Executive summary
In 2025, the conversation around AI for medical diagnosis in the NHS is moving decisively from pilots to governed, large-scale deployments. For any tool to be considered "approved" or "vetted," it must navigate a clear and rigorous pathway. This includes holding the correct UKCA/CE marking as a medical device, aligning with NICE guidance (often through a Medical Technologies Guidance or Early Value Assessment), and meeting the NHS’s baseline procurement standard, the DTAC.
This guide provides a practical, specialty-by-specialty map of the regulated, evidence-backed AI tools that are already in active use across NHS pathways. We will explore real-world examples, from the established, cost-saving HeartFlow FFRCT in cardiology and Annalise CXR in radiology, to the innovative Mia® for breast screening and Skin Analytics' DERM for dermatology triage. While tools for stroke AI are in widespread use, they remain under evidence-generation conditions as per NICE's DG57, underscoring the non-negotiable need for human oversight.
What counts as “approved” in the UK (and why it matters)
The regulatory spine for regulated clinical AI UK is built on four key pillars:
- UKCA/CE Marking (MHRA): Any Software or AI as a Medical Device (SaMD/AIaMD) must have the appropriate device classification and marking to be placed on the market in Great Britain.
- NICE pathways: The National Institute for Health and Care Excellence provides evidence-based recommendations through its Technology Guidance (MTG), Diagnostics Guidance (DG), and, for promising but early-stage tech, its Early Value Assessment (EVA) pathway.
- DTAC (Digital Technology Assessment Criteria): This is the NHS’s mandatory procurement baseline, ensuring any new tool meets stringent standards for clinical safety, data protection, security, interoperability, and usability.
- AIDRS (AI & Digital Regulations Service): This "one-stop shop," led by NICE and involving the MHRA, CQC, and HRA, provides a unified route-map for both developers and NHS adopters to navigate the complex regulatory landscape.
For NHS buyers, the takeaway is clear: prefer tools with a clear device class, traceable evidence that aligns with a NICE pathway, and proven conformance with the DTAC.
Fast map: live NHS AI by specialty
Cardiology
HeartFlow FFRCT is a long-standing success story for AI in the NHS. Recommended by NICE in MTG32, it uses AI to create a 3D model of the coronary arteries from a standard CT scan, providing a non-invasive fractional flow reserve (FFR) analysis. NICE's cost modelling found it could deliver a £391 per-patient saving compared to traditional functional imaging by avoiding the need for more invasive tests.
Radiology – chest imaging & lung cancer pathways
Annalise CXR is a prime example of a tool scaled through national funding. Supported by the NHS AI Diagnostic Fund (AIDF), it provides an AI "second read" of chest X-rays to help prioritise findings. It is now available across more than 40 NHS Trusts and six imaging networks, including Greater Manchester and the North East and North Cumbria (Digital Health).
Breast screening
Mia® (from Kheiron Medical) is a CE-marked Class IIa AI tool for breast screening. It is being deployed in NHS settings to act as a second reader for mammograms, with case studies from trusts like East Midlands Radiology Consortium (EMRAD) showing a potential uplift in cancer detection rates. An Aberdeen trial reported it helped find approximately 12% more cancers than the routine double-reading process (The Royal College of Radiologists, NHS England Digital).
Dermatology
Skin Analytics' DERM is a UKCA Class IIa marked AI medical device used to triage skin lesions in the urgent suspected cancer pathway. It is now live in multiple NHS deployments. The tool allows a clinician to take a dermoscopic image of a lesion, and the AI assesses whether it is benign (allowing for safe discharge from the pathway) or requires urgent review by a dermatologist. This is a leading example of an AI skin cancer triage tool being used at scale in the NHS (Skin Analytics, PMC).
Stroke (decision support & imaging logistics)
AI tools from providers like e-Stroke, RapidAI, and Viz.ai are in widespread use across the UK, with 99 of 107 English stroke units having access to AI support by 2023. However, it is critical to note that under NICE DG57, these technologies are still in an evidence-generation phase. NICE supports their use to help interpret brain scans and inform decisions about thrombectomy, but only alongside continued clinician decision-making and as part of ongoing data collection.
Pathology
The Paige Prostate Suite is another example of a tool being evaluated through a major national initiative. It is part of an NHS AI in Health and Care Award evaluation across leading NHS labs in Oxford, Coventry, and North Bristol, with a real-world deployment study currently underway to validate its performance in a UK setting (Oxford University Hospitals).
Long-term conditions / community
Healthy.io provides a smartphone-based urine albumin-to-creatinine ratio (ACR) testing service to help identify chronic kidney disease earlier. An NHS case study has documented its successful use at scale to support home-based albuminuria testing, improving access and adherence for at-risk populations (NHS England Digital).
Procurement & assurance playbook
- Step 1 – Regulatory status: Confirm the tool's medical device class, its UKCA/CE marking, and its MHRA registration. Check if its use is aligned with the conditions of any relevant NICE guidance (e.g., an EVA).
- Step 2 – Evidence & value: Map the tool's claims to the NICE evidence standards and create a local benefits realisation and audit plan.
- Step 3 – Safety assurance: Obtain the supplier's DCB0129 clinical safety case. Appoint a Clinical Safety Officer (CSO) and complete your local DCB0160 safety case.
- Step 4 – Procurement baseline: Ensure the vendor has a passed DTAC pack, covering data protection, cybersecurity, and interoperability.
- Step 5 – Data protection: Complete a full Data Protection Impact Assessment (DPIA) and ensure the tool's use aligns with ICO AI guidance.
- Step 6 – Post-market monitoring: Define your local KPIs, establish a clear process for incident reporting, and align with the resources available on the AIDRS portal.
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 (Evidence-generating) | Stroke Networks | Specialist oversight mandatory | Further real-world data required |
FAQs
- Is there a single list of “approved” NHS AI tools?
- No. The UK uses a multi-layered assurance system. You should use the AIDRS portal, check for NICE guidance (MTG/DG/EVA), ensure the tool has passed the DTAC, and look at the NHS AI Knowledge Repository for real-world case studies.
- Are stroke AI tools fully recommended by NICE?
- No. Under DG57, they are recommended for use with evidence generation. This means you must maintain full clinician oversight and participate in local or national data collection to evaluate their ongoing performance.
- What’s demonstrably cost-saving today?
- HeartFlow FFRCT has a formal, per-patient cost-saving model published by NICE. Other tools, particularly in ambient scribing and triage, are demonstrating strong operational efficiency gains in local evaluations.