Three layers to a wearable ultrasound

Ultrasound sees muscle that electrodes cannot. Making it wearable takes an interface that stays on skin, a transducer that bends, and a circuit that reads it untethered.

Yeeun Lee  이예은

M.S. candidate, Mechanical Systems Engineering, Sookmyung Women's University
B.S., Chemical and Biological Engineering

LinkedIn Manuscript in preparation
skin muscle Flexible circuit planned Wearable transducer in progress Acoustic interface built · 1 mm thick 18 mm core 10 mm adhesive edge beam path
The system in cross-section, drawn to the dimensions that exist. The couplant is built and written up; the transducer and the circuit above it are outlines so far.

Abstract

Precise muscle monitoring needs force and joint motion read together. IMUs give kinematics but no force, and surface EMG gives activation but cannot reach deep muscle, so our lab is building one wearable module that fuses ultrasound, EMG and IMU. Ultrasound is the only one of the three that reaches deep muscle, and the only one that still cannot stay on skin for a working day — the couplant dries within hours and never adheres. I show that in AMPS hydrogels acoustic transparency and skin adhesion are governed by the same compositional variable, so no single composition satisfies both, and resolve that constraint geometrically rather than chemically. The resulting couplant stays coupled for 24 h and supports 71 consecutive measurement cycles from one attachment.

ContextMuscle monitoring needs more than one signal

skin + fat rectus femoris vastus intermedius femur IMU kinematics Surface EMG activation Ultrasound structure no depth information — limb motion only attenuates with depth; superficial muscle only resolves every boundary above skin surface depth bands drawn at the boundary depths measured on the thigh
Why the fusion module needs ultrasound at all. IMU and surface EMG stop at the skin; only ultrasound resolves the deep boundaries the deformation index is measured between. It is also the only one of the three that cannot yet stay on skin for a working day.

Layer 1 · builtThe couplant is the bottleneck

Two literatures have grown around hydrogel couplants — one measuring acoustic transparency, the other adhesion — and neither measured both on one compositional series. I came to this from the materials side, which is why the bottleneck I saw was interfacial chemistry rather than transducer design.

No composition is both transparent and adhesive

Across 20 to 40 wt% AMPS, attenuation rises 2.59-fold and 90° peel strength 3.6-fold along the same axis.

A single-composition couplant is therefore a compromise whose size is set by the material rather than by the design — a negative result nobody had published.

00.10.20.30.4 02468 peel strength [N m⁻¹] α [dB mm⁻¹] AMPS20AMPS30AMPS40 soft tissue 0.215 dB mm⁻¹ commercial hydrogel 3.99 N m⁻¹ no composition here transparent AND adhesive means of 4–9 specimens per composition
A usable couplant must sit below the soft-tissue line and right of the commercial-adhesion line. The series walks past that corner: AMPS20, the one transparent composition, reaches 45% of commercial adhesion.

Separating the two functions in space resolves it

Earlier work stacked adhesive on acoustic, under a quarter wavelength so the stack would not register. I divided them radially instead: a transparent core inside an adhesive edge, sized so the beam never reaches the edge.

The condition contains no wavelength, so it does not depend on the operating frequency, and it carries no thickness budget. Freed from the attenuation constraint, the edge can be chosen on adhesion alone.

transducer exposed face 8 mm SKIN MUSCLE core 10 mm · AMPS20 edge 4 edge 4 the beam never leaves the core Ø18 Ø10 Ø8 AMPS37 + 3 wt% DMA edge AMPS20 core
The 10 mm core exceeds the 8 mm exposed face, so the adhesive composition never enters the beam. The edge reaches 22.6 N m⁻¹, 5.6× the commercial hydrogel.

It holds for a working day

MetricCouplantBenchmark
90° peel on glass22.6 · 5.6×3.99
90° peel on skin33.5 · 3.3×10.2
Mass kept at 5 h77%3× lower
Phantom coupling24 h · 5/512.0 h · 0/5

Peel strength in N m⁻¹ against a commercial adhesive hydrogel; mass against ultrasound gel; coupling as runs still intact at 24 h. Gel lost 40 dB of back-wall echo over the same day while the couplant held within one. Log-rank p = 0.0018.

The transducer housing lifted from the forearm; the couplant stays bonded and tents the skin upward with it.
Lifting the housing tents the skin rather than releasing it.

One attachment shows a timescale discrete assessment cannot

Across 13 sessions and 71 cycles, the couplant was never reapplied within a session.

Illustration of a seated participant on a leg-extension machine, with a callout showing the transducer over the concentric couplant on the quadriceps, between two stimulation electrodes.
The measurement as it stands: a 5 MHz probe over the couplant on the quadriceps, with electrical stimulation either side. The housing, the strap and the cable are what layers 2 and 3 remove.
baseline
stimulation windows
Thirty seconds of M-mode through the quadriceps: skin and fat at the top, then rectus femoris, vastus intermedius and the femur echo, which steps deeper during each of four stimulation windows. skin + fat rectus femoris vastus intermedius femur
5 10 15 20 25 30
time [s]depth increases downward
What the probe returns. Each step in the femur echo is the muscle thickening under one stimulation window, and the deformation index is that thickness change, measured between the boundaries drawn here.
1008050 deformation index [%] Within one cycle four windows, 40 s apart exercise 66 79 PrePostRec 1 Rec 2Rec 3 recovers Across one session pre-exercise value, cycle to cycle 82676458 12345 declines, front-loaded
Two timescales, moving in opposite directions: within a cycle the index recovers, while across the session the pre-exercise value itself falls −41.6% — in every one of 12 sessions. Both components of fatigue are visible, swelling at +7.8% and loss of contractility at −37.6%. Variation between attachments is 2.8× the per-cycle effect, so reapplying the couplant would have buried this signal.
Four limits on these claims
  1. Skin adhesion was compared on one volunteer, without same-day pairing, so it does not resolve inter-individual variation.
  2. Coupling stability was measured on a phantom, excluding wearer motion; post-collapse gel values sit at the noise floor, making 24 h a floor estimate.
  3. Torque was not measured here, so the decline is not independently confirmed as fatigue within this study.
  4. Stimulation current was not recorded, so the fall in evoked deformation cannot be separated from change at the electrode–skin interface.

Layer 2 · nextMake the element bend with the body

Solving the interface moved the limit one layer out. A conformal element removes the housing, the strap and the pressure point at once — and the specification it has to meet is already fixed by layer 1.

SurvivesThe core-wider-than-aperture condition holds no wavelength, so a new element does not invalidate it.
Re-deriveIt was set against a flat 8 mm face; a curved element needs the core sized for a curved footprint.
Gets easier33.5 N m⁻¹ on skin can supply the contact pressure the strap supplies now.
Becomes the testBack-wall amplitude, temporal SNR and reverberation ratio transfer as the acceptance bar.

The open question is whether bending walks the beam out of the core.

Layer 3 · plannedCut the cable

Every number on this page was acquired through a coaxial cable to a benchtop pulser-receiver. The metrics that decide whether a couplant failed also decide whether a front end is good enough, so this layer inherits a specification rather than a wish list.

Dynamic rangeGel collapse spans 40 dB; the couplant holds within 1 dB. The front end must resolve a decibel across that span.
Noise floor43–56 dB temporal SNR is what the benchtop system delivers today — the number to approach.
Duty cycleAcquisition is sparse rather than continuous, which changes the power budget entirely.
DepthReaching bone through the quadriceps at 5 MHz sets the time-of-flight window and sampling rate.

Each layer removes one reason the measurement has to happen in a laboratory. The couplant removed reapplication; the transducer removes the housing; the circuit removes the room.

OutputsManuscript and figures

Y. Lee et al. “Decoupling Adhesion from Acoustic Attenuation: A Concentric Hydrogel Couplant for Long-Duration Wearable Ultrasound.”
First author · prepared for Advanced Healthcare Materials · 9,025 words, 5 figures, 2 tables, 22 SI items · IRB approved

Materials

AMPS and DMA hydrogel synthesis, two-stage molding, 90° peel

Acoustics

Impedance, attenuation, phantom design, pulse-echo quality metrics

Biosignal

A-mode and M-mode deformation, evoked contraction protocols

Computation

Python, survival analysis, figure automation

Keywords

wearable ultrasound · hydrogel couplant · acoustic attenuation · skin adhesion · muscle fatigue monitoring · sonomyography · flexible electronics · biosignal analysis

Key references
  1. C. Wang et al., Science 2022, 377, 517 — bioadhesive hydrogel, axial separation.
  2. K. W. K. Tang et al., Nat. Commun. 2025, 16, 4940 — AMPS couplant, states the dehydration problem.
  3. J. T. Alvarez et al., IEEE TNSRE 2025, 33, 10 — deformation index against dynamometer torque.
  4. M. Lin et al., Nat. Biotechnol. 2024, 42, 448 — integrated wearable ultrasound imaging.
  5. X. Gao et al., Nat. Electron. 2024, 7, 1035 — single-element wearable muscle sensing.
  6. L. A. Negron et al., IEEE TUFFC 2002, 49, 1543 — attenuation against polymer concentration.
  7. J. Yi et al., ACS Appl. Bio Mater. 2020, 3, 8943 — catechol adhesive for intraoral ultrasound.

Yeeun Lee · Mechanical Systems Engineering, Sookmyung Women's University · linkedin.com/in/yeeunlee-smwu