Radiation_oncology · Deep-learning auto-contouring for radiotherapy planning
DLCExpert
Mirada Medical
DLCExpert was the first FDA-cleared AI auto-contouring product, and it remains one of the most widely evaluated. Its evidence, though, is exactly the kind a contouring tool tends to have: retrospective studies of geometric agreement with expert contours and of the downstream dose impact, not prospective trials of patient outcomes. Independent work — including a five-vendor head-to-head in which Mirada is one system among several rather than a standout — supports it as a clinically usable time-saver that still requires a clinician to review and correct every structure. England's health-technology assessor reached the same posture, judging the outcome evidence for AI contouring as a class immature and recommending use alongside evidence generation. One mark reflects a genuinely cleared, broadly adopted tool whose published evidence is real but geometric and retrospective, not outcome-level.
Performance Metrics
Clinical Evidence
DLCExpert's evidence base is characteristic of auto-contouring software: it measures how closely the software's contours match expert-drawn ones, and how much the differences would change a dose plan, rather than whether the tool changes what happens to patients. Read in that frame, the literature is supportive and, importantly, includes independent work. The foundational validation is Wong and colleagues (Radiotherapy and Oncology 2020, volume 144, pages 152-158), which compared deep-learning auto-segmentation of organs at risk and clinical target volumes against expert inter-observer variability — the honest benchmark for a contouring tool, since experts themselves disagree. The most useful neutral evidence is Doolan and colleagues (Frontiers in Oncology 2023;13:1213068), an independent clinical evaluation that put five commercial AI contouring systems — Mirada among them, alongside MVision, Radformation, RayStation and TheraPanacea — through the same test on 80 patients spanning breast, head-and-neck, lung and prostate cases, scoring geometric similarity with Dice and surface-distance metrics. That Mirada appears as one credible system among several, rather than as a singled-out winner, is the point: the independent comparison neither inflates nor dismisses it. The limiting context is set by England's National Institute for Health and Care Excellence, whose early value assessment of AI contouring technologies (HTE11, 2024) concluded that the evidence for the category is not yet mature enough to establish clinical and cost benefit, and recommended these tools be used only alongside further evidence generation. The catalogue judgment follows that posture: DLCExpert reliably produces a good, consistent contouring draft that saves time, and the standing requirement — never optional — is that a clinician reviews and corrects every structure before it is used to plan a dose.
| Study | Design | n | Sensitivity | Specificity | AUC | Published |
|---|---|---|---|---|---|---|
| Wong J, et al. | RetrospectiveRetrospective | 0 | — | — | — | Radiotherapy and Oncology, 2020;144:152-158; deep-learning auto-segmentation of organs at risk and clinical target volumes benchmarked against expert inter-observer variability |
| Doolan PJ, Charalambous S, Roussakis Y, et al. (independent) | RetrospectiveRetrospective | 80 | — | — | — | Frontiers in Oncology, 2023;13:1213068; independent clinical evaluation of five commercial AI contouring systems (Mirada among them) across breast, head-and-neck, lung and prostate cases using Dice and surface-distance metrics |
Clinical Pulse
Verified clinician sentiment
Verified clinician reviews launching soon · Apply to contribute →
Reviews are signed by clinicians verified against their regulator's public register.
Inside the algorithm
Editorial featureHow DLCExpert drafts a set of contours.
Five stages — from raw input to verdict — drawn from manufacturer documentation and the public regulatory record.
- INGEST
- NORMALISE
- DETECT
- LOCALISE
- VERDICT
Stage 01 · INGEST
A radiotherapy planning CT arrives for contouring.
The input is the planning CT acquired for treatment planning — the same scan the dosimetrist and oncologist use to design the radiation plan. The software takes the volume as acquired; there is no need for a clinician to pre-mark or pre-sort the anatomy before analysis.
Input
Planning CT (DICOM)
Inference
Per-scan, automated
Inside the Auris+ Listing
Five more sections complete this device’s Auris+ Listing.
Decision Ledger
ProPro unlocks a private, cross-vendor log of every case you read on this device — what the AI called, what you concluded, and a one-line reason if you overrode it. Ready for the EU AI Act's deployer logging obligations when they land in 2028.
Clinical Evidence Deep Dive
ProPro unlocks the structured clinical-evidence summary — study count, target patient population, and a tabular accuracy-metrics view drawn from peer-reviewed sources.
Peer-Reviewed Publications
ProPro unlocks the curated peer-reviewed publication list with PubMed cross-links — the citation backbone of every editorial verdict.
Post-Market & Regulatory Conditions
ProPro unlocks the post-market surveillance summary, recall record, and the conditions of approval that bound real-world use.
AI Algorithm Version History
ProPro unlocks the chronological record of algorithm version changes — what changed when, drawn from manufacturer changelogs and regulatory filings.
Regulatory Approvals
Safety Record
No DLCExpert or Workflow Box recalls, field safety notices or FDA safety communications were identified in publicly available sources as of August 2026 — this should be read as "none found in public reporting" rather than an exhaustive vigilance audit, because the FDA MAUDE adverse-event database, the openFDA recall endpoint and the primary 510(k) database could not be queried directly from this environment (the proxy returned policy denials). The clinically relevant risk is intrinsic to auto-contouring rather than a device fault: an automated contour that is plausible but subtly wrong — a slightly mis-drawn organ boundary — propagates into the dose plan if it is accepted unchecked, which is precisely why the software is a review-and-edit draft and not an autonomous step. The organ-sparing benefit and the risk both rest on the clinician correcting the contours before planning.
Intended Use & Indications
DLCExpert is deep-learning auto-contouring software used in the radiotherapy planning workflow. Given a planning CT, it automatically generates contours for organs at risk across common treatment sites and presents them as an editable starting point in the treatment-planning environment. The contours are a draft: a qualified clinician reviews every structure, edits it as needed, and remains responsible for the final contours used to compute the dose plan. The software is a workflow tool intended to reduce the time and inter-observer variability of manual contouring; it does not autonomously delineate the clinical target volume, does not generate or optimise the radiation dose distribution, and does not make any clinical or treatment decision. It ships within Mirada's Workflow Box automation platform.