Software
What ran
The pipeline, its version and its configuration are content-addressed and signed. Change a single byte of code or settings and the fingerprint changes, so “the approved version” becomes a fact you can check.
Trust layer · with EQTY Lab
Every result, traceable to its source
Medlea attaches cryptographic evidence to each functional lung analysis: which data went in, which software processed it, on which hardware, and what came out. Your Data Protection Officer gets records to inspect, not promises to accept.
Hospitals adopting AI hear one question from their data protection officers before any other: can you show what happened to this patient’s data? Medlea answers it case by case, with a signed provenance manifest generated inside the hospital’s own perimeter.
Three things attested, on every run
Hardware
Where it ran
When the compute platform supports confidential computing, the record carries hardware evidence from CPU, GPU and TPM, verified against the manufacturers’ roots of trust. Software integrity and hardware attestation are reported separately, each with its own scope.
Data
What it touched
Inputs and outputs are fingerprinted at every stage, and each stage’s output must match the next stage’s declared input. Alter a file and the chain breaks visibly. The manifest holds hashes and identifiers, never patient data.
What it gives your Data Protection Officer
Under the GDPR, accountability means being able to demonstrate compliance, not only to state it. Attestation turns the processing of each case into evidence that stands on its own.
- ● Per-case processing records: inputs, processing steps and outputs with their timestamps, each linked to its source.
- ● Independent verification: the manifest is checked with open tooling and public keys, with no access to Medlea systems.
- ● Tamper evidence: an altered input, model or result is detected and named, never silently absorbed.
- ● Minimisation by design: processing stays on-premise and the record contains no patient data, so it can be handed to auditors.
- ● Audit-ready packages: approvals, transfers and processing steps for a case gathered for review, with open questions flagged rather than hidden.
- ● Material for your DPIA and for the record-keeping and traceability expectations of the EU AI Act.
Partnership
Built with EQTY Lab
EQTY Lab develops verifiable-compute and AI-governance technology: it records what an AI system did and lets anyone check it. Medlea integrates the EQTY Lab SDK around the LeaFlow pipeline as an observational layer, designed to record each step without altering the clinical computation. The resulting lineage opens in EQTY Lab’s Lineage Explorer, from the result back to every recorded input, model and processing step.
Visit EQTY Lab ↗See it end to end
From a clinical question to an audit review: CT images become a lung model, the simulation runs on accelerated compute, the agent works inside a controlled runtime, the Lineage Explorer traces the result, and a Data Protection Officer prepares the evidence package for a single case.
Illustrative edit of the demo environment, not a live capture. Narration in English, with captions.
What attestation proves, and what it does not
- ● It proves which data, software and hardware produced a result, and that nothing changed afterwards.
- ● It does not establish clinical validity. That remains the job of the device’s clinical evaluation and of the clinician’s judgment.
- ● It records what the AI assistant did. It does not decide for the clinician, and it does not declare compliance on your behalf.
- ● Every report states the limits of its own checks, including what was not verified.
Bring attestation to your hospital
We are running proof-of-concept deployments with EQTY Lab. If your DPO or IT security team wants to see how a manifest is built and verified, get in touch.
Contact us →