Radiology AI Directory

FDA-cleared profile

AutoContour (Model RADAC V4)

Radformation, Inc.

AutoContour Model RADAC V4 assists radiation treatment planners with contouring and reviewing structures in medical images for radiation therapy treatment planning. It generates initial machine-learning contours for user review and modification rather than independently approving contours or treatment plans.

Radiotherapy planning
Evidence status: each field states its source quality, applicability, and review date. Research pending, information not established, and vendor documentation pending remain distinct outcomes.

Clinical details

What this tool is for

Start with the authorized purpose, then verify how it fits your service line and reading workflow.

Exact purpose
AutoContour Model RADAC V4 assists radiation treatment planners with contouring and reviewing structures in medical images for radiation therapy treatment planning. It generates initial machine-learning contours for user review and modification rather than independently approving contours or treatment plans.
  • FDA source
  • Exact FDA submission
  • Checked 2026-09-02
Intended users
Radiation treatment planners qualified to select, review, edit, and export contours. Local governance should assign responsibilities across radiation oncologists, dosimetrists, medical physicists, radiation therapists, and clinical IT.
  • FDA source
  • Exact FDA submission
  • Checked 2026-09-02
Care setting and population
Professional radiation oncology treatment-planning workflows using DICOM-compatible imaging and treatment-planning systems.
  • FDA source
  • Exact FDA submission
  • Checked 2026-09-02
Workflow role
Receives CT or MR image sets, generates initial structure contours, supports manual or automatic rigid registration and automatic deformable registration, presents contours for review and editing, and exports DICOM RTSTRUCT, REGISTRATION, and DOSE objects for downstream treatment-planning review.
  • FDA source
  • Exact FDA submission
  • Checked 2026-09-02
Required input
DICOM CT or MR images for contouring or registration and fusion; PET/CT for registration or fusion only; and DICOM RTSTRUCT and REGISTRATION objects as supported inputs.
  • FDA source
  • Exact FDA submission
  • Checked 2026-09-02
Output and human action
User-reviewed and editable contours plus DICOM RTSTRUCT, REGISTRATION, and DOSE files for downstream review in a treatment-planning or independent registration-QA system. The FDA summary states that the device has no built-in reporting.
  • FDA source
  • Exact FDA submission
  • Checked 2026-09-02
Limitations
Every generated contour and registration must be reviewed and edited as needed before planning use. K242729 reports manufacturer nonclinical validation and explicitly states that no clinical studies were conducted for V4. Aggregate DSC and reviewer averages do not replace structure-level review: some model-level lower confidence bounds fall below the nominal size-category threshold, and small structures and very small external test sets remain important limitations. The summary describes training anatomy from adult male and female patients, so pediatric, postsurgical, unusual-anatomy, positioning, contrast, and local-guideline performance require local validation.
  • Public source
  • Exact FDA submission
  • Checked 2026-09-02

Regulatory identity

FDA record and catalog context

The FDA listing establishes the regulatory identity. It does not by itself establish local workflow fit, pricing, security, or performance in your environment.

FDA submission
K242729
  • FDA source
  • Exact FDA submission
  • Checked 2026-08-31
Modality
Radiotherapy Planning Imaging
  • FDA source
  • Exact FDA submission
  • Checked 2026-09-02
Anatomy
Multi-region
  • FDA source
  • Exact FDA submission
  • Checked 2026-09-02

The FDA summary lists head and neck, thorax, abdomen, and pelvis models.

Clearance type
510(k)
  • FDA source
  • Exact FDA submission
  • Checked 2026-08-31
Decision date
2024-12-09
  • FDA source
  • Exact FDA submission
  • Checked 2026-08-31
FDA status
FDA-cleared
  • FDA source
  • Exact FDA submission
  • Checked 2026-08-31

Implementation

Questions for IT, informatics, and operations

Use these fields to structure a vendor demo, security review, and implementation estimate.

Integration
The FDA-cleared architecture uses DICOM data, a Windows .NET client, a Windows agent that can monitor network storage for new CT or MR datasets, and a Linux-compatible cloud contouring service. V4 can export DICOM deformable REGISTRATION and deformed DOSE objects for review in external treatment-planning and registration-QA systems. Current product-family materials also describe direct Eclipse read/write and standalone DICOM workflows.
  • FDA source
  • Exact FDA submission
  • Checked 2026-09-02
Deployment and data flow
K242729 describes a Windows .NET client and local Windows agent connected to a Linux-compatible cloud automatic-contouring service. Current product-family materials also offer cloud-based or local compute resources, but the V4 summary does not establish that every current deployment option applies to this cleared version.
  • FDA source
  • Exact FDA submission
  • Checked 2026-09-02
Security and privacy
Radformation's current license terms state that cloud-processing configurations transmit selected CT or MR images, structures, plan, and dose data after software anonymization, encrypt the data in transit, and delete the imaging-plan data after returning results. The customer remains responsible for preventing PHI disclosure. The reviewed public sources do not establish which current controls apply to an installed V4 environment or describe its authentication, at-rest encryption, retention logging, vulnerability, backup, and incident-response controls.
  • Vendor supplied
  • Product family
  • Checked 2026-09-02

Obtain version-specific security and support evidence and validate anonymization, identity, access, encryption, logging, deletion, recovery, and breach-response behavior.

Training and support
Radformation's current support terms include online educational materials, videos, and a remote training session with new access. NICE states that trained professionals must always review and edit AI autocontours and maintain manual contouring skills. Local V4 training should cover model selection, contouring-guideline alignment, review and editing, unusual-anatomy failure modes, registration QA, DICOM routing, downtime, and escalation.
  • Vendor supplied
  • Product family
  • Checked 2026-09-02
Monitoring and change control
Monitor performance by exact software and model version, anatomy, structure, scanner and protocol, patient position, contrast, unusual or postsurgical anatomy, reviewer role, edit time, unusable or missing contours, DICOM transfer failures, registration and dose-export QA, incidents, and adverse events. NICE calls for ongoing error and adverse-event reporting, while independent AutoContour studies show that geometric averages can conceal clinically important outliers and contour-definition mismatches.
  • Public source
  • Product family
  • Checked 2026-09-02

Define signed acceptance thresholds and revalidation triggers for software, model, scanner, protocol, routing, or local contouring-guideline changes.

Exact cleared model expansion

K242729 adds 77 CT and 18 MR contouring models to the V3 predicate across head and neck, thorax, abdomen, and pelvis and retains user review and editing.

exact submission · checked 2026-09-02

Create an enabled-model inventory tied to the installed build, FDA labeling, local anatomy, local contour definitions, and acceptance status.

Source

DICOM and treatment-planning integration

The exact FDA architecture supports CT and MR contouring, PET/CT registration or fusion, DICOM RTSTRUCT and REGISTRATION inputs, and RTSTRUCT, REGISTRATION, and DOSE exports. Current product-family materials describe direct Eclipse and standalone DICOM workflows.

exact submission · checked 2026-09-02

Test each scanner, TPS, service class, object mapping, frame of reference, naming rule, routing path, retry, and failure-recovery workflow.

SourceSource

Human review and local commissioning

The device produces initial contours for review and modification. NICE and independent studies support model-by-model commissioning, trained review, retained manual skills, and special attention to small structures, unusual anatomy, contrast, positioning, and local contouring conventions.

product family · checked 2026-09-02

Commission with representative local cases and predefine accept, edit, reject, and manual-fallback criteria for each enabled model.

SourceSourceSourceSource

Cloud imaging-data governance

Current terms describe software anonymization, encrypted transmission, processing by Radformation, return of results, and deletion of imaging-plan data, while making the customer responsible for preventing PHI disclosure.

product family · checked 2026-09-02

Confirm the terms and controls that apply to the installed V4 environment and validate DICOM private tags, burned-in annotations, logs, support access, deletion evidence, and exception handling.

Source

Economic evaluation

The vendor uses quote-based licensing. NICE found autocontouring economics depend on technology cost, professional grade, review and edit time, and true net time saved, with published time endpoints varying substantially.

product family · checked 2026-09-02

Model costs per completed plan and measure total hands-on and elapsed time, edit burden, rejected contours, downstream rework, throughput, and avoided overtime.

SourceSourceSource

Lifecycle and upgrade planning

Current support terms condition support on current releases and installed updates. V4 users should obtain explicit support status, security-update commitments, migration scope, data portability, regression requirements, and rollback terms.

product family · checked 2026-09-02

Treat migration to a later model family as a controlled clinical change rather than assuming prior acceptance transfers.

Source

Evidence

Performance evidence

Metrics are shown only when they are tied to a source, endpoint, population, and tested product version. Candidate literature without exact product and version linkage is not shown as product evidence.

Evidence summary
V4 has an exact FDA technical-validation package and independent product-family evidence from earlier software versions. The evidence supports assisted contour generation with mandatory review; it does not establish autonomous use, universal generalizability, reduced toxicity, improved tumor control, or other patient outcomes.
  • Public source
  • Exact FDA submission
  • Checked 2026-09-02
Reported sensitivity
Not applicable
  • Not applicable
  • Not applicable
  • Checked 2026-09-02

Sensitivity is not an applicable endpoint for this contour-generation device. Review the structure-specific DSC, distance, clinician-review, and local edit-burden evidence instead.

Reported specificity
Not applicable
  • Not applicable
  • Not applicable
  • Checked 2026-09-02

Specificity is not an applicable endpoint for this contour-generation device. Review the structure-specific DSC, distance, clinician-review, and local edit-burden evidence instead.

New CT models versus RADAC V3

77 models

  • EndpointNew CT contouring models added in RADAC V4 relative to its V3 predicate
  • PopulationHead and neck, thorax, abdomen, and pelvis structures listed in K242729
  • Tested versionAutoContour Model RADAC V4 cleared in K242729
Source

New MR models versus RADAC V3

18 models

  • EndpointNew MR contouring models added in RADAC V4 relative to its V3 predicate
  • PopulationBrain and pelvis MR structures listed in K242729
  • Tested versionAutoContour Model RADAC V4 cleared in K242729
Source

External CT mean DSC, small structures

0.76 DSC

  • EndpointMean Dice similarity coefficient across externally reviewed small CT structures
  • PopulationExternal CT datasets described in the K242729 validation summary
  • Tested versionAutoContour Model RADAC V4 cleared in K242729
Source

External CT mean DSC, medium structures

0.84 DSC

  • EndpointMean Dice similarity coefficient across externally reviewed medium CT structures
  • PopulationExternal CT datasets described in the K242729 validation summary
  • Tested versionAutoContour Model RADAC V4 cleared in K242729
Source

External CT mean DSC, large structures

0.94 DSC

  • EndpointMean Dice similarity coefficient across externally reviewed large CT structures
  • PopulationExternal CT datasets described in the K242729 validation summary
  • Tested versionAutoContour Model RADAC V4 cleared in K242729
Source

External CT reviewer average

4.57 points on 1-5 scale

  • EndpointAverage clinical-expert rating across CT structure models; 5 meant no edits and 1 meant full manual re-contouring
  • PopulationExternal CT review image sets in the K242729 validation summary
  • Tested versionAutoContour Model RADAC V4 cleared in K242729
Source

External MR mean DSC, small structures

0.61 DSC

  • EndpointMean Dice similarity coefficient across externally reviewed small MR structures
  • PopulationExternal MR brain and pelvis datasets described in K242729
  • Tested versionAutoContour Model RADAC V4 cleared in K242729
Source

External MR mean DSC, medium structures

0.84 DSC

  • EndpointMean Dice similarity coefficient across externally reviewed medium MR structures
  • PopulationExternal MR brain and pelvis datasets described in K242729
  • Tested versionAutoContour Model RADAC V4 cleared in K242729
Source

External MR mean DSC, large structures

0.8 DSC

  • EndpointMean Dice similarity coefficient across externally reviewed large MR structures
  • PopulationExternal MR brain and pelvis datasets described in K242729
  • Tested versionAutoContour Model RADAC V4 cleared in K242729
Source

External MR reviewer average

4.6 points on 1-5 scale

  • EndpointAverage clinical-expert rating across MR structure models; 5 meant no edits and 1 meant full manual re-contouring
  • PopulationExternal MR review image sets in the K242729 validation summary
  • Tested versionAutoContour Model RADAC V4 cleared in K242729
Source

Older-version mean contouring time saved

36.6 minutes

  • EndpointMean time saved across the tested breast, head-and-neck, lung, and prostate contour sets after correction
  • PopulationTiming subset of three cases from an 80-patient, five-vendor evaluation
  • Tested versionAutoContour v1.0.25.0 tested in April 2022; not RADAC V4
Source

Older-version average physician score

1.96 points on 1-5 scale

  • EndpointAverage physician score for Radformation contours; lower scores indicated fewer required changes
  • PopulationSixteen organs from 47 patients, reviewed by at least three physicians
  • Tested versionAutoContour v2.2.8; not RADAC V4
Source

K242729 AutoContour Model RADAC V4 verification and validation

  • DesignManufacturer nonclinical validation using held-out training-test partitions, external CT and MR datasets, size-stratified DSC criteria, and clinical-expert Likert review
  • PopulationCT and MR structures across head and neck, thorax, abdomen, and pelvis, with CT data from four institutions in the United States and Switzerland and additional MR sources
  • Scopeexact submission
  • Tested versionAutoContour Model RADAC V4 cleared in K242729
  • IndependenceSponsor or vendor study

K242729

The FDA summary states that no clinical studies were conducted. External testing was separate from training, but the evidence was manufacturer generated and some structure-level confidence bounds were below category thresholds.

Source

A clinical evaluation of the performance of five commercial artificial intelligence contouring systems for radiotherapy

  • DesignIndependent single-center multi-vendor benchmark against manually drawn expert contours with geometric metrics and correction-time measurement
  • Population80 patients: 20 breast, 20 head and neck, 20 lung, and 20 prostate; 45 structures
  • Scopeproduct family
  • Tested versionAutoContour v1.0.25.0 tested in April 2022
  • Samplen=80
  • Sites1
  • IndependenceIndependent study

PMID 37601695

The authors reported no commercial or financial conflicts and acknowledged vendor support and guidance. Timing estimates used a three-patient subset and cannot be transferred to V4 or another site.

Source

Evaluation of multiple-vendor AI autocontouring solutions

  • DesignIndependent single-center blinded physician review and geometric comparison of three commercial CT autocontouring systems against approved physician contours
  • PopulationSixteen organs, ten patients per organ, 47 distinct patients, and at least three physician reviewers per contour
  • Scopeproduct family
  • Tested versionAutoContour v2.2.8
  • Samplen=47
  • Sites1
  • IndependenceIndependent study

PMID 38822385

The study found generally comparable performance but documented poor results with unusual anatomy and local contour-definition differences. It informed implementation across five facilities, but the exact relationship between v2.2.8 and RADAC V4 was not established.

Source

Evaluation and failure analysis of four commercial deep learning-based autosegmentation software for abdominal organs at risk

  • DesignIndependent retrospective multi-vendor geometric evaluation with explicit outlier and failure-mode analysis
  • Population111 abdominal cases evaluating liver, stomach, and kidney contours
  • Scopeproduct family
  • Tested versionAutoContour v1.7.11
  • Samplen=111
  • Sites1
  • IndependenceIndependent study

PMID 39946266

Radformation outliers included liver spill into heart or stomach, incomplete stomach contours with barium, and kidney differences driven by whether the renal pelvis was included. The authors reported no conflicts.

Source

Costs and setup

Budget and ongoing governance

These are common procurement questions; unknown values remain visible until a source supports them.

Pricing and total cost
No public list price for AutoContour Model RADAC V4 was found on the reviewed product page. Radformation's current license agreement places fees in a customer quote, treats pricing as confidential, and permits additional charges for special support. A V4 support or upgrade quote should separate licensing, implementation, compute, storage, network needs, training, support, upgrade, renewal, and end-of-support terms.
  • Vendor supplied
  • Product family
  • Checked 2026-09-02
Reimbursement and coding
No separate named-product Medicare payment for AutoContour was identified in the reviewed CMS Radiation Oncology Model material. Economic value should be evaluated within the radiation-therapy planning service line using net staff time saved after review and editing, implementation and recurring costs, throughput effects, rework, and downstream plan-quality measures.
  • Public source
  • Not applicable
  • Checked 2026-09-02

No product-specific code

No separate named-product payment identified in the reviewed CMS Radiation Oncology Model page

Medicare · United States

This is service-line payment context, not a comprehensive coding, coverage, or procurement determination.

Source

Safety and lifecycle

Postmarket record

Recall and adverse-event records are shown only after product matching. Adverse-event reports do not establish incidence or causality.

Postmarket safety review
The 2026-09-02 openFDA device-recall snapshot contained no record matched to K242729 by exact submission identifier.
  • Public source
  • Exact FDA submission
  • Checked 2026-09-02

A zero-result exact-identifier search does not prove that no recall, correction, MAUDE report, or other safety signal exists under a model, catalog, software-version, or product-family identifier. Maintain ongoing FDA and vendor surveillance.

Buyer worksheet

Open questions to take to the vendor

Open evaluation checklist

Research record

What has been checked

This audit distinguishes completed source review from fields that have not yet been researched.

Human reviewedStatus
2026-09-04Last searched
24Fields reviewed
12Source classes checked
20Awaiting review PubMed leads
0Awaiting review trial leads
0Unreviewed FDA recall leads

Recovered from the prior exact-submission extraction and normalized under the current provenance rules. The exact FDA summary, current product and license terms, NICE implementation and economics guidance, automated PubMed and ClinicalTrials.gov candidates, three directly relevant independent product-family studies found through broader literature review, CMS payment context, and the exact-identifier FDA recall search were manually reviewed. Current contract price, V4 support status, deployment-specific security package, enabled-model inventory, interoperability, and local acceptance remain organization-specific evidence. Automated exact-name discovery found 20 PubMed and 0 ClinicalTrials.gov candidate records. Candidates require human product and version matching; zero candidates is not evidence that no studies exist. Native FDA recall identifiers produced 0 postmarket candidate records; 0 have been reviewed (0 published, 0 rejected) and 0 remain unreviewed.

Candidate leads remain unpublished until a human confirms the exact product and tested version.

Source classes: fda ai list, fda decision summary, vendor product page, vendor legal documentation, peer reviewed publication, health technology guidance, trial registry, reimbursement policy, fda device recall, pubmed, clinical trials, openfda device recall

Sources & history

How this profile was documented

Open the ledger for source dates, scope, and research-record updates.

View source ledger and history

Source ledger

  1. K242729 AutoContour Model RADAC V4 510(k) summary

    U.S. Food and Drug Administration - Published 2024-12-09 - Accessed 2026-09-02

    Scope: purpose, modality, anatomy, intendedUsers, careSetting, workflowRole, input, output, limitations, integration, deployment, evidenceSummary, evidenceMetrics, evidenceStudy, implementationCapabilities

  2. Artificial Intelligence-Enabled Medical Devices

    U.S. Food and Drug Administration - Accessed 2026-08-31

    Scope: regulatoryIdentity, modalityContext

  3. PubMed biomedical literature index

    U.S. National Library of Medicine - Accessed 2026-09-01

    Scope: evidenceDiscovery

  4. ClinicalTrials.gov study registry

    U.S. National Library of Medicine - Accessed 2026-09-01

    Scope: evidenceDiscovery, studyStatus

  5. openFDA Device Recall API

    U.S. Food and Drug Administration - Accessed 2026-09-02

    Scope: recallDiscovery, nativeSubmissionMatching

  6. Radiation Oncology Model

    Centers for Medicare & Medicaid Services - Accessed 2026-09-02

    Scope: reimbursement, radiationTherapyPaymentContext

  7. AutoContour v2.7 product page

    Radformation, Inc. - Accessed 2026-09-02

    Scope: productFamily, integration, deployment, pricingResearch, currentCommercialContext, implementationCapabilities

  8. Radformation Software License Agreement and support terms

    Radformation, Inc. - Accessed 2026-09-02

    Scope: security, training, pricing, monitoring, dataGovernance, implementationCapabilities

  9. Artificial intelligence technologies to aid contouring for radiotherapy treatment planning

    National Institute for Health and Care Excellence - Published 2023-09-27 - Accessed 2026-09-02

    Scope: limitations, training, pricing, evidenceSummary, economics, monitoring, implementationCapabilities

  10. A clinical evaluation of the performance of five commercial artificial intelligence contouring systems for radiotherapy

    Frontiers in Oncology - Published 2023-08-08 - Accessed 2026-09-02

    Scope: evidenceSummary, evidenceStudy, evidenceMetrics, economics, implementationCapabilities

  11. Evaluation of multiple-vendor AI autocontouring solutions

    Radiation Oncology - Published 2024-05-31 - Accessed 2026-09-02

    Scope: evidenceSummary, evidenceStudy, evidenceMetrics, implementationCapabilities

  12. Evaluation and failure analysis of four commercial deep learning-based autosegmentation software for abdominal organs at risk

    Journal of Applied Clinical Medical Physics - Published 2025-02-13 - Accessed 2026-09-02

    Scope: limitations, evidenceSummary, evidenceStudy, evidenceMetrics, monitoring, implementationCapabilities

Research history

  1. K242729 FDA decision

    Regulatory record

    510(k)

  2. Source-review record updated

    Research record

    human reviewed

Alternatives

Related tools to compare

FDA-cleared K260509

AutoContour (RADAC V5)

Radformation, Inc.

AutoContour Model RADAC V5 assists radiation treatment planners with contouring and reviewing structures in medical images for radiation therapy treatment planning. It produces...

Modality
Radiotherapy Planning Imaging
Anatomy
Multi-region
Decision
2026-03-19
Radiotherapy planning
FDA-cleared K253091

ART-Plan+ (v3.1.0)

Therapanacea Sas

ART-Plan+ assists trained radiation-therapy professionals with multimodal image review, registration, contouring, and the release-specific planning functions described in K253091....

Modality
Radiotherapy Planning Imaging
Anatomy
Multi-region
Decision
2025-12-23
Breast imagingRadiotherapy planningProstate imaging
FDA-cleared K242748

Oncospace

Oncospace, Inc.

Oncospace supports trained radiation therapy professionals with protocol configuration, machine-learning achievable-dose prediction, planning-objective exchange, and plan review...

Modality
Radiotherapy Planning Imaging
Anatomy
Multi-region
Decision
2025-04-11
Radiotherapy planning