For clinicians and researchers
A neurological wearable earns clinical use only when it can explain itself.
Accuracy without an audit trail is a demo. We are building a platform whose every output traces to a rule you could read in a chart note, and we would like your help deciding what it should measure first.
The clinical problem as we see it
Neurological care runs on snapshots.
A patient with Parkinson's disease may be assessed for twenty minutes twice a year. A seizure is described from memory, days later. Cognitive decline is caught once it is already visible. The disease is continuous; the data is not.
Wearables were supposed to close that gap, and by and large they have not, for two reasons. The signals are noisy and the states they encode are genuinely graded, so systems tuned to binary thresholds either cry wolf or stay silent. And the models that do perform well are opaque, which means no clinician can defend acting on their output, and no regulator should let them.
Our position. A wearable earns clinical use when it is right about ambiguous states and can explain itself in language a neurologist would accept. We have designed the platform around both constraints rather than solving one and hoping the other follows.
Explainability
Every output decomposes into rules a person can read.
Our inference layer is built on fuzzy logic rather than a deep network. In practice that means three things for the clinician receiving its output.
Graded, not thresholded
Severity is a degree
A tremor is not present or absent. Inputs are mapped to membership functions that describe how much, so a signal can be 0.7 elevated and 0.3 normal rather than forced across a cutoff no one would defend.
Transparent by construction
The rules are the model
Conclusions are reached by a rule base written in plain clinical sentences and refined against data. The explanation is not a saliency map bolted on afterward. It is the mechanism.
Arguable
You can disagree with it
Each output arrives with the rules and inputs that produced it and a stated confidence. Disagreement is possible, which is the point. A model that cannot be argued with cannot be trusted.
What the platform measures
Eleven sensors, seven sites, three modalities.
The HemiSense Array places sensors at the headband, chest and waist on the midline, and at elbow, wrist, knee and ankle on both sides. A product uses only the sites its indication needs.
Neural
Electrophysiology
Cortical rhythm, event morphology and state, read at the head.
Motor
Movement and motor signal
Efferent motor signal at the limb sites, and inertial sensing for tremor, bradykinesia, gait and limb use.
Autonomic
Physiologic context
Cardiac, electrodermal and temperature signals that carry autonomic state and often precede neurological events.
First add-on
SureHand: a closed loop on the forearm.
The HemiSense Array is the sensing and inference layer and, worn alone, a detection, monitoring and prediction instrument. SureHand is the first add-on that acts on it, using two of the seven sites. The elbow sensor reads the efferent motor signal before the movement occurs. The inference layer separates intended movement from the tremor component and predicts the tremor's next phase. Conductive wire woven into an elastic forearm sleeve delivers an out of phase impulse to oppose the tremor component. A wrist accelerometer measures the actual hand motion and closes the loop, tuning the model continuously to the individual.
The intervention is the counter-impulse itself. There is no drug delivery and no rigid actuator on the moving segment. The same measurements that drive the loop are recorded, so the device is a longitudinal instrument as well as an intervention.
Full product detailStatus. Pre-clinical. Engine development and evaluation against public and partner neurological datasets. No human testing has been performed. We are seeking collaborators for a feasibility study under IRB oversight, and we would rather design the outcome measures with you than present them to you.
Indications
Four we are evaluating, in the order the evidence allows.
Parkinson's disease
Motor fluctuation, tremor and gait, and response to medication across the day. Lead candidate, because SureHand gives it a concrete first instrument.
Epilepsy
Event detection, and the harder problem underneath it: forecasting risk before onset from autonomic and cortical precursors.
Cognitive decline
Early signal and progression tracking in Alzheimer's disease and related dementias, ambulatory rather than episodic.
Stroke recovery
Objective measurement of neurorehabilitation between therapy sessions, with limb use and gait as primary signals.
Working with us
What we are asking for, and what we offer in return.
We are looking for
- Clinicians in movement disorders willing to define what a meaningful tremor outcome looks like and how it should be measured
- Research groups with neurological datasets, cohorts or clinical access for feasibility work
- Advisors with regulatory depth in neurological devices
- Honest criticism of the approach before it hardens
We offer
- Co-design of study protocols and outcome measures from the start
- Full transparency of the inference layer, since explainability is the product
- Authorship and attribution appropriate to contribution
- A platform designed to be argued with
Help us design the first study.
The outcome measures, the cohort and the reference standard are all still open. That is deliberate.