenges. The increased number of optodes significantly raises system complexity, including channel count, power consumption, data throughput, and thermal load [162]. Dense optode packing also exacerbates mechanical constraints, as maintaining stable optical coupling across many sources and detectors is difficult on a curved, deformable scalp, especially during motion [163]. Recent wearable HD‐fNIRS prototypes illustrate how this can be achieved through modular packaging and optode miniaturization. For example, a mechanically flexible HD‐DOT system weighing only 70 g was developed for preventive neuroimaging in infants, using a modular architecture [164]. The optoelectronics were embedded within rigid PCB “islands” connected by flexible interconnects, eliminating bulky cabling while preserving signal integrity and enabling tight bend radii for scalp conformity (Figure 4c). Additionally, soft silicone encapsulation with optically transparent windows and curved contact surfaces improves optical coupling, distributes pressure, and allows integration into stretchable headgear for stable, wearable operation. Representative measurements show localized right‐hemisphere activation in a 2D array, characterized by increased HbO and a concurrent decrease in HbR, consistent with typical functional hemodynamic responses (Figure 4d). In contrast, another study explored a different design strategy, in which individual source and detector optodes were implemented in thin‐wall 3D‐printed housings aligned with polished acrylic light pipes and secured with dielectric epoxy potting [165]. To preserve SNR in high‐density layouts, electromagnetic interference shielding was provided by brass housings and conductive films. Overall, these examples illustrate the importance of balancing increased channel density with mechanical stability, flexibility, and consistent optical coupling when translating HD‐fNIRS into wearables.

5. Imaging Patches

Flexible imaging patches have recently attracted widespread interest in biomedical imaging, enabling continuous, non‐invasive monitoring of disease progression [44, 72, 108, 166]. Compared with conventional instruments in clinical settings, wearable imaging patches are compact, lightweight, and flexible, improving comfort, convenience, and accessibility. A key materials consideration underlying these designs is the mechanical modulus, which governs the device's ability to conform to soft, curved, and dynamically moving biological tissues [167]. Materials with a modulus comparable to that of skin minimize interfacial gaps and motion‐induced artifacts, thereby stabilizing acoustic coupling in ultrasound and preserving optical alignment in photoacoustic systems. This mechanical matching also enables long‐term applications, such as monitoring disease progression, which can ultimately lead to early detection, timely intervention, and improved outcomes in preventative medicine. In this section, we review the different modalities of flexible imaging patches, focusing on their design, clinical applications, and limitations. Table 2 summarizes the key contributions of each study.

TABLE 2.

Recent developments in wearable imaging patches.

Refs.	Modality	Year	Contributions to wearable imaging
Hu et al. [46]	Ultrasound	2018	Introduction of a skin‐conformal ultrasound imaging patch capable of operation on nonplanar, deformable surfaces
Wang et al. [171]	Ultrasound	2018	Translation of ultrasound patches to physiological sensing in human subjects
Wang et al. [172]	Ultrasound	2021	Integration of doppler‐mode imaging to assess hemodynamic signals up to 14 cm beneath the skin. Adaptive beam steering to focus on regions of interest
Kenny et al. [173]	Ultrasound	2021	Hands‐free application of doppler‐mode imaging to large arteries.
Gao et al. [44]	Photoacoustic Tomography	2022	First photoacoustic‐based imaging patch. Achieved sub‐millimeter resolution for the 3D imaging of hemoglobin.
Wang et al. [167]	Ultrasound	2022	Bioadhesive ultrasound imaging of various internal organs, such as the lungs, the myocardium, and the gastrointestinal tract.
Hu et al. [24]	Ultrasound	2023	Improved electromechanical coupling between the transducer and skin. A deep learning model to extract key cardiovascular parameters, such as stroke volume, cardiac output, and ejection fraction.
Hu et al. [174]	Ultrasound	2023	Quantitative elastography data obtained using imaging algorithms.
Lin et al. [45]	Ultrasound	2024	Wireless and fully integrated with flexible circuit, battery, and probe. Utilized machine learning to automatically track moving targets.

5.1. Ultrasound Patches

Ultrasound is a widely adopted modality for flexible imaging patches, owing in part to its ability to achieve high‐resolution imaging of tissues several centimeters beneath the skin while operating within established clinical safety limits and modest power budgets [168]. Modern approaches to materials engineering have enabled the wearability of ultrasound. Unlike conventional ultrasound probes, which are optimized for planar contact and rely on rigid housings, ultrasound patches are designed to conform to curved and dynamic body surfaces, maintaining stable acoustic coupling during motion and extended wear [46]. This shift from handheld probes to skin‐conforming layouts enables hands‐free operation, improves repeatability, and facilitates access to anatomically complex regions, thereby motivating the rapid development of wearable ultrasonic imaging platforms [169].

One of the major challenges for ultrasound patches is balancing acoustic performance with mechanical compliance. Early work by Hu et al. demonstrated that this trade‐off could be addressed by localizing strain away from acoustically active regions through targeted materials and structural design [46]. In particular, a 10 × 10 array of piezoelectric transducers was embedded in a silicone matrix using an island‐bridge architecture (Figure 5a). The silicone matrix provided high compliance, and copper electrode islands, which mechanically and electrically anchored the transducers, were connected to it via polyimide bridges. Each piezoelectric transducer, spaced at a 2 mm pitch, used a custom 1–3 composite, in which the PZT phase promoted longitudinal vibration, while the epoxy matrix mechanically isolated neighboring elements to suppress lateral crosstalk. Furthermore, an Ag‐epoxy backing layer was incorporated to provide acoustic damping and broaden the effective bandwidth. With this design, the system achieved a spatial resolution of 610 µm across irregular surface geometries.

FIGURE 5.

Wearable ultrasonic patch and applications. (a) Schematic showing the island‐bridge structure of transducers. Reproduced with permission from Hu et al. (2018) [46]. Copyright American Association for the Advancement of Science. (b) Schematic of the ultrasonographic window used in cardiac activity monitoring. (c) Ultrasonic Doppler sensing for central blood flow monitoring. Reproduced with permission from Wang et al. (2021) [172]. Copyright Springer Nature. (d) Placement of a rigid probe to image the carotid artery. (e) Imaging of carotid arteries before and after 30 min of exercise, showing an increase in carotid artery diameter and blood flow rate. Reproduced with permission from Wang et al. (2022) [167]. Copyright American Association for the Advancement of Science. (f) Workflow of using deep learning to train a model to extract features from cardiac ultrasound imaging. Reproduced with permission from Hu et al. (2023) [24]. Copyright Springer Nature. (g) Mapping and surveillance of delayed‐onset muscle soreness in the lateral shoulder joint, comparing B‐mode images and corresponding strain mapping results. The stretchable ultrasonic array is wired for power and data communication. Reproduced with permission from Hu et al. (2023) [174]. Copyright Springer Nature. (h) Fully integrated USoP on the chest for measuring cardiac activity. Reproduced with permission from Lin et al. (2024) [45]. Copyright Springer Nature.

This compliant patch architecture served as a basis for many subsequent studies that aimed to measure specific physiological parameters [46, 170]. In cardiovascular monitoring, Wang et al. adapted the conformal array to continuously track central arterial and venous blood pressure waveforms at the carotid and jugular sites, prioritizing an ultrathin form factor, low contact pressure, and stable recordings for extended wear [171]. To maximize spatial resolution while enabling selective interrogation of deep organs, another work implemented a larger 12 × 12 transducer array capable of beam steering via programmed phase delays across individual elements, thereby electronically controlling the acoustic focus and incidence angle without mechanical repositioning [172]. The frequency was reduced to 2 MHz to enable deeper acoustic propagation. Using this platform, the authors demonstrated two complementary cardiovascular sensing modalities: cardiac tissue Doppler imaging enabled detection of myocardial motion at depths of several centimeters (Figure 5b), while pulsed‐wave Doppler measurements of central blood flow resolved velocity spectra arising from ensemble backscattering by red blood cells (Figure 5c). A complementary emphasis on longitudinal vascular monitoring is illustrated by the hands‐free Doppler patch introduced by Kenny et al., which employs a simplified acoustic architecture designed for continuous cardiovascular assessment rather than for spatially resolved imaging [173]. By using a broad, unfocused beam with a fixed angle to the vessel, the system favors stable alignment over spatial selectivity, enabling reliable beat‐to‐beat tracking of carotid velocity during transient changes in cardiac preload. Together, these studies show how cardiovascular applications shape the architecture of ultrasound patches, from material selection to array complexity.

Another challenge regarding ultrasound patches is achieving stable skin adhesion under motion and long‐term wear without introducing excessive attenuation or discomfort. In contrast to stretchable‐array approaches, a parallel line of work has addressed this challenge by engineering more effective skin‐transducer interfaces. One study introduced a bioadhesive ultrasound (BAUS) platform in which a chitosan‐polyacrylamide hydrogel‐elastomer couplant anchored a rigid piezoelectric array to the skin while preserving the overall structure's flexibility [167]. The hydrogel was encapsulated within a thin polyurethane membrane to ensure high water retention, while the outer bioadhesive layer was functionalized with NHS ester to enable both physical and covalent bonding to the epidermis. This architecture suppressed relative probe‐tissue motion [11], enabling stable flow measurements in the carotid artery and jugular vein (Figure 5d,e) for up to 48 h [167]. Subsequent work further refined this interface by integrating adhesion, electrical interconnection, and mechanical matching into a unified materials system. Hu et al. employed a triblock copolymer styrene‐ethylene‐butylene‐styrene (SEBS) substrate with a Young's modulus comparable to that of skin, reducing interfacial stress concentrations during deformation [24]. Liquid–metal–elastomer composite electrodes derived from eutectic gallium‐indium were introduced not only to provide stretchable, low‐resistance electrical routing, but also to enhance bonding strength between the transducer elements and surrounding polymer layers (∼250 kPa), exceeding that of conventional medical adhesives (∼236 kPa). In addition, embedded electromagnetic shielding reduced motion‐induced noise in the received radiofrequency signals.

A further consideration for ultrasound patches is the development of computational frameworks that can transform continuous, high‐dimensional acoustic data over long‐term wear. In contrast to conventional ultrasound, in which image acquisition and interpretation are episodic, wearable platforms require automated analysis pipelines that operate continuously with minimal human intervention. Recent studies have therefore integrated deep learning and advanced signal‐processing methods to enable robust segmentation, tracking, and parameter estimation directly from wearable ultrasound data. For example, Hu et al. employed a convolutional neural network to automatically segment the left ventricle from continuous cardiac image streams, allowing real‐time reconstruction of ventricular volume and the derivation of stroke volume, cardiac output, and ejection fraction (Figure 5f) [24]. Complementarily, signal‐processing pipelines have enabled the extraction of mechanical tissue properties, such as the elastic moduli of local tissues, from wearable ultrasound data. In elastographic ultrasound patches, sub‐wavelength tissue displacements are estimated from radiofrequency echoes via cross‐correlation‐based motion tracking, followed by regularized strain estimation and inverse‐elasticity solvers to recover quantitative modulus maps at centimeter depths [174]. By decoupling biomechanical inference from direct image interpretation, these approaches extend the functionality of ultrasound patches beyond kinematic imaging toward quantitative assessment of tissue state.

Lastly, the translation of ultrasound patches from laboratory prototypes to practical wearables hinges on full‐system integration, including onboard power, data acquisition, and wireless communication. Many early patch‐based systems relied on external cables for power delivery and signal transmission (Figure 5g), thereby constraining subject mobility and limiting use during daily activities or exercise [174]. Recent work has addressed this bottleneck by developing fully integrated ultrasonic systems‐on‐patch (USoP) that combine stretchable transducer arrays with miniaturized control electronics, batteries, and wireless modules in a soft, skin‐mountable form factor (Figure 5h) [45]. By co‐designing flexible printed circuits, analog front‐end electronics, and wireless data links, these platforms enable on‐board signal conditioning and streaming without tethered connections. Importantly, system‐level integration enables tailoring of sensing performance via selectable operating frequencies and array configurations, thereby balancing penetration depth and resolution across different cardiovascular targets. Coupled with real‐time algorithmic channel selection and autonomous target tracking, such fully integrated patches support continuous monitoring during vigorous motion, as demonstrated during high‐intensity exercise.

5.2. Photoacoustic Patches

Physiological and metabolic processes typically correlate better with biomarkers in deep tissue, as opposed to those near the epidermal surface [175]. Accordingly, advances in deep‐tissue wearable electronics have motivated the development of photoacoustic patches for continuous hemoglobin monitoring [44]. These devices leverage the photoacoustic effect, combining optical excitation with ultrasonic detection to achieve high‐resolution imaging at greater penetration depths.

In one study by Gao et al., the design of a photoacoustic patch consists of vertical‐cavity surface‐emitting lasers (VCSELs) and arrays of piezoelectric ultrasonic transducers (Figure 6a) [44]. VCSEL diodes emit laser pulses with an average irradiance of 1.8 × 10³ W/m² toward the skin surface, propagating into tissue to reach hemoglobin. Absorption induces thermoelastic expansion in hemoglobin, generating acoustic waves. These waves are detected by the transducer array, which produces high‐resolution spatial maps of the target biomolecule. All components are encapsulated in Ecoflex, enabling the patch to undergo multimodal deformation (Figure 6b). The robust design enables the photoacoustic patch to undergo multiple bends without affecting the VCSEL's performance, as shown in the infrared image in Figure 6c. The patch integrates 24 VCSELs arranged in four columns, with each column connected in series to ensure uniform illumination. It also includes 240 transducers organized into 15 columns, each with 16 transducers, interleaved between the lasers. During image reconstruction, four adjacent transducer elements within each column are virtually connected in parallel and combined to form a single linear array element. This process generates 13 overlapping linear arrays with a three‐element overlap between neighboring arrays, improving the SNR and lateral sampling density. After calibration, this architecture enables reconstruction of a 3D temperature map for continuous monitoring (Figure 6d).

FIGURE 6.

Design, layout, working principles, and application of a wearable photoacoustic patch. (a) Schematic of the device structure and working principles. The system is encapsulated in Ecoflex, with serpentine copper electrodes interconnecting the arrays of piezoelectric transducers and VCSELs. (b) Optical photographic representation of a photoacoustic patch wrapped on a non‐developable surface as a mode of deformation. (c) Infrared images of the photoacoustic patch under the bending working mode with the VCSEL under operation (Laser wavelength: 850 nm). (d) The photoacoustic patch comprises alternating columns of laser sources and ultrasound transducers. The transducer array consists of 15 columns, each containing 16 elements. During image reconstruction, four adjacent transducer elements within each column are virtually connected in parallel and combined to form a linear array element, producing 13 overlapping linear arrays with a three‐element overlap between neighboring arrays to improve SNR and lateral sampling density. This architecture generates 13 slices in the y‐direction for 3D imaging, enabling calibrated volumetric temperature mapping of blood vessels. (e) Photoacoustic image slices of an artificial blood vessel placed 2 cm under an ex vivo porcine tissue. (f) The corresponding 3D hemoglobin distribution of the ex vivo porcine tissue. (g) Forearm placement of the photoacoustic patch. (h) Cross‐sectional changes of the vein before, during, and after cuff inflation. The white outline is used to measure the target vein's size. Reproduced with permission from Gao et al. (2022) [44]. Copyright Springer Nature.

For ex vivo evaluation, the patch mapped hemoglobin in porcine tissue using 850 nm excitation, a wavelength favorable for deep‐tissue penetration [44]. An artificial cyst containing multiple biofluids (including bovine blood) was prepared to compare optical absorption. The results indicate that blood has the highest absorption coefficient, approximately 6.114 cm⁻¹, across the wavelength range 700–1000 nm. This makes hemoglobin a suitable biomarker for deep‐tissue detection at 850 nm. The system reconstructed thirteen 2D photoacoustic images and fused them to obtain a 3D hemoglobin distribution (Figure 6e,f).

The patch was also tested on healthy volunteers to reconstruct 3D maps of forearm vasculatures (Figure 6g) [44]. Venous occlusion plethysmography was used to assess hemodynamics, including vascular resistance. By inflating a cuff, venous return was obstructed, leading to arterial inflow and a ∼19% change in vessel diameter over 60 s. The vascular response was monitored by measuring the vein's cross‐sectional area during inflation and deflation (Figure 6h). Researchers also applied the photoacoustic patch on the subject's neck for the detection of the jugular vein. The photoacoustic dual‐mode imaging consisted of one sliced photoacoustic image superimposed on B‐mode ultrasound images. Whereas ultrasound alone showed low contrast for the jugular vein, the photoacoustic channel yielded high contrast due to stronger hemoglobin‐based signals. Leveraging the photoacoustic effect, this imaging approach provided high sensitivity and contrast, enabling vascular detection via hemoglobin mapping.

6. Conclusion and Perspectives

We have reviewed recent advances in wearable bioimaging, including those that detect the electrical, optical, and ultrasonic properties of tissue [109, 166]. High‐resolution modalities, such as ultrasound and PAT, are critical to assessing precise anatomical changes, whereas functional modalities, such as EIT and fNIRS, serve an important role in characterizing bulk tissues. These modalities have been integrated into everyday wearables, offering several advantages in portability, accessibility, and patient comfort. EIT devices, such as belts and textiles, can provide real‐time monitoring of electrical impedance within deep tissues [48]. The emergence of flexible imaging patches based on ultrasound and PAT can provide high‐resolution images of anatomical structures while adopting a comfortable, lightweight design [44, 46]. Current and emerging applications of these wearables include assessment of lung function, muscle activity, and cardiac performance. Despite this promise, most systems remain at early research or preclinical stages. Below, we outline key challenges and research directions for wearable imagers; Figure 7 summarizes the main obstacles for translation.

FIGURE 7.

Future directions for wearable bioimaging technologies. Future directions for wearable imagers may involve techniques for natural energy harvesting, such as biomechanical, biochemical, and thermal, to maximize mobility and patient comfort. The working mechanisms of fundamental imaging modalities can also be refined to further optimize tissue penetration depth, contrast, and spatial resolution. Lastly, to realize the full clinical potential of wearable imagers, we can consider developing closed‐loop theragnostic systems, integrating IoT, and working with industry to meet safety standards for full market translation.

6.1. Considerations for Long‐Term Stability

Although many of the wearable imaging platforms surveyed in this review are intended for continuous use, extended operation introduces limitations that are not apparent in proof‐of‐concept experiments. One factor affecting long‐term performance is the device's ability to adhere to the skin. Over time, sweat, skin oils, epidermal turnover, and repeated motion progressively weaken adhesion, leading to partial detachment and a decline in signal quality [176]. In many systems, hydrogels are incorporated to enhance adhesion while simultaneously providing compliant electrical or acoustic coupling at the skin‐device interface [177]. However, prolonged wear leads to gradual hydrogel dehydration, which increases stiffness, alters viscoelastic behavior, and increases parasitic impedance [178]. These changes can compromise conformal contact and coupling efficiency, thereby reducing signal transmission, image contrast, or sensitivity. The rate and extent of dehydration are strongly influenced by environmental conditions such as temperature and humidity, as well as hydrogel composition and encapsulation.

Another consideration for long‐term stability is biofouling, the accumulation of biological residues such as lipids and proteins, on device surfaces during prolonged skin contact [179]. These contaminants can alter the electrical conductivity, optical transmission, or refractive index at the skin‐device interface, thereby reducing image contrast and SNR. In addition, fouling layers promote microbial growth, thereby triggering local immune responses in the skin. From a materials perspective, biofouling may be mitigated through surface engineering approaches, including low‐surface‐energy coatings, hydrophilic or zwitterionic polymers, and antifouling encapsulation layers that reduce nonspecific protein adsorption and bacterial adhesion [180, 181]. However, the long‐term stability of such coatings under repeated mechanical deformation and exposure to sweat remains a significant challenge for wearable imaging systems.

Beyond interfacial effects, long‐term operation also imposes significant demands on the mechanical integrity of wearable imaging devices. Continuous cyclic loading arising from respiration, muscle contraction, and daily activities can induce mechanical fatigue in soft substrates, stretchable interconnects, and integrated transducer or emitter arrays [182]. Over time, these stresses may lead to microcracking, delamination, or permanent deformation, resulting in calibration drift, spatial misalignment, and reduced measurement repeatability [183]. Addressing mechanical fatigue and structural reliability is therefore essential for maintaining stable imaging performance over clinically relevant wear durations.

6.2. Autonomous Power Sources

A central challenge of wearable imaging technology is delivering sufficient, safe power in a miniaturized, comfortable form factor [184, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197]. Many prototypes still rely on wired tethers to benchtop supplies [44, 46, 118, 135, 172] or bulky commercial batteries [45, 173], which restrict mobility and add weight. Wireless power transfer (e.g., inductive coupling) can reduce on‐board mass but typically requires proximity to a primary transmitter, limiting everyday use [198]. Consequently, on‐body and self‐powered strategies are needed for continuous monitoring.

One approach is to harvest energy from biomechanical motion, such as respiration, arterial pulses, or voluntary movements, using nanogenerators [199, 200, 201, 202, 203, 204, 205]. These include triboelectric nanogenerators (TENGs) [119, 131, 199, 200, 201, 202, 203, 204, 205, 206, 207, 208, 209, 210, 211, 212, 213], piezoelectric nanogenerators (PENGs) [31, 66, 214, 215, 216, 217, 218, 219, 220, 221], and magnetoelastic generators (MEGs) [132, 133, 134, 222, 223, 224, 225, 226, 227, 228, 229, 230, 231, 232, 233]. TENGs exploit contact electrification along the triboelectric series to generate charge during cyclic contact separation [234]. The triboelectric series is a qualitatively ranked list of materials based on their ability to acquire positive or negative charge and the strength of that charge [190, 235, 236, 237, 238, 239, 240, 241, 242, 243, 244]. PENGs convert cyclic stress and strain into charge using the piezoelectric effect [66]. MEGs couple a magnetomechanical (MC) layer to a magnetic‐induction (MI) layer: motion perturbs magnetization in the MC layer, inducing an electric signal in the MI layer [132, 133, 134, 223, 224, 225, 226, 227, 228, 229, 230]. While these nanogenerators are conducive to the miniaturization of many wearable imaging platforms, their output power is often limited to the microwatt to milliwatt range [245]. As such, they are well‐suited for low‐power sensing modalities, such as EIT, but are unlikely to meet the higher power demands of ultrasound, optical, or photoacoustic systems.

Wearable imaging platforms can also be powered by natural sources [126]. The human body constantly generates heat as a byproduct of metabolic processes, reaching a maximum output of approximately 525 W [246, 247, 248, 249]. If two dissimilar materials are placed in proximity with a temperature gradient, electricity would be generated in a phenomenon known as the thermoelectric effect [126, 250]. Accordingly, a thermoelectric generator (TEG) may be placed on the skin to power the wearable imager via this effect. To construct such a device, we must choose two thermoelectric materials with a high figure of merit (ZT), a dimensionless quantity that depends on electrical and thermal conductivities. Another source of natural energy is solar [251, 252, 253, 254]. When light reaches a solar cell (SC), it excites electrons in the semiconductor material, creating electron‐hole pairs [245]. These electron‐hole pairs generate an electric current, which is then stored for later use. In direct sunlight, this process can produce up to 100 mW/cm², demonstrating great potential for powering wearable imagers [126]. Hence, self‐powered energy sources may be integrated into wearable imaging technology in the future to enhance patient comfort and mobility.

6.3. Safety Considerations

Adherence to established safety limits is a fundamental requirement for translating wearable imaging platforms, as they may remain in close contact with the body for several hours or even days. This longer timescale places stricter constraints on cumulative exposure, local heating, and tissue stimulation [255]. For EIT, safety is primarily governed by limits on injected current amplitude, frequency, and waveform [109]. Standards such as IEC 61508 and IEC 60601 provide a framework for functional safety in electrical systems, emphasizing risk analysis, fault tolerance, and fail‐safe operation [109, 256]. EIT platforms can align with these principles through hardware‐ and software‐based current limiting, electrical isolation, continuous electrode‐contact and impedance monitoring, watchdog circuitry, and automatic shutdown during abnormal operating conditions [257, 258]. Wearable EIT systems typically use low‐amplitude alternating currents in the microampere‐to‐milliampere range, at frequencies selected to minimize neuromuscular stimulation and electrochemical reactions at the electrode‐tissue interface [50, 259]. Nevertheless, prolonged operation introduces additional risks, including electrode polarization, localized heating, and skin irritation under sustained current injection. Careful selection of excitation frequency, electrode materials, and current density, combined with impedance monitoring and fail‐safe current limiting, is therefore critical for safe EIT measurements.

For ultrasound‐based wearable systems, safety is commonly evaluated using the Mechanical Index (MI) and Thermal Index (TI), which capture different aspects of exposure risk [260]. MI is primarily influenced by the peak negative acoustic pressure and the transmitting frequency, with lower frequencies and higher‐pressure amplitudes increasing the likelihood of cavitation. TI, in contrast, reflects the potential for tissue heating and depends on the time‐averaged acoustic power, duty cycle, and exposure duration, as well as on local heat dissipation at the skin‐transducer interface. Although wearable ultrasound patches typically operate at acoustic outputs well below clinical diagnostic limits, prolonged or repeated operation can increase cumulative thermal load, particularly for conformal devices that limit airflow [45].

Optical excitation introduces distinct safety constraints that depend strongly on wavelength, pulse structure, and tissue site. For photoacoustic patches, optical exposure must comply with the maximum permissible exposure (MPE) limits defined for skin and underlying tissue, particularly when using pulsed or high‐peak‐power sources [261]. Wearable photoacoustic patches typically operate in the near‐infrared window and rely on low average irradiance combined with short pulses to generate detectable acoustic signals while remaining below MPE thresholds [44]. However, continuous, repeated illumination of the same skin region requires careful management of the repetition rate, beam profile, and thermal accumulation. Similar considerations apply to fNIRS systems, which employ continuous or quasi‐continuous scalp illumination [155]. Although fNIRS operates at optical power densities that are orders of magnitude lower than those used in therapeutic devices, prolonged illumination can still lead to local heating or discomfort if optode contact is poor or ventilation is restricted [155]. Wearable fNIRS designs therefore benefit from conservative source power, distributed illumination across multiple optodes, and materials that promote heat dissipation and skin breathability. These factors become increasingly important for high‐density fNIRS systems, where many sources may be active simultaneously.

6.4. Toward Intelligent and Connected Wearable Imaging Systems

A promising direction for wearable imagers is closed‐loop theragnostic systems that integrate a therapeutic module, such as a microneedle patch for drug delivery, which activates upon detection of diseased tissue [262, 263]. This approach combines the diagnostic capabilities of ultrasound and photoacoustic tomography with targeted treatment, providing preventive, timely care. To determine the medication and proper dosage to be administered by the therapeutic module, we can integrate a neural network, specifically designed to diagnose medical conditions, and predict their severity. The model can also consider individual patient characteristics, such as medical history, genetic factors, and social determinants of health, to tailor the treatment approach for each patient [264]. Once the neural network has made a prediction, the appropriate dosage of a prescription may be released from a reservoir, enabling a painless, highly regulated method of drug delivery.

Another direction is the transmission of medical information from wearables to physicians worldwide as our society enters the era of the Internet of Things (IoT) [265, 266]. The work by Lin et al. has already demonstrated the wireless transmission of images from an ultrasound patch to the cloud, and future systems could offer end‐to‐end platforms that deliver results to patients and route referrals to specialists in real time [45]. Such connectivity is particularly impactful for patients in rural or under‐resourced regions who have internet access but limited proximity to advanced care facilities [267]. However, one concern with this model is adversaries’ ability to gain access to sensitive patient information through security vulnerabilities in the IoT environment [268]. Accordingly, research is needed on robust encryption, authentication, and access control to protect data flows among patients, primary care physicians, and medical specialists.

6.5. Commercialization

Recent advances in wearable imaging point to a future in which continuous physiological monitoring becomes more accessible outside traditional clinical settings. To realize the widespread deployment of the wearable imaging platforms surveyed in this review, translating laboratory prototypes into commercial end‐products is essential. This transition introduces manufacturing challenges that are largely absent at the proof‐of‐concept stage but are critical for reliable, cost‐effective production.

For soft and flexible imaging patches, multilayer fabrication requires precise alignment of electrodes, interconnects, sensing elements, and encapsulation layers across large‐area substrates [269]. Minor misalignments that are tolerable in benchtop prototypes can significantly degrade performance or yield in high‐volume manufacturing. Yield optimization is further complicated by the use of soft, stretchable, and heterogeneous materials, which can introduce device‐to‐device variability due to variations in thickness, conductivity, or adhesion [270]. Ensuring materials uniformity across batches and over extended production runs remains a key challenge for maintaining consistent imaging performance. In this context, thorough metrology testing is essential for identifying defects early and minimizing scrap rates.

Packaging and system integration also present nontrivial hurdles. Wearable imaging systems must integrate sensing elements with power management, data acquisition, wireless communication, and mechanical support, all within a compact form factor [271]. Encapsulation strategies must simultaneously protect sensitive electronics from moisture ingress and mechanical damage, while maintaining breathability and patient comfort. Another barrier to commercialization is the requirement for long‐term reliability. Wearable imaging patches must withstand repeated mechanical deformation, prolonged skin contact, exposure to sweat and oils, and environmental stressors without degradation in electrical, optical, or acoustic performance [272]. Accordingly, reliability testing protocols must account for mechanical fatigue, biocompatibility, and environmental aging.

Beyond manufacturing considerations, the shift to mass production requires extensive regulatory testing, sufficient capital investment, and the execution of clinical trials to demonstrate safety and efficacy at scale [273]. We therefore advocate early and sustained collaboration among researchers, manufacturing engineers, startups, regulatory agencies, clinicians, and end users. Such coordinated efforts will be essential for addressing fabrication challenges, improving yield and reliability, and ultimately accelerating the responsible commercialization of wearable imaging platforms.

Conflicts of Interest

The authors declare no conflicts of interest.