What Does It Mean When a Diagnostic Device Is “AI-Enabled”?
An AI-enabled diagnostic device is a piece of software or hardware that uses machine learning to help produce a health-related output — a score, a flag, or a reading — instead of relying only on fixed rules a person wrote. The U.S. Food and Drug Administration (FDA) requires transparency about how these outputs are produced, but that information isn’t always visible on the screen where you see the result.
Imagine a connected otoscope, a skin-imaging app, or a wearable sensor produces a result labeled “AI algorithm” or “smart analysis.” The screen shows something confident-looking: a score, a flag, a phrase like “no concern detected.” It’s easy to treat that as a finished answer. It’s better understood as a starting point for a conversation with a healthcare provider, not a diagnosis.
What Is a Machine Learning-Enabled Medical Device?
The FDA calls software of this kind a Machine Learning-Enabled Medical Device, or MLMD. In FDA guidance, “transparency” means how clearly a device’s intended use, development, performance, and — where available — internal logic are communicated to the people who use it or are affected by it.
Two terms matter for reading a result:
- Logic — how the device reached its output or the basis for its decision.
- Explainability — how well that logic can be described in a way an ordinary person can follow. Not every device can fully explain its own logic in plain terms, and that limitation itself is something worth knowing.
The FDA also maintains a public list of AI-enabled devices authorized for marketing in the United States. That list is explicitly not comprehensive — it’s built by scanning marketing-authorization summaries for AI-related terms, so a device can use AI and still not appear on it, and the list is only updated periodically.
What Six Questions Should You Ask Before Trusting an AI Diagnostic Output?
FDA transparency guidance organizes the information a device should communicate around who, why, what, where, when, and how. Turned into reader questions, they form a practical checklist for any AI-enabled result you’re handed.
1. Who is this device actually for?
Transparency information is meant to reach everyone involved in a health decision — not just the person using the device, but also the person receiving care with it, and anyone else who might act on the result, such as a caregiver or clinician. If you can’t tell who the output is written for, that’s a gap.
2. What is the device’s stated intended use?
A clear, accurate device description should explain its medical purpose, the condition or function it addresses, and its intended users, use setting, and target population. An output that doesn’t match the device’s stated intended use — for example, a screening tool being read as a diagnostic one — is a mismatch worth noticing.
3. How does the output fit into a health decision?
Good practice is for a device to describe whether its output is meant to inform a clinician’s judgment or, in some cases, stand in for part of it. If a device doesn’t say whether its output is advisory or closer to conclusive, don’t assume the answer.
4. What are the device’s known limits?
FDA guidance specifically calls out communicating limitations as good practice, including known biases or failure modes, confidence intervals attached to outputs, and known gaps in the training or testing data — for instance, patient groups underrepresented during development who may see more biased results. It also includes situations where a person’s real-world data simply won’t match what the device was built and validated on.
5. Is the underlying evidence available anywhere?
The guidance describes it as good practice for a device maker to characterize its training and testing data, summarize relevant clinical studies, and provide ongoing updates on performance monitoring and any issues identified over the device’s lifecycle. None of this needs to appear on the device screen itself — but it should exist somewhere accessible, such as labeling, a user manual, or a public database entry.
6. Is the information current?
Devices can be updated, and their performance can shift over time. Good practice includes providing timely notification when a device is updated or modified, or when new information comes to light. A device with no visible update history or last-reviewed date makes this harder to check.
When Should You Slow Down Before Acting on an AI-Generated Result?
If the output conflicts with how you feel, don’t let a “normal” or low-risk reading override a real, ongoing symptom — every device has gaps in what it was tested on. If you can’t find any information about the device’s limitations, treat that absence as a gap in disclosure, not as a clean bill of health.
If a device presents its output as final rather than advisory, treat the result as informational only until a clinician reviews it — the intended role of the output should be clear, and if it isn’t, caution is warranted. If you have urgent or worsening symptoms, no home device output, AI-enabled or not, is a substitute for emergency care; contact local emergency services right away.
What’s the Decision Path Before You Act on an AI Diagnostic Output?
- If you receive an AI-generated output, then read the device’s stated intended use first and confirm the output category (screening, monitoring, informational) matches what you assumed.
- If you can’t find a stated limitation or bias disclosure, then look for one in labeling, a manufacturer FAQ, or an included insert before assuming there isn’t one.
- If the device doesn’t show a review or update date, then check whether that date is tied to the software or algorithm version, not just the app version number.
- If the result affects a real health decision, then bring the raw output — not just your interpretation of it — to a healthcare provider, so they can weigh it against your history and exam findings.
- If the “why” behind an output is unclear or unexplained, then treat that as a limitation in itself, not proof the device is wrong, and seek a second source of information before acting.
What This Guide Does Not Cover
This guide explains how to evaluate the transparency of an AI-enabled device’s output. It does not identify, evaluate, recommend, or rank any specific device, app, or manufacturer, and it is not a substitute for professional medical advice, diagnosis, or treatment. Always talk to a qualified healthcare provider about a specific health concern or device result. If you are experiencing a medical emergency, call your local emergency number immediately.
Frequently Asked Questions About AI-Enabled Diagnostic Devices
Is an AI health app result the same as a medical diagnosis?
No. An AI-generated output from a health app or device is a computer-generated estimate based on the inputs it received and the data it was built on. It is not a diagnosis, and FDA guidance frames these outputs as information meant to support — not replace — a clinician’s judgment.
How can I tell if an AI-enabled device has been reviewed by the FDA?
The FDA publishes a list of AI-enabled medical devices that have received marketing authorization in the United States. That list isn’t complete, since it’s compiled by scanning authorization summaries for AI-related terms, so a device can be authorized without appearing on it yet.
What if an AI device’s output doesn’t match how I feel?
Trust your symptoms over a reassuring output. Every device has known and unknown limits in the data it was trained and tested on, and a “normal” reading doesn’t rule out a real, ongoing problem. Bring both the symptom and the device’s output to a healthcare provider.
Why won’t some AI devices explain how they reached a result?
Some machine learning models are more “explainable” than others — meaning their internal logic can be described in terms a person can follow. FDA guidance treats explainability as a goal, not a guarantee, so some devices may only be able to share their output, not the reasoning behind it.
Should I stop using a device if I can’t find its limitations listed anywhere?
Not necessarily, but treat that gap as a reason for extra caution rather than reassurance. Communicating known limitations is described as good practice by the FDA, so their absence may reflect incomplete disclosure rather than a device with no limitations at all.
Sources
To understand how these images or sensor readings are produced in the first place, see Start Here. This guide is based on two FDA guidance documents: Transparency for Machine Learning-Enabled Medical Devices: Guiding Principles and the FDA’s AI-Enabled Medical Devices list. For more on how this publication vets sources like these, see How We Research.
Everyday Imaging Evidence is an independent educational publication and is not affiliated with, and does not continue, endorse, or represent, any medical device manufacturer past or present. This article was last reviewed for accuracy in September 2026.
By Everyday Imaging Evidence Editorial Team
Leave a Reply