How are you taking part in this consultation?

You will not be able to change how you comment later.

You must be signed in to answer questions

    The content on this page is not current guidance and is only for the purposes of the consultation process.

    EARLY VALUE ASSESSMENT PROGRAMME

    Closed for comments This consultation ended on at   Request commenting lead permission

    1 Recommendations

    1.1 Nine artificial intelligence (AI) technologies can be used in the NHS while more evidence is generated to aid contouring for radiotherapy treatment planning in people having external beam radiotherapy. AI technologies must be used with healthcare professional review of contours.

    The following technologies can only be used once they have Digital Technology Assessment Criteria (DTAC) approval:

    • AI-Rad Companion Organs RT (Siemens Healthineers)

    • ART-Plan (TheraPanacea, Oncology Systems)

    • DLCExpert (Mirada Medical)

    • INTContour (Carina Medical)

    • Limbus Contour (Limbus AI, AMG Medtech)

    • MIM Contour ProtégéAI (MIM Software)

    • MRCAT Prostate plus Auto-contouring (Philips)

    • MVision Segmentation Service (MVision AI Oy, Xiel)

    • RayStation (RaySearch).

    1.2 The technology developers must confirm that agreements are in place to generate the evidence (as outlined in NICE's evidence generation plan) and contact NICE annually to confirm that evidence is being generated and analysed as planned. NICE may withdraw the guidance if these conditions are not met.

    1.3 At the end of the evidence generation period (3 years), the technology developers should submit the evidence to NICE in a form that can be used for decision making. NICE will review the evidence and assess if the technologies can be routinely adopted in the NHS.

    Evidence generation

    1.4 More evidence needs to be generated on the following key outcomes:

    • clinical acceptability of contours and amount of edits needed

    • impact on radiation dose

    • time saving including time for healthcare professional review and edits

    • resource use defined by healthcare professional grade and time

    • adverse events associated with AI autocontouring and contouring errors.

    Potential benefits of early use

    • System benefit: AI technologies may help healthcare professionals to produce contours more quickly. This may make the workflow more efficient. It may also improve the consistency of contours and increase compliance with national and international guidelines.

    • Clinical benefit: Clinical evidence suggests that AI technologies generally produce similar quality contours as manual contouring, with most structures needing only minor edits.

    • Resources: The evidence suggests that AI autocontouring is quicker than manual contouring even when including time for healthcare professional review and edits. This could have potential cost savings. It may also free up healthcare professional time for patient-facing tasks or more complex cases when AI autocontouring may not be appropriate.

    Managing the risk of early use

    • Clinical review: All AI autocontours must be reviewed and edited as needed by a trained healthcare professional before being used in radiotherapy treatment planning.

    • Costs: Potential cost savings depend on technology costs including setup and maintenance, time saving and the healthcare professional grade of the person doing the contouring. With a band 7 radiographer doing the contouring, cost analysis suggests that the highest priced AI technology (£50 per plan) would need to save around 47 minutes to be cost neutral, compared with 4 minutes for the lowest priced technology (£4 per plan). This guidance will be reviewed within 3 years and the recommendations may change. Take this into account when negotiating the length of contracts and licence costs.

    • Information governance: Local NHS hospitals and trusts should have appropriate information governance policies for using AI technologies.

    • Equality: AI models can contain algorithmic bias depending on the population used in training, which may not be representative of populations in clinical practice. This may affect the performance of AI autocontouring for some populations such as children and young people or people with atypical anatomy.