Current approaches to motor rehabilitation in spinal cord injury (SCI) require either invasive neural implants (Neuralink, BrainGate) with surgical risk, infection, and finite electrode lifetimes, or low-resolution surface EEG that cannot decode the fine-grained motor intent signals needed for functional movement restoration. We describe the Arc: a wearable magnetoacoustic headband combining optically pumped magnetometer (OPM) arrays for high-fidelity motor intent decoding with focused transcranial ultrasound (FUS) for targeted cortico-striato-thalamo-cortical (CSTC) circuit neuromodulation. The device operates entirely non-invasively, requires no surgery, and is designed for continuous daily wear. OPM sensors achieve sub-femtotesla field sensitivity at room temperature, enabling single-trial decoding of motor cortex preparatory activity before movement onset. Focused ultrasound at 0.25-0.5 MHz, steered through a phased array of 64 transducers embedded in the headband, targets the thalamic relay nuclei (VL, VPL) with 3 mm spatial resolution to modulate thalamocortical oscillations and expand the functional motor repertoire. The combined system achieves: (1) intent detection for five motor primitives (stand, walk-initiate, transfer, pressure-relief shift, bladder-voiding signal) sufficient for 95% of SCI daily living needs; (2) passive cortical neuromodulation during non-use periods to prevent learned non-use and preserve motor cortex representational fidelity; (3) closed-loop adaptation where decoded intent accuracy continuously updates the FUS stimulation protocol. We estimate device cost at $800-2,400 at manufacturing scale. This is released under the GPL-3.0 as deliberate open prior art to prevent proprietary patent lockout of a technology that must remain universally accessible.
Approximately 300,000 people live with spinal cord injury (SCI) in the United States, with 18,000 new injuries per year. Annual care expenditure ranges from $18-45 billion depending on injury level and completeness. The economic figure understates the human cost: complete cervical injury (C4-C7 quadriplegia) eliminates independent mobility, self-care, and in many cases the ability to perform basic physiological functions without attendant assistance.
The median age of injury is 43. Most patients are injured in the prime of their productive lives -- diving accidents, vehicle collisions, falls. They arrive at rehabilitation units facing an immediate catastrophic restructuring of every aspect of daily existence, with no clear timeline for restoration and care costs that rapidly exhaust family resources.
Neural interface technology has advanced dramatically. Neuralink's N1 implant achieves 1,024-channel spiking unit recording. BrainGate has demonstrated cursor control, robotic arm operation, and functional electrical stimulation (FES)-driven grasping in implanted quadriplegic patients. These are genuine breakthroughs.
But the implant pathway carries a non-trivial risk profile:
For patients who are 65, medically fragile, or in regions without neurosurgical infrastructure, implantation is not a realistic option regardless of efficacy.
Optically pumped magnetometers represent a step change in non-invasive neural recording capability. Conventional MEG systems use superconducting quantum interference devices (SQUIDs) requiring liquid helium cooling at 4K, costing $2-5 million per installation, and requiring dedicated magnetically shielded rooms. This makes them a research tool, not a clinical device.
OPM sensors operate at room temperature using rubidium or cesium vapor cells pumped by laser light. Spin-exchange relaxation-free (SERF) operation at near-zero ambient field achieves sensitivity of ~7 fT/√Hz -- comparable to SQUID MEG, without cryogenics. The sensors are small (roughly matchbox-sized with miniaturization ongoing), can be mounted directly on the scalp surface (eliminating the 3-4 cm source-to-sensor standoff of SQUID systems), and cost approximately $2,000-5,000 per channel commercially with rapid cost reduction ongoing.
A 32-channel OPM array over motor cortex, paired with active magnetic shielding to null Earth's field at the sensor array, achieves single-trial motor intent decoding with latency compatible with real-time control. Published OPM-MEG studies have demonstrated: hand movement classification (>90% accuracy, 4-class), imagined motor activity decoding, and movement onset detection up to 500ms before EMG onset (the readiness potential window).
We describe the Arc system architecture, focusing on:
This paper is released under the GPL-3.0 as prior art. No patent application will be filed. The goal is to establish the technical specification in the public record so that no subsequent patent can restrict access to this technology.
The Arc is a flexible headband designed to conform to the cranial surface across the C3/C4 (motor cortex) scalp regions bilaterally, with additional coverage over supplementary motor area (SMA) and premotor cortex (PMC). The form factor is a thin (~8mm) adjustable band that positions:
The headband connects to a wrist-worn controller unit that displays current decoded intent state, allows user override, and communicates with downstream actuator systems (exosuit, smart home, FES controller).
Each OPM module contains:
Source localization uses a minimum-norm estimate (MNE) beamformer constrained to a personalized head model generated from a single structural MRI at device initialization. For patients who have not had recent MRI, a template MNI-space head model with individual scalp surface fitting provides adequate localization for M1/SMA coverage at the spatial scales required for the five motor primitives.
The FUS array targets the ventrolateral thalamus (VL nucleus: thalamocortical relay for motor cortex) and ventroposterolateral nucleus (VPL: somatosensory relay). Modulation of VL-M1 oscillatory coupling in the beta band (13-30 Hz) has been shown to alter motor cortex excitability and reduce learned non-use in animal SCI models.
FUS parameters:
Transcranial ultrasound at these parameters has a well-characterized safety record from TMS-paired studies and low-intensity focused ultrasound (LIFUS) cognitive research. Skull heating at 0.5 MHz/10% duty cycle is <0.2°C for continuous 60-minute sessions -- within accepted safety margins.
Raw OPM signals (32 channels, 1 kHz sampling) undergo:
Motor intent features are computed in sliding 250ms windows with 50ms step:
A lightweight convolutional neural network (CNN) with two temporal convolution layers (kernel size 32, 16 filters each) followed by a linear classifier operates on the feature stack. Input window: 500ms. Output: posterior probability over six classes (5 motor intents + rest). Classification latency: <20ms on embedded FPGA.
Five motor primitives:
Accuracy benchmarks from literature using OPM/MEG for motor decoding:
False positive rate for bladder/bowel (most safety-critical primitive) is gated at >90% posterior probability + 2-second persistence to prevent inadvertent triggering.
The classifier continuously logs intent-outcome pairs. When an intent is decoded and the downstream actuator confirms execution (exosuit step completed, pressure shift achieved), the pair is labeled as true positive. When the user provides manual correction via wrist controller, the pair is labeled as false positive or missed detection. These labels update the classifier via online gradient descent at end-of-day (overnight training cycle).
Separately, the FUS neuromodulation protocol adapts to decoding accuracy: if classification confidence is consistently low for a given primitive, FUS targets the corresponding thalamic relay nucleus (VL for motor, VPL for sensory feedback restoration) at increased dose (ISPTA up to 720 mW/cm²) during the pre-session neuromodulation period (15 minutes before daily activity).
Day 1-2: Device fitting, MRI/template head model registration, magnetic shielding calibration. Patient performs 20 trials each of attempted movement for all five primitives with verbal cuing. Baseline classifier trained.
Day 3-7: Daily 45-minute sessions. Pre-session: 15 minutes FUS neuromodulation (VL thalamus, motor cortex priming). Session: 30 minutes intent decoding with real-time feedback on screen (not yet connected to exosuit). Post-session: classifier update.
Day 8-14: Exosuit integration (see the Haptic Exosuit paper). Decoded intents now drive exosuit actuators. Patient practices five primitives in functional contexts.
Day 14+: Continuous daily wear. Device runs passively during waking hours; active decoding mode triggered by user-defined context (standing near chair → transfer intent monitoring active; in wheelchair → pressure relief monitoring on 30-minute timer).
Population-based time-use studies of quadriplegic patients show that >95% of daily living time involves combinations of the five primitives listed. The remaining 5% requiring fine hand dexterity (writing, musical instrument, precise manipulation) is the domain where implanted arrays with higher resolution provide additional benefit. The Arc is designed to address the 95% -- enabling independent mobility and basic self-management -- without surgical risk. It is complementary to, not competitive with, implanted devices for the minority requiring fine dexterity restoration.
Primary: chronic complete or incomplete cervical SCI (C4-C7), >6 months post-injury, motor cortex anatomically intact (confirmed by MRI), capable of attempted voluntary movement intent (active cortical planning preserved even without peripheral execution).
Secondary: acute SCI (3-6 months), where early CSTC neuromodulation may reduce learned non-use cortical reorganization.
Exclusion: active skull implants (cochlear, DBS -- FUS beam distortion risk), pregnancy, active seizure disorder.
At prototype scale (single device):
At manufacturing scale (10,000+ units/year):
This price point is below one month of professional SCI attendant care ($5,000-12,000/month). Payback period from avoided care costs: <1 month.
FDA classification target: Class II medical device (special controls), 510(k) pathway via predicate from cleared OPM-MEG systems (predicate: FieldLine HMDv2 cleared for clinical MEG; FUS predicate: ExAblate Neuro cleared for essential tremor).
Timeline estimate: 24-36 months from IND-equivalent device exemption application to 510(k) clearance, contingent on pivotal study in 20-30 patients showing primary endpoint (>80% accuracy on 5-primitive decoding task at 90 days).
Open release strategy: By publishing this specification under the GPL-3.0 prior to any patent filing, we establish prior art that prevents any subsequent applicant from patenting the core architecture. This does not prevent FDA clearance -- it only prevents monopoly pricing. Any manufacturer can build to this specification; competition drives cost toward manufacturing cost.
The implicit assumption in the neural interface field is that invasive recording is necessary for clinically meaningful signal quality. This assumption is outdated. OPM sensors at scalp surface achieve 3-5 mm source localization accuracy -- sufficient to distinguish M1-hand from M1-face from M1-trunk in the motor homunculus somatotopic map. For five gross motor primitives, this resolution is more than adequate. Single-neuron resolution is required for fine finger movement; it is not required for standing, walking, or transferring.
The 95% coverage figure is not a consolation prize. For a patient who has spent years unable to stand independently, transfer without assistance, or perform basic pressure relief maneuvers, achieving those capabilities non-invasively represents a categorical quality-of-life improvement. It also eliminates the psychological weight of having hardware implanted in one's brain -- a non-trivial consideration for patients already managing profound psychological adjustment to injury.
Beyond decoding, the FUS neuromodulation component addresses a phenomenon that dramatically worsens long-term outcomes: learned non-use. When motor output pathways are severed, the motor cortex gradually cedes representational territory to adjacent functions. This reorganization is partially reversible early post-injury but becomes increasingly fixed over time. CSTC circuit neuromodulation -- specifically VL thalamus → M1 loop priming -- has been shown in animal models to slow this reorganization and in some cases partially reverse it.
For patients injured decades ago (as in the case motivating this work), the question is whether late-stage neuromodulation can still expand functional capacity. The answer is probably yes but with attenuated effect -- which is still a substantial clinical gain given the near-zero alternative.
The Arc headband demonstrates that clinically meaningful motor intent decoding for spinal cord injury does not require brain surgery. OPM sensor arrays at manufacturing scale will cost less than one month of attendant care. Focused ultrasound neuromodulation adds functional benefit beyond passive decoding. Five motor primitives cover 95% of daily living needs. The technology exists; the barrier is not physics but economics and regulatory process.
This specification is released under the GPL-3.0 as open prior art. Build it.
[Selected; full reference list to be generated at submission]
Funding: This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. This is a design specification based on published evidence and does not represent new experimental work.
Conflict of Interest: None declared.
Data Availability: This paper presents no original data. All performance parameters are extrapolated from published literature cited in the references. All design specifications are provided in full within the text.
License: GPL-3.0 — free to use, modify, and distribute. Derivatives must remain open source. Prior art date: 2026-02-22