BioSIGNAL-AI is an international network that brings clinicians, technical researchers, device makers, regulators, ethicists, patients and health systems together, so biosignal models are judged by whether they serve clinical decisions, not by benchmark scores alone.
Imaging and text have moved into routine care. Continuous signals from monitored beds, clinic archives and wearables are already captured at scale, but much of it reaches clinicians as waves and numbers, not as a reading.
Only a few subspecialists can read subtle patterns reliably, and their time is scarce. Models could bring that reading to every monitored patient.
The promise: expert-level reads where specialists are few.
Far more is collected than clinicians can review. A cheap, routine signal could reduce the need for a slow or invasive test, or reveal a marker nobody thought to look for.
The promise: more from what is already being recorded.
A trajectory can replace a single label, so clinicians and patients have time to act before deterioration becomes clinically apparent.
The promise: earlier warning, with global reach, because these are among the cheapest and most widely acquired measurements.
Biosignal AI is advancing fast, but the way it is evaluated has drifted away from the clinic.
Whatever labels a public dataset happens to carry end up defining the task. Many are outputs of older algorithms rather than answers to a clinician’s problem.
Our answer: every task starts from a stated clinical need.
The recording device, its settings and the state of the patient shape every result, yet rarely travel with the data. A model called general makes a claim no single dataset can test.
Our answer: the conditions are written down with every claim.
Averaged scores hide what matters at the bedside, where some mistakes cost far more than others. Benchmarks rarely ask whether a model adds to current practice, or how it will run in a real hospital.
Our answer: evaluation is tied to the clinical decision and compared with current practice.
Left as it is, a strong benchmark score is easily mistaken for clinical readiness. That risks patient and clinician trust, and points research at problems that are not the clinic’s.
No single discipline, institution or country can do this alone. BioSIGNAL-AI gives stakeholders everywhere a common language and a path to testing claims against real needs, so those needs reflect the settings where care is actually delivered.
Clinicians describe what current care cannot resolve. Developers describe what a model claims to do. Both use one template and shared terms, so each claim can be matched to the need it answers and followed as it is revised.
Who writes: clinicians state needs, developers state claims.
As a need matures, the network will work with the people who hold the data to assemble shared testbeds and evaluation suites, so claims can be compared on well-defined tasks.
How: together with data holders, as each need matures.
Device manufacturers, regulators, ethicists, patients and health systems take part alongside clinicians and researchers, both in contributing to needs and claims and in the discussions the network will convene.
Where: seminars and meetings the network will convene.
No single group can write a complete need or claim alone. Each brings something the others cannot supply.
State the problem current care cannot resolve, say which patients a model should serve, which decision it would support, and which errors would matter most.
Turn a clinical problem into a task that can be tested: which label captures it, how that label is assigned, and which measures would show a solution works.
Manufacturers document how recordings are made. Regulators set out the evidence expected for an intended use. Patients name the outcomes that matter to them. Health systems describe the settings a model must run in, and ethicists join the discussions the network convenes.
A need or claim is complete when each of its readers can act on it: a clinician knows whether a model applies to the patient in front of them, a researcher knows which problems deserve attention, and a regulator knows what evidence supports the claim.
Participation is tiered, from registering a single need or claim to curating a whole clinical task. The network launches with a founding steering group of clinicians and technical researchers.
Tell us about your work and attach your CV.
Current members read your application and vote. You get a confirmation email straight away.
Approved applicants get a welcome email. If demand is high, you go on the waiting list.
An international network of clinicians, technical researchers, device makers, regulators, ethicists, patients and health systems working to make the evaluation of biosignal AI reflect real clinical needs. The name stands for Biomedical Standards and Infrastructure for Global Network on biosignAL Artificial Intelligence.
Benchmarks rarely say which clinical need a task serves, under what conditions a result holds, or how success is judged against the decision it informs. A strong benchmark score is then easily mistaken for readiness for the clinic.
No. Curated datasets, benchmark suites and data platforms are valuable and the network builds on them. It gives them a clinical specification to consolidate around, so results become easier to compare and trust.
Everyone whose work depends on biosignal AI: clinicians, technical researchers, device manufacturers, regulators, ethicists, patients and health systems. Clinicians and technical researchers write needs and claims together, and the others contribute what they cannot supply. The network launches with a founding steering group of clinicians and technical researchers.
No. The shared way of stating needs and claims is a foundation, not a finished standard. Existing efforts can build on it rather than compete with it.
It complements them. Reporting guidelines make studies transparent, and community frameworks have shown that agreed definitions work within one area. The network aims to make that practice the norm, one clinical need at a time.
Such systems can only rely on a biosignal model that performs as claimed under its stated conditions, and they cannot check that from the signal alone. Clinically grounded evaluation is what makes the claim trustworthy, whether a model works alone or inside a larger system.
Any biosignal. The network is not limited to one type of signal or one clinical area.
Everyone whose work depends on biosignal AI is welcome, from clinicians and technical researchers to device makers, regulators, ethicists, patients and health systems. Current members review every application.