Skip to main content

AI/ML-Based SaMD — ANVISA Guidance

ANVISA's approach to AI/ML SaMD​

ANVISA is actively developing guidance on AI/ML-based Software as a Medical Device (AI/ML SaMD), informed by its participation in IMDRF working groups and monitoring of FDA, EU, and other major regulatory authority guidance.

Key concepts for AI/ML SaMD in Brazil​

Locked vs. adaptive algorithms​

ANVISA currently evaluates locked algorithms (fixed at registration) using the standard SaMD technical dossier framework. For adaptive algorithms (continuously learning / updating post-deployment), ANVISA expects:

  • A predetermined change control plan (PCCP) in the registration dossier — describing the types of changes that may occur, the limits of acceptable change, and how changes will be validated before deployment;
  • Transparency about the algorithm's learning methodology; and
  • Post-market monitoring of algorithm performance.

Transparency and explainability​

ANVISA expects AI/ML SaMD manufacturers to address:

  • How the algorithm makes its decisions (explainability);
  • What data was used to train the model and its limitations (training data bias);
  • The algorithm's performance characteristics (accuracy, sensitivity, specificity) on test datasets; and
  • Plans for ongoing performance monitoring in real-world use.

Clinical evidence​

Clinical evidence for AI/ML SaMD must demonstrate the algorithm's performance on a test dataset that is representative of the intended Brazilian patient population. ANVISA may require supplementary validation on Brazilian clinical data if the training and test datasets were drawn from non-Brazilian populations.

Official resources​

Official sources

Verify all information against official ANVISA sources before making regulatory decisions.