6 must-have properties of AI as a Medical Device and how to approach the FDA guiding principles

Authors: Paulo Pinheiro, Head of Software

In the ever-evolving landscape of modern healthcare, a profound transformation is taking place at the intersection of Artificial Intelligence (AI) technology and medicine in the shape of a new paradigm shift of Artificial Intelligence as a Medical Device (AIaMD).

What is AIaMD?

AIaMD refers to the use of AI technologies in medical applications with the intention of diagnosing, treating, monitoring, or preventing diseases. Just like traditional medical devices, AIaMDs are designed to assist healthcare professionals and improve patient outcomes.

What makes AIaMD different?

In the medical practice, the concept of replicating human actions is not novel. Take automatic blood pressure monitors (sphygmomanometers) as an example – these devices simply mimic the actions of skilled medical professionals by identifying and relaying the Korotkoff sounds that indicate systolic and diastolic blood pressures. Nevertheless, such technologies do not operate autonomously from human logic; instead, they rely on established clinical protocols that have been previously validated to diagnose medical ailments or administer treatment. They are static systems governed by predefined rules and validated protocols, programmed to generate specific results based on the input values they receive.

On the other hand, AIaMD leverages extensive datasets and intricate statistical methodologies to unveil novel connections among inputs, actions, and outcomes. These data-centric or machine-learning systems do not follow explicit programming to yield predetermined results; instead, they adopt a heuristic approach, capable of learning and generating outputs.

What are AlaMD opportunities and challenges?

As AIaMD gains momentum, it brings forth opportunities, accompanied by challenges that demand careful consideration.

It is widely acknowledged that AIaMD can have a positive impact by enabling:

  • Early Disease Detection: The ability to swiftly analyse vast datasets and identify subtle patterns that might escape human perception. Enabling early detection of diseases, allowing more timely interventions and improved patient outcomes.
  • Accurate Diagnostics: The use of advanced algorithms with heightened precision has the potential to reduce the margin of error, enabling healthcare professionals to make accurate diagnostics.
  • Personalised Therapeutics: The ability to simultaneously analyse all individual patient data allows for tailored treatment plans that may consider a wide range of inputs such genetic, lifestyle, environmental factors, patient history, and test results. This offers potential for enhancing treatment effectiveness while minimising adverse effects.
  • Learning, Adaptation, and Improvement: The capability of continuously learning from new data and adapting to evolving medical knowledge. This iterative learning process enables devices to improve over time and refine their diagnostic and treatment capabilities.

Conversely, there are challenges:

  • Fit-for-Purpose Development: Ensuring AIaMD is both effective and safe requires rigorous development processes in specific medical contexts and patient populations.
  • Minimisation of Bias: Biases present in training data can lead to skewed AI outcomes, perpetuating disparities in healthcare. Addressing and minimising these biases is imperative to ensure equitable healthcare provision.
  • Oversight for Adaptive Systems: AI devices that can learn and adapt autonomously pose challenges in terms of regulatory oversight. Monitoring and controlling the evolution of these systems, to ensure patient safety while fostering innovation is a delicate balance.
  • Transparency to Users: Explaining AIaMD decisions to healthcare professionals and patients can be challenging. The "black box" nature of some AI models can hinder trust and acceptance.
  • Interpretability: Complex AI algorithms can generate results that are difficult for clinicians and patients to interpret. Ensuring outputs are understandable and actionable is essential for effective utilisation.

What are 6 must-have properties of an AIaMD?

AIaMD has several key properties to ensure its effectiveness, safety, and ethical use within the healthcare domain. Here are 6 must-have properties with recommendations around how to approach each within the FDA guiding principles framework



As AI continues to integrate with medical devices, the key lies in navigating these opportunities and challenges with a commitment to patient safety, ethical considerations, and regulatory adherence. Striking a balance between innovation and responsible implementation is essential to harness the full potential of AIaMD for the betterment of healthcare.

Contact us to discuss AIaMD.

Get in touch

By submitting your details you agree to us holding the personal data you've supplied for the purpose of processing your enquiry. For information about how we handle your data, please read our privacy policy