What is AI auto-contouring in radiotherapy and does it save time?
AI auto-contouring uses deep learning to outline tumours and nearby healthy organs on planning CT scans, a step that manual contouring can take one to four hours per case. Published studies report large time savings, often 30% to 70%, with comparable geometric accuracy, but a clinician must still review and correct every outline.
Radiotherapy plans must deliver dose to the tumour while sparing organs at risk (OARs) such as the spinal cord, parotid glands, heart or rectum. Before a plan can be calculated, someone has to draw those outlines slice by slice. That work is slow, varies between clinicians and can delay the start of treatment.
This page explains how AI contouring works, what recent real-world and multicentre studies show, where it falls short, and what patients should ask when they are planning radiotherapy in India, China or elsewhere.
Key facts from radiotherapy contouring studies
Published results, each from a different design:
- Manual contouring of organs at risk: typically one to four hours per case (Communications Medicine, 2025)
- Head and neck, single centre (2026): contouring time for model-covered structures fell 48.6% (21.8 to 11.2 minutes), and 30.0% overall (35.3 to 24.7 minutes) once all required organs were included
- Brain organs at risk (in-house model, eight experts from four institutions): evaluating and adjusting AI contours averaged 22 minutes versus 69 minutes for manual contouring
- Multicentre real-world evaluation (British Journal of Radiology, 2026): 80% of contour sets needed no edits or only minor edits; median time to start contouring was 0 days versus 3 days
- Network of 18 oncology centres in a low- and middle-income country (Frontiers in Oncology, 2026): 116 planning cases rated by six physicians; significant clinician time and labour cost savings reported
- Seven-institution head and neck study (npj Digital Medicine, 2025): every clinic saved contouring time, but not all structures benefited equally
- Lung radiotherapy review: 24 studies reported significant time reduction with comparable accuracy to manual contours
Key takeaways: AI auto-contouring in radiotherapy planning
- AI auto-contouring consistently saves clinician time in studies, from about 30% to roughly two-thirds depending on the site and workflow.
- Most AI contours needed no or only minor edits (80% in one real-world evaluation), but every outline still needs expert review.
- Benefits vary by structure and clinic: some lymph node levels showed no significant time saving in a seven-institution head and neck study.
- Faster contouring can mean planning starts sooner: the median wait to start contouring fell from 3 days to 0 in one multicentre evaluation.
- No study reviewed shows that AI contouring improves tumour control or survival; its demonstrated benefits are time, consistency and workload.
How does AI auto-contouring work in radiotherapy planning?
Radiotherapy starts with a planning CT, sometimes combined with MRI or PET. Deep learning models trained on thousands of expert-drawn outlines automatically segment organs at risk and, in some products, target volumes. The clinician reviews the result, edits errors and approves the contours, after which physicists and dosimetrists calculate the treatment plan. Common uses include breast cancer, prostate cancer, lung cancer and head and neck tumours.
Earlier atlas-based tools matched a new patient to a library of previous cases. Deep learning contours are reported to be comparable in accuracy to manual contours and faster than atlas-based methods: a review of lung radiotherapy studies found time savings of 18% to 19% compared with atlas-based contours, and 24 studies reported significant time savings against manual contouring.
How much time does AI contouring save? Evidence from 2025 and 2026 studies
Time savings are consistent but vary with site and workflow. In a 2026 single-centre head and neck study, a deep learning prototype cut contouring time for the structures it covered by 48.6% (from 21.8 to 11.2 minutes), and by 30.0% overall (35.3 to 24.7 minutes) once structures it did not cover were included. A 2025 framework paper evaluating an in-house brain model found that evaluating and adjusting AI contours averaged 22 minutes against 69 minutes for manual work. A 2026 comparison in 30 patients across head and neck, abdomen and pelvis reported 5 to 10 minutes with AI versus about an hour manually.
These figures include human editing, which is the realistic comparison. The overall saving is smaller than the saving on covered structures because clinicians still draw whatever the model does not handle, so the benefit depends on how many of the required structures a product supports.
What did multicentre and real-world studies show about AI contouring?
A multicentre real-world evaluation in the British Journal of Radiology in 2026 looked at AI-assisted organ-at-risk contouring across whole treatment-planning pathways. Across four main tumour sites, 80% of contour sets needed no edits or only minor edits. The median time to start contouring was 0 days for AI-assisted cases compared with 3 days for manual contouring, and while every manual contour review involved clinical oncologists, 38% of AI contour reviews did, showing that work moved between staff groups.
A two-phase multicentre head and neck study across seven institutions in npj Digital Medicine found that every clinic benefited from reduced contouring time, but benefits varied by structure and by clinic; several lymph node levels showed no significant difference between manual contouring and editing AI output. A 2026 study across 18 oncology centres in a low- and middle-income country, using 116 planning cases rated by six physicians, reported meaningful clinician time and labour cost savings.
Key real-world results for AI contouring at a glance
Reported in the 2026 multicentre evaluation and related studies:
- Contour sets needing no or only minor edits: 80% across four tumour sites
- Median time to start organ-at-risk contouring: 0 days AI-assisted versus 3 days manual
- Contour reviews involving clinical oncologists: 38% (AI) versus 100% (manual)
- Head and neck single-centre time reduction: 48.6% for covered structures, 30.0% overall
- Brain organ-at-risk evaluation time: 22 minutes versus 69 minutes for manual contouring
Is AI auto-contouring accurate and safe for radiotherapy patients?
Accuracy is usually measured by geometric overlap between AI and expert contours, such as the Dice similarity coefficient, and by whether the dose to organs at risk changes. Reviews report broadly comparable accuracy and inter-observer variability between deep learning and manual contours, and one 2026 study found comparable geometric accuracy while contouring time fell. Framework studies also assess the dosimetric impact on treatment plans.
Safety rests on human review. AI can produce plausible but wrong outlines, particularly in unusual anatomy, after surgery, with implants or artefacts, or where a tumour distorts normal tissue. Quality assurance, local validation before clinical use and clinician sign-off on every plan are therefore essential. Variation between clinics in how benefits arise is a reminder that products should be validated locally.
Can AI auto-contouring reduce radiotherapy waiting times?
Contouring is one of the slower steps between CT simulation and the first treatment, so shortening it can help. In the multicentre real-world study, the median time to start organ-at-risk contouring fell from 3 days to 0, and review effort moved from oncologists toward other staff. In low- and middle-income settings with few radiation oncologists, the 18-centre study argues that recovered clinician time becomes significant when institutions produce thousands of plans a year.
Waiting time to the first radiotherapy fraction also depends on machine availability, physics checks, patient fitness and referral delays, so faster contouring alone will not clear a backlog. Patients waiting for radiotherapy can ask how long planning takes at their centre and whether AI tools are used.
Limits of the evidence for AI contouring
The time-saving evidence is consistent, but its scope is narrower than product marketing may suggest.
- Many studies are single-centre, retrospective or exploratory, and some test prototype software
- Savings depend on the structure, the clinic and how many required structures a product covers
- Studies measure time, geometry and workload; none reviewed shows improved tumour control or survival
- Target volumes, which involve more clinical judgement than organs at risk, are generally harder for AI and need close oversight
Who could ask about AI in their radiotherapy planning?
AI contouring is a back-office tool, but patients and families can still ask informed questions. Situations where it is worth asking:
- You are about to start radiotherapy and want to know how long planning will take
- Your cancer is in the head and neck, brain, chest, abdomen or pelvis, where many organs at risk sit close to the tumour
- You are comparing centres, including abroad, and want to know about their technology and quality assurance
- A planning delay has been explained by staff shortages or contouring workload
- You are considering a second opinion on a radiotherapy plan or target volume
- You want to know who checks and approves the contours before treatment
How does AI contouring compare with manual and atlas-based methods?
Manual contouring is the traditional standard but is slow and varies between clinicians. Atlas-based auto-segmentation, an earlier automated approach, matches a patient to previous cases and often needs heavy editing. Deep learning contouring is generally faster than atlas-based methods and comparable to manual in geometric accuracy, with editing times that depend on the structure.
Human experts remain responsible in every approach. A useful way to think about it is that AI changes who does the drafting and how much is left to correct, not who is accountable. Clinics that adopt AI well measure edit rates, audit outliers and train staff to spot typical AI errors.
Contouring evidence and access by region for AI tools
What has been published, with gaps stated:
- Low- and middle-income country network: AI-Rad Companion Organs RT evaluated across 18 oncology centres and 116 cases (Frontiers in Oncology, 2026)
- Multicentre real-world evaluation: published in the British Journal of Radiology, 2026
- Head and neck: a published comparison of five commercial AI auto-segmentation solutions reported high-quality contours in significantly reduced time
- India and China: we have not verified which hospitals use AI auto-contouring or which products hold local approval
Is AI auto-contouring used in radiotherapy in India and China?
AI contouring products are commercially available and are being evaluated worldwide, including in a multicentre study across a network of centres in a low- and middle-income country. We have not verified which hospitals in India or China use AI auto-contouring, which products they use or how their quality assurance works.
If you are planning radiotherapy abroad, ask the centre about its planning workflow, whether physicians review all AI outlines, how quickly treatment can start and what quality assurance is in place.
What should patients expect during radiotherapy planning with AI tools?
Patients usually notice nothing different. You lie in the treatment position for a planning CT, sometimes with a mask or immobilisation device, and possibly additional MRI or PET images. Behind the scenes, software drafts outlines of organs and the radiation oncologist defines and approves the tumour volume. Physicists and dosimetrists then design the plan, which is checked before treatment begins.
Patients can ask whether AI was used, who reviewed the outlines and whether the plan was checked independently. These are reasonable questions, and a good centre will answer them. AI contouring does not change how radiotherapy is delivered, the dose prescribed or its side effects; those depend on the cancer, the plan and the patient.
What happens next for AI in radiotherapy planning?
Research is expanding from organs at risk to target volumes, adaptive radiotherapy that reshapes the plan as the tumour changes, and workflow automation with language models. A 2026 review describes large language models being used for clinical summarisation, trial identification and even multimodal target volume contouring in radiation oncology, although these uses are early. Evaluation frameworks such as COMMUTE have been proposed so that clinics can test AI contouring systematically on geometry, expert ratings, time and dose rather than on a single measure.
For patients, the trend should mean shorter planning times and more consistent outlines, provided clinics validate tools locally. Related AI developments, such as AI clinical trial matching and AI pathology risk tools, also aim to speed decisions without removing specialists.
Questions to ask your radiation oncologist about AI and treatment planning
- Does your centre use AI to outline organs at risk or tumour volumes, and who reviews the outlines?
- How long will planning take before my first treatment, and is anything delaying it?
- How are plans checked for safety before I start?
- Are there features of my case, such as surgery or implants, that make automated outlines less reliable?
- Would a second opinion on the radiotherapy target and dose be helpful?
- Are there clinical trials of new radiotherapy approaches for my cancer?
Sources
Where the figures in this guide come from: