Executive summary
The use of AI for medical diagnosis in the NHS is rapidly moving from research pilots to regulated, real-world practice. However, its adoption is governed by a strict and maturing framework. Any tool that contributes to a diagnosis is likely to be classified as a medical device (SaMD/AIaMD) and must adhere to MHRA rules, local clinical safety standards (DCB0129/0160), and the national procurement baseline, the DTAC (GOV.UK, NHS England Digital, NHS Transformation Directorate).
To accelerate the safe adoption of promising technologies, NICE is increasingly using its Early Value Assessment (EVA) pathway. This allows for the time-limited, conditional use of new AI tools—such as those for fracture detection on X-rays and skin lesion triage—in the NHS while further real-world evidence is generated (NICE). This guide provides a practical map of the regulatory pathways, highlights the diagnostic AI that is already delivering value in NHS pathways, and outlines the essential governance that must be in place for safe deployment.
Definitions and scope: AI “diagnosis” vs decision support
It is critical to distinguish between AI that provides decision support and AI that performs a diagnostic function.
- An AI tool becomes a Software as a Medical Device (SaMD) or AI as a Medical Device (AIaMD) in UK law when its intended purpose is to provide information for diagnostic or therapeutic purposes.
- An assistive tool that helps a clinician detect a potential finding (e.g., highlighting a possible fracture on an X-ray for a radiographer to review) is different from an autonomous tool that provides a definitive diagnostic output. This distinction has significant implications for the device's risk classification, the level of evidence required, and the necessary clinical oversight (GOV.UK, NICE).
The UK regulatory pathway at a glance (“how AI reaches patients”)
- MHRA (Medicines and Healthcare products Regulatory Agency): The MHRA regulates all medical devices in the UK. Its software and AI change programme is continually updating the rules. All devices must have a UKCA mark (though valid CE marks are accepted for a transitionary period) and are subject to new, stricter post-market surveillance duties (GOV.UK).
- AIDRS (AI & Digital Regulations Service): This is a "one-stop shop" portal, led by NICE in partnership with the MHRA, CQC, and HRA. It provides a unified, practical route-map to help both developers and NHS adopters navigate the complex regulatory and evaluation landscape (digitalregulations.innovation.nhs.uk, NICE).
- NICE evidence routes: For innovative technologies, NICE increasingly uses its Early Value Assessment (EVA) pathway. This allows for conditional recommendation and time-limited use in the NHS, with a requirement for the manufacturer and adopters to generate further real-world evidence. This sits alongside its classic Medical Technologies Guidance (MTG) and Medtech Innovation Briefings (MIB) (NICE).
- Clinical safety & procurement: The supplier of any health IT must provide a DCB0129 clinical safety case and hazard log. The adopting NHS organisation must complete a local DCB0160 safety case. For procurement, any tool must pass the DTAC baseline, which covers clinical safety, data protection, cybersecurity, interoperability, and usability (NHS England Digital, NHS Transformation Directorate).
- Data protection: The Information Commissioner's Office (ICO) provides specific guidance for the use of AI, covering fairness, Data Protection Impact Assessments (DPIAs), and transparency (Information Commissioner's Office).
What NICE currently says about AI for diagnosis
NICE has already evaluated several AI technologies for diagnostic use, providing a clear picture of where the evidence is strongest.
Fracture detection on X-rays in urgent care (EVA HTE20, Jan 2025)
NICE has conditionally recommended that AI tools such as TechCare Alert, BoneView, and others can be used in the NHS to help identify fractures on X-rays, provided they are used alongside clinician review and as part of a plan to generate further evidence (NICE).
Skin-lesion triage for the urgent suspected cancer pathway (EVA HTE24, May 2025)
In another EVA, NICE has recommended that certain AI technologies may be used to assist in the assessment and triage of skin lesions for patients on a two-week-wait pathway for suspected cancer. Again, this adoption is tied to strict evidence generation plans (NICE).
Coronary CT: HeartFlow FFRCT (MTG32)
HeartFlow FFRCT has a long-standing positive NICE recommendation (MTG32). It uses AI to analyse coronary CT angiograms to assess for coronary artery disease, improving diagnostic accuracy and potentially avoiding the need for invasive angiography (NICE).
Stroke decision-support (DG57 evidence update)
In contrast, for AI tools that assist in stroke decision-support, NICE has concluded that there is currently insufficient evidence to recommend their routine use in the NHS. Further research is required before these tools can be adopted (NICE).
MHRA’s AI Airlock: sandboxing novel diagnostic AI
For truly novel diagnostic AI that pushes the boundaries of current regulations, the MHRA AI Airlock provides a crucial pathway. Launched in May 2024 and expanded with a second cohort in 2025, this regulatory sandbox allows developers to test their AIaMD in a supervised, real-world NHS environment. The learnings from the Airlock are used to inform and shape future UK regulations (GOV.UK, Digital Health).
Where AI helps today in NHS diagnostic pathways
- Imaging triage & detection: Primarily in fracture detection and chest/CT use-cases, currently under the NICE EVA pathway.
- Dermatology triage: Assisting in the prioritisation of patients on the two-week-wait pathway, also under EVA.
- Cardiac CT analysis: The established use of HeartFlow FFRCT in the stable chest pain pathway. It is important to note that some high-profile domains, such as stroke decision-support, are still considered to be in the evidence-generating phase and are not yet recommended for routine use.
Deployment playbook for NHS organisations
- Regulatory status: Confirm the device's class, its UKCA/CE mark, and its MHRA registration. Check if its use falls within the conditions of a NICE EVA.
- Evidence & value: Map the tool's claims to the relevant NICE guidance (MTG/EVA) and create a local benefits realisation and audit plan.
- Safety assurance: Obtain the supplier's DCB0129 safety case. Appoint a Clinical Safety Officer (CSO) and complete your local DCB0160 safety case.
- Procurement baseline: Ensure the vendor has a passed DTAC pack, covering data protection, cyber security, and interoperability standards (e.g., FHIR/HL7).
- Data protection: Complete a full DPIA, ensuring transparency and data minimisation in line with ICO AI guidance.
- Post-market monitoring: Establish local KPIs and a clear process for incident reporting, aligning with the resources on the AIDRS portal.
Risks, limits and how to mitigate them
- Over-reliance and automation bias: Mitigate by enforcing dual-reading or mandatory clinician oversight as specified by NICE guidance for EVA technologies.
- Generalisation & bias: Mitigate by scrutinising the representativeness of the training datasets, a key requirement flagged in NICE EVAs.
- Governance drift: Mitigate with periodic re-validation of the AI's performance, strict software update controls, and ongoing maintenance of your DCB0160 hazard log.
Quick comparison table
| Use-case | NICE Status | Typical Setting | Oversight Required | Evidence Notes |
|---|---|---|---|---|
| Fracture detection (X-ray) | EVA HTE20 | Urgent Care / ED | Clinician review mandatory | Use during evidence generation |
| Skin-lesion triage (2WW) | EVA HTE24 | Dermatology Referral | Within EVA protocol | Equity and follow-up data required |
| Coronary physiology (CCTA) | MTG32 | Cardiology / Imaging | MDT use | Proven cost & diagnostic benefits |
| Stroke decision-support | DG57 (Insufficient) | Stroke Networks | Specialist oversight | Further research required |
FAQs
- Is AI “diagnosis” allowed in the NHS?
- Yes, but only when it is appropriately regulated as a medical device by the MHRA, has a sufficient evidence base as assessed by NICE (often via an EVA), and is deployed with local DCB/DTAC/ICO governance.
- What’s the fastest route for a new AI diagnostic tool into practice?
- The NICE EVA pathway is designed to enable time-limited, conditional use in the NHS while the required real-world evidence is generated.
- Do CE-marked AI tools still qualify for use in the UK?
- For many devices, a valid CE mark is acceptable for placing a product on the Great Britain market for a transitionary period (check the latest MHRA updates), but it must still be registered with the MHRA.
- Where can NHS teams get end-to-end guidance on this?
- The AI & Digital Regulations Service (AIDRS) portal is the official "one-stop shop" for both adopters and developers.