Skip to main content

healthcare

What is ZK Heart Health?​

ZK Heart Health is a ZkAGI application that proves a private heart-risk score meets a policy threshold without revealing raw medical values. It uses zk-SNARK (Groth16) proofs for privacy-preserving analytics with compact, verifiable selective disclosure.

  • Goal: Prove a private heart-risk score meets a policy threshold without revealing raw medical values
  • Selective disclosure: The verifier learns only a YES/NO claim and a model commitment hash (which binds the model used); no feature values are shown
  • Privacy AI fit: Demonstrates privacy-preserving analytics with verifiability and compact proofs (zk-SNARK: Groth16)

How does ZkAGI prove heart risk scores privately?​

ZkAGI uses a three-step cryptographic pipeline — build, prove, verify — where the build step produces keys tied to the circuit and model weights, the prover keeps inputs private and outputs a succinct proof, and the verifier checks correctness learning only the claim and model hash. This ensures verifiable results with minimal disclosure and prevents model swaps.

  • Build step: Produces the cryptographic keys tied to the circuit and weights (prevents model swap)
  • Prover step: Keeps inputs private, outputs only a succinct proof
  • Verifier step: Checks correctness fast, learns only the claim and model hash — no raw data

ZK Heart Health — private risk proof demo

That's the full promise: Verifiable results with minimal disclosure, ready to extend to genomics and broader medical predicates.

What are the future healthcare and genomics applications of ZkAGI?​

ZkAGI's zero-knowledge proof technology extends to on-device patient proofs, polygenic risk scores, cross-site cohort queries, and regulatory compliance — all without exposing raw medical data or genomes. Technical extensions include PII commitments, on-chain verification via Verifier.sol on L2s, and integration with FHIR/SMART health data standards.

Clinical and patient apps

  • On-device proofs: Patients compute proofs on their phone; clinics/insurers verify a predicate (e.g., PRS >= T) without accessing genomes
  • Selective disclosures: "Lab value in range" or "eligibility met" proofs instead of sharing raw labs or charts
  • Consent receipts: Publish proofs that an access decision met policy/consent, without exposing data

Genomics-specific

  • Polygenic Risk Score (PRS): Prove PRS thresholds for specific conditions without revealing variants
  • Variant status (PGx/rare disease): Use Merkle-tree membership to prove presence/absence in a panel without revealing which variant
  • Cohort queries across sites: Hospital A and B return count proofs for inclusion criteria; meta-analysis runs without sharing genomes

Provider, payer, regulator

  • Utilization controls: Prove compliance with coverage criteria (e.g., biomarkers present) before reimbursement
  • Auditable privacy: Third parties can re-verify proofs later; combine with signed policies for compliance trails

Technical extensions

  • PII commitments: Hash(Name, salt) as a public handle; identity stays hidden but proofs are linkable when desired
  • Per-verifier salts: Prevent cross-verifier linkage; same user looks unrelated across orgs
  • On-chain verification: Use generated Verifier.sol for public, timestamped verification on L2s
  • MPC/HE + ZK: Use MPC/HE for collaborative compute; ZK to attest results and policies were followed
  • FHIR/SMART integration: Wrap ZK proofs as FHIR DocumentReference or SMART app artifacts

What do ZkAGI's ZK proof commands guarantee?​

ZkAGI's ZK proof commands guarantee that the cryptographic keys are bound to a specific circuit and model (preventing model swaps), that user inputs remain private with only a succinct proof as output, and that verification is fast while revealing only the claim and model hash. Together, these steps deliver verifiable results with minimal disclosure.

  • Build step produces the cryptographic keys tied to this circuit and weights (prevents model swap)
  • Prover step keeps inputs private, outputs only a succinct proof
  • Verifier step checks correctness fast, learns only the claim and model hash — no raw data