2 The technologies

2.1 Contouring is an important part of the radiotherapy treatment planning process. It involves outlining the target volumes and organs at risk (OAR) to guide radiotherapy so that treatment is effective and radiation toxicity is minimised. Artificial intelligence (AI) technologies aim to improve contouring efficiency by automatically contouring the OAR and sometimes the target volumes before radiotherapy. The technologies have been trained to process images from CT, cone-beam CT or MRI scans to produce an initial contour. Images and contours are then reviewed by trained healthcare professionals and modified as needed before use.

2.2 NICE has assessed 11 AI technologies to aid contouring for radiotherapy treatment planning. The criteria for including technologies in this assessment are in the final scope in the project documents for this guidance. Nine technologies have regulatory approval for use in the NHS:

  • AI-Rad Companion Organs RT (Siemens Healthineers) is a standalone software that contours over 60 OAR structures on CT scans including the abdomen, head and neck, pelvis and thorax.

  • ART-Plan (TheraPanacea, Oncology Systems; Brainlab) is a standalone software that contours over 150 structures including OAR and lymph nodes in the abdomen, brain, head and neck, thorax and pelvis on CT images and the abdomen, brain and male pelvis on MRI scans.

  • DLCExpert (Mirada Medical) is deployed on Mirada Medical's Workflow Box platform. It contours over 160 structures on CT and MRI images, including the abdomen, breast, head and neck, prostate and thorax.

  • INTContour (Carina Medical) is a standalone software that contours over 60 target and OAR structures from the abdomen, head and neck, male pelvis and thorax.

  • Limbus Contour (Limbus AI, AMG Medtech) is a standalone software that contours over 200 OAR and target volumes including lymph nodes, the abdomen, breast, central nervous system, head and neck, lungs, pelvis and prostate on CT images, and the central nervous system, gynaecological and brachy structures on MRI scans.

  • MIM Contour ProtégéAI (MIM Software) is a standalone software that contours the head and neck, thorax, lungs and liver, prostate and abdomen structures from CT images and the prostate from MRI scans.

  • MRCAT Prostate plus Auto-contouring (Philips) is a clinical application integrated in the Philips MR-RT systems for MRI in radiation therapy. It provides automatic contours of the prostate and associated OAR.

  • MVision Segmentation Service (MVision AI Oy, Xiel) is a standalone software that contours over 160 structures including OAR and target volumes in the abdomen and thorax, brain, breast, head and neck, and pelvis.

  • RayStation (RaySearch) is a radiotherapy external beam and brachytherapy planning system with AI autocontouring functionality. It contours over 70 structures on CT images including the breast and lymph nodes, head and neck, male pelvis, thorax and abdomen.

    Two technologies are awaiting CE or UK Conformity Assessed (UKCA) mark approval so cannot be used yet:

  • AutoContour (Radformation) is a standalone software that contours over 200 structures including OAR and lymph nodes in the chest and abdomen, head and neck, and pelvis on CT images and the brain on MRI scans.

  • OSAIRIS (Cambridge University Hospitals NHS Foundation Trust) is an open-source standalone software that contours up to 26 head and neck and prostate treatment site structures on CT images.

    The Medicines and Healthcare products Regulatory Agency (MHRA) advised that most AI autocontouring technologies will likely be classified as class IIa or higher under the MHRA guidance on software as a medical device. The government has extended the transition period for CE marked devices in the UK.

Care pathway

2.3 The National Cancer Registration and Analysis Service (NCRAS) and Cancer Research UK (CRUK) statistics on chemotherapy, radiotherapy and surgical tumour resections in England reported that between 2013 and 2019, about 40% of people with cancer had radiotherapy as part of their cancer treatment. Clinical experts advised that AI technologies could be used to aid contouring for everyone having external beam radiotherapy.

2.4 Manual contouring is the most common contouring method in standard care. Manual contouring of target regions is usually done by clinical oncologists. Contouring of OAR may also be done by clinical technologists, dosimetrists or therapeutic radiographers. There are guidelines and consensus statements on contouring from organisations such as the European Society for Radiotherapy and Oncology, the Global Quality Assurance of Radiation Therapy Clinical Trials Harmonization Group and the Royal College of Radiologists. Other contouring methods include atlas-based contouring and model-based segmentation, but these are not as widely used. Regardless of contouring method, contours should be reviewed before being used in treatment planning in line with guidance such as the Royal College of Radiologists guidance on radiotherapy target volume definition and peer review.

2.5 AI technologies would be used to provide an initial contour as part of the standard workflow for radiotherapy treatment planning. AI autocontours would then be reviewed by trained healthcare professionals and edited as needed before being used in treatment planning.

The comparators

2.6 Comparators are contouring methods used in standard care to contour OAR and target volumes. These include manual contouring, atlas-based contouring and model-based segmentation. Comparators may also include no contouring for cases when AI technologies may produce contours for structures not routinely contoured in standard care.

  • National Institute for Health and Care Excellence (NICE)