90%
Agreement with the physician's report
Hundreds of real reportsLaiBench v3.10
Platform overview
One layer, one clinical flow
For radiologistsDictation, structuring and review, without switching windows
For clinics and networksLive dashboard: output, TAT and critical findings with SLA
EnterprisePACS/RIS integration, template governance and audit
For institutionsThe answer for leadership, clinical staff, IT and quality.
Laudos.AI AgentNEWThe integration layer: you speak, the report lands in your RIS
Historical production snapshot
Fixed snapshot from April 9 to May 9, 2026 · does not measure total TAT or clinical quality. Sample and definitions are available in the methodology.
Evidence
Evidence with scope and limits
Fidelity in a controlled set
90% score in the v3.10 snapshot of 120 cases. It is not external clinical validation.
Try it in your real routine14 days free or up to 30 reports
Your exams, your templateon your own shift, not a demo environment
No credit cardno automatic charge
After the trial, if it makes sensePro: R$ 219/month, up to 1,000 reports
The platform
A new layer of intelligence for whoever reports.
From dictation to the signed report back in the RIS: AI, classification at the right point and governed critical findings in a single experience.
REPORT — Dictate the way you reason
The editor built for radiology — not transcription: the AI grasps clinical context, organizes your reasoning and writes the structured report while you keep looking at the image.
GUIDE — Classifications right where they belong
BI-RADS, LI-RADS, PI-RADS and the rest show up with the criteria in view, at the point in the report where you need them.
CRIT — Governed critical findings
Critical information at the right time: identification, physician confirmation and structured communication under SLA — every step recorded on the exam.
LaudAI — Structured and integrated reporting
Every report comes out exactly in your house style and goes straight back to your RIS/PACS. Less inconsistency, less rework, no unnecessary steps.
FILM · 77 SECONDS
The post-image layer in 77 seconds
Speak and the template fills in. Describe and the classification is there. Edit by voice, report multiple exams in one recording and leave a trail in the workflow, with CFM Resolution 2.454/2026 along the way.
ASSISTIVE AI THE GOVERNED FLOW
You gain speed.
The report gains a standard.
You need to cut repetitive tasks and spend more time on diagnosis, without giving up your institution's standard. In production, that is close to 10 radiologist hours recovered per 100 reports — and it is the same platform that gives the time back and shows where it went.
Context ready before the first click −
SOURCES
Worklist · indication · priors
ORIGIN
RIS / PACS, with no manual upload
Live structuring, review visible throughout +
Critical findings and audit trail +
AUDIO · CONFIGURABLE PROCESSING
DATA · DISTINCT LIFECYCLES
Data flow documented for each deployment
Audio, transcription, draft, final report, metadata, logs, and backups are distinct categories. Transmission, persistence, and retention depend on the feature and contracted configuration.
REGION AND SUBPROCESSORS · DOCUMENTED
→
CT
Dr. Marina Duarte
Chest CT · 09:30
EXAM CONTEXT
Non-contrast chest CT to investigate a chronic cough of 3 months. Former smoker, 20 pack-years. May radiograph with a right upper lobe opacity for correlation. 2024 prior in the PACS; indication and history imported from the RIS.
3 sources · PACS/RIS
Dictated findings
LIVE
✓
Ground-glass opacity
Right upper lobe, 8 mm
✓
Mediastinal lymph nodes
Calcified
✓
2024 prior
Stable in size
✓
No effusion or pneumothorax
Osseous structures preserved
REC 12:04
❚❚ PAUSE
-
BEFORE THE EXAMThe radiologist interprets
You keep your natural flow: look at the image, decide and dictate — no commands, no memorized phrases.
-
DURING DICTATIONThe AI keeps up
It structures, organizes and accelerates: the report assembles in your house style in real time while your eyes stay on the image.
-
AFTER THE SIGNATUREThe institution follows along
You review and sign — and indicators, quality and performance stay visible, with an auditable trail. More productive physicians, smarter institutions, better-served patients.
REAL CASES
The dictation as you give it.
The report that comes out.
Running speech, with pauses, mid-sentence corrections and on-call slang. On the right, the text that reaches you to review, edit and sign.
DICTATED SPEECH
“Look, there are findings suggestive of pulmonary embolism, yes. It is saddle there, so central PE, right? And besides that I am seeing an aerated consolidation here in the posterior basal segment of the right lower lobe, about 6.3 centimeters. Since it is in the setting of PE, I think it could be an infarction.”
“And wow, this interventricular septum is bulging too, that could even point to right heart strain. At the thoracoabdominal transition just signs of a prior cholecystectomy and bilateral calyceal microcalculi, about two on each side, up to 0.3, non-obstructive. That is all, all right?”
GENERATED REPORT · CHEST CT ANGIOGRAPHY
FINDINGS
Pulmonary trunk and main pulmonary arteries: vascular filling defect extending through the bifurcation of the pulmonary trunk and the main pulmonary arteries (saddle pulmonary embolism).
Cardiac chambers: bulging of the interventricular septum, indicating signs of right heart strain.
Lung parenchyma: consolidation with air bronchograms in the posterior basal segment of the right lower lobe, extending 6.3 cm, suggestive of pulmonary infarction.
Thoracoabdominal transition: prior cholecystectomy. At least two calyceal microcalculi on the right and two on the left, measuring up to 0.3 cm, non-obstructive.
IMPRESSION
Signs of acute central (saddle) pulmonary embolism, with right heart strain and consolidation in the right lower lobe consistent with pulmonary infarction.
CRIT
Acute central pulmonary embolism flagged as a critical finding, for recorded review and communication.
Examples of dictation and drafts generated on the platform, with identifying data removed. The AI is assistive: the radiologist provides the findings, reviews, edits and signs. See the interactive demonstration.
Science and compliance
Artificial intelligence is only the beginning.
The goal is to organize the post-imaging layer with fewer operational steps and explicit controls. The physician provides the findings, reviews, edits, validates, and signs; the data flow is documented for each deployment.
Test dateJuly 2026
Version testedLaiBench v3.10
Number of reportsHundreds of real cases
Who signsThe physician, every time
v4Pulmonary noduleSource cited
v2Carotid stenosisSource cited
v1Breast densityUnder review
What changed in the review
3.2 cm nodule in the right upper lobe
3.8 cm nodule in the right upper lobe
Reviewed and signed by the physician2nd version of the report
14:38
CFM 2.454/2026
LGPD · ANPD
Rule announced
1st week
100% compliant
COMPARISON
The usual workflow and the post-image layer, criterion by criterion
By what changes in the routine, not by the feature list. The Laudos.AI numbers are the ones measured in production and described in the methodology.
The manual workflow time comes from the peer-reviewed literature cited in the references on this page. See the comparisons by criterion →
FIG. 01 · LAIBENCH
Fidelity measured in a controlled internal set.
The v3.10 snapshot records a 90% score across 120 cases. It is a fidelity benchmark, not external clinical validation.
See LaiBench
laudos.ai
Manifesto
Everything that comes after the image.
Artificial intelligence does not replace specialists — it lets them do more. The future of radiology is not human versus machine: it is physicians and technology together for better medicine.
DEPLOYMENT
From first contact to the real flow
-
01A demo in your context
Your exam types, your institution’s template, your flow. No slides, straight into the product.
-
02A pilot with the team
Radiologists using it in a real routine, with the institution’s templates, before any decision is made.
-
03Assisted integration
PACS, RIS and worklist connected. The signed report goes back into your system.
Security documentation ready for IT and DPO review: architecture, encryption, subprocessors and data flow.
WHO USES IT
Radiologists who leave the shift
with less friction.
“ Laudos.AI helps me move faster on the cases I know well and describe less familiar findings with more structure. I cannot imagine working without it.
Dra. Stephanie A.MSK radiologist · CRM-SP 192818“ I use Laudos.AI for my Dopplers. The precision in structuring the findings is what stands out most.
Dr. Bernardo HimeVascular surgeon · CRM-SP 192561“ As a first-year resident, I do not know what my routine would be without it. I dictate a finding and get back a structured description that makes sense.
Dr. Petrus ParaísoRadiology resident“ The reporting assistant has helped me a lot, not only in speeding up reports but also in feeling more confident that I am conveying the information clearly. I particularly like the diagnostic impressions it generates.
Dra. TabitaRadiologist · Fortaleza, Brazil“ Your platform created a real disruption. We are not talking about a simple transcription tool, but about an outstanding way to organise your thinking and structure the report.
Dr. Fábio BortoliniRadiologist“ I can report every exam within the shift itself. I no longer take any work home. And it prevents rework too.
Dr. Carlos IlhaRadiologist · Paraná, Brazil
DIRECT QUESTIONS
Before you
book
Who is the author of the report? −
The radiologist, every time. The AI is assistive: it structures and proposes. Review, editing and signature are the physician’s.
Is my data used to train models? +
Where does the data live? +
Does it integrate with my PACS and RIS? +
And our own vocabulary? +
Do I need to change my workflow? +
The pilot maps templates, permissions, connectors, and contingency. The degree of workflow change depends on the current environment and the modules adopted.
How are critical findings handled? +
CRIT is in a controlled pilot.
- Detection depends on the findings provided by the physician.
- Confirmation, channels, recipients, timeframes, and closure are defined and tested for each deployment.
What is the regulatory status? +
A medium-risk assistive application under CFM Resolution 2,454/2026. It is not SaMD; it does not autonomously interpret images, diagnose, sign, or release reports. The physician provides the findings, reviews, edits, validates, and signs in the official system. See the control matrix.
How does deployment work? +
Within days your team is up and reporting.
- Workflow mapping (PACS/RIS, modalities, and templates).
- A demo with synthetic cases.
- A pilot with baseline metrics.
- Monitored production with support and auditing.
What if the AI is wrong or unavailable? +
Incorrect suggestions are to be edited or discarded by the physician. During unavailability, the institution's contingency plan applies; we do not presume every workflow continues without impact. The public status shows only dated checks.
How do the trial and the subscription work? +
You try it for 14 days or up to 30 reports, no card and no automatic charge. If it makes sense to continue, Pro costs R$ 219/month for up to 1,000 reports — close to R$ 0.22 per report. See the details at /precos.
WHO IT SERVES
Different roles,
explicit controls.
Use scenarios, not testimonials. The radiologist remains responsible for interpretation and the report; the institution governs deployment, access, integration, and contingency.
RADIOLOGIST
From dictation to a structured draft
Provides findings and retains full control over review, editing, validation, and signing.
CLINICAL MANAGEMENT
Templates and governance
Defines standards, permissions, indicators, and contingency criteria for the local context.
IT AND DPO
Verifiable deployment
They assess connectors, data flows, subprocessors, access, retention, and evidence before production.
REFERENCES
- European Society of Radiology (ESR). ESR paper on structured reporting in radiology. Insights Imaging. 2018;9(1):1-7. doi:10.1007/s13244-017-0588-8
- Ringler MD, Goss BC, Bartholmai BJ. Syntactic and semantic errors in radiology reports associated with speech recognition software. Health Informatics J. 2017;23(1):3-13. doi:10.1177/1460458215613614
- Hawkins CM, Hall S, Hardin J, Salisbury S, Towbin AJ. Prepopulated radiology report templates: a prospective analysis of error rate and turnaround time. J Digit Imaging. 2012;25(4):504-511. doi:10.1007/s10278-012-9455-9
- Larson DB, Towbin AJ, Pryor RM, Donnelly LF. Improving consistency in radiology reporting through the use of department-wide standardized structured reporting. Radiology. 2013;267(1):240-250. doi:10.1148/radiol.12121502
FROM THE BLOG
Written by people who report
L
What the CBR's PADI accreditation assesses in a diagnostic imaging service, how the technical review of images and reports works, and what to prepare in operations before applying.
Live dashboard · SLA
PADI accreditation: what the CBR standard demands of the report, and what that demands of operations
BY NATAN · QUALITY
Does your AI reporting software need Anvisa registration? How to read RDC 751 before signing
BY NATAN · COMPLIANCE
How to read Brazil's medical demography and Radiology Atlas data without generalising, and how to build the capacity, queue and backlog maths for your own imaging service.
TAT ↓
Exam demand and radiologist supply: the maths each service has to do for itself
BLOG · MARKET
Liability for a report produced with AI support: what CFM Resolution 2,454/2026 requires of the physician, what the contract splits between service and vendor, and which record holds up the signature in an audit.
laudos.ai
+
CRIT
Who answers for a report signed with AI support
BLOG · COMPLIANCE
OPERATIONAL LIBRARY · CFM 2.454/2026
The question is proving the AI was used correctly.
CFM Resolution 2.454/2026 comes into force on 26 August 2026 and demands evidence produced inside the workflow, not internal policy. It is 16 pages, six playbooks and 36 implementation steps, sourced only from official standards: CFM, LGPD, ANPD, ACR, IHE, HL7, DICOM and NIST.
01
Auditable record
What has to be recorded per exam, and in what format, to hold up in an audit.
02
Demonstrable clinical supervision
How to turn medical review into evidence, instead of a statement of intent.
03
Preserved medical authorship
Who signs, what changed after the suggestion and how that is recorded.
NEXT STEP
More intelligence for every report.
Start reporting with REPORT, GUIDE and CRIT in your real routine.
Individual plan: 14 days or 30 reports, no card.
More control across the entire operation.
Bring REPORT, GUIDE, CRIT, structured data and an audit trail to your institution.
Demo with synthetic data and pilot design.
Report faster.
FAQ
Questions and answers
Individual plan: 14 days or 30 reports, no card.
- What it is
- The post-imaging layer of Brazilian radiology: from dictation to the structured report, back into the RIS and PACS.
- Who it is for
- Radiologists who report and institutions: clinic, hospital and teleradiology.
- What it measures
- Median of 52 seconds from editor open to signature; 54.6% of reports under one minute; close to 10 hours returned per 100 reports.
- What it does not do
- It does not interpret images, does not diagnose, and does not sign or release reports autonomously.
- In summary
- Structured reporting from dictation to signature, back into the RIS and the PACS, with an audit trail per exam; the AI does not sign in the physician's place.
Figures published on this page
- Median from editor open to signature
- 52 s
- April 9 to May 9, 2026 · 5,200 finalized reports
- Metrics methodology
- Resolution 2.454/2026
- In force from August 26, 2026
- Published on February 27, 2026
- Operational Library for CFM 2.454
Product
What is Laudos.AI?
Laudos.AI is the post-imaging layer of Brazilian radiology: it turns dictation into a structured report and returns it to the RIS and the PACS, with a median of 52 seconds from editor to signature, measured in production.
- For radiologists who report and for clinics, hospitals and teleradiology.
- The AI is assistive: it structures and proposes; review, editing and signature are the physician's.
What does structured report mean at Laudos.AI?
Structured reporting, at Laudos.AI, refers to the report text organized into fixed sections, with findings, measurements and classifications in the right fields, generated from free dictation. This means that the physician dictates the way they reason, the AI organizes the text, and interpretation stays with the physician. Product
What are the platform's modules?
REPORT in production; GUIDE and LaudAI in beta; CRIT in a controlled pilot; Agent and the institutional dashboard in beta.
- REPORT: the editor that structures the report from dictation.
- GUIDE: BI-RADS, LI-RADS, PI-RADS and other classifications at the point of the report.
- CRIT: critical findings with identification, physician confirmation and recorded communication.
- LaudAI: the report in the institution's standard, back into the RIS and PACS.
Does the platform interpret images?
No: interpretation belongs to the physician. The platform structures the text from the dictated findings and proposes classifications.
- The physician provides the findings; the AI structures the text and proposes classifications.
Does the AI sign the report?
No: the physician always signs. The AI structures the text and proposes; signature and release stay in the official system.
- The platform does not interpret images or make autonomous diagnoses.
- The physician provides the findings, reviews, edits, validates, and signs in the official system.
What is a governed critical finding?
A governed critical finding refers to a result that has to be communicated within a deadline, with the physician in control: CRIT asks for the radiologist's confirmation, communicates with an SLA per level and records each step in the exam's audit trail. As stated on the product page, CRIT is in a controlled pilot. Product
How does the flow from dictation to the RIS work?
How it works, in five steps: the worklist, the indication and prior exams arrive from the RIS and the PACS with no manual upload; the radiologist dictates; the AI structures the text and proposes classifications; the physician reviews, edits and signs in the official system; the report returns to the RIS and the PACS with no retyping. According to the integrations page, that return is validated per deployment. Integrations
Does it integrate with my PACS and my RIS?
Yes: through API, HL7, DICOM and the Agent, with the report's return validated in your deployment.
- Compatibility depends on the system, version, connector, and contracted scope.
- Return to the RIS or PACS is validated in each deployment, so there is no universal integration promise.
What is RadCommons?
RadCommons is defined as Laudos.AI's repository of classification systems: 121 versioned systems, with cited sources and served over an API. In the editor, the category is shown with the criterion alongside it and the versioned reference; the radiologist validates it. RadCommons
Pricing and deployment
What does it cost?
R$ 219 per month on the individual plan, with up to 1,000 reports a month, close to R$ 0.22 per report; institutions receive a proposal sized by exam volume and active modules.
- 14-day or 30-report trial, no card.
- No hidden per-user fee.
Is there a trial before subscribing?
Yes: 14 days or 30 reports, whichever comes first, with no card and no automatic charge.
- Up to 5 of your own templates during the trial.
How long does deployment take?
Up to two weeks, in three phases, without replacing the PACS or the RIS.
- The pilot maps templates, permissions, connectors and contingency.
Does it work for clinics, hospitals, and teleradiology?
Yes, and also for the radiologist who reports alone: an individual plan with public pricing and an institutional proposal sized by exam volume and active modules.
- Radiologist reporting solo: individual plan with a public price and a new-user trial with no card required.
- Clinic, hospital, and teleradiology: a proposal by exam volume and active modules, with integration validated for each deployment.
Bottom line: what does Laudos.AI deliver?
In summary: structured reporting from dictation to signature, back into the RIS and the PACS, with governed critical findings, operational data and an audit trail per exam: measured in production, with the physician signing every time. Product
Figures
Where does the 52-second figure come from?
52 seconds is the median, measured in production, from opening the editor for an exam to signing the report.
- Fixed historical snapshot: April 9 to May 9, 2026.
- Sample: 5,200 completed reports.
- It does not measure total service TAT, clinical quality, causality, or future performance.
What is the difference between median and mean reporting time?
The 52-second median is the typical report; the mean, between 3 and 5 minutes, includes the complex cases. Laudos.AI publishes both.
- 54.6% of reports come in under one minute in the same measurement.
- We publish median and mean side by side so the tail is not hidden.
Who measures the figures published on this page?
Laudos.AI itself, in production, with a public methodology. According to the methodology page, the window is fixed, April 9 to May 9, 2026, with 5,200 finalized reports; according to the same page, the figure measures neither the total TAT of the service nor clinical quality. Metrics methodology
What is TAT, and how does it differ from the 52 seconds?
TAT, also known as turnaround time, refers to the total time from the exam being performed to the report being released, and it is shown on the institutional dashboard; the 52 seconds measure only the stretch from editor open to signature, the part the radiologist performs inside the platform, because that is where the platform acts. This means the two figures are neither added together nor compared. Metrics methodology
How do you prove that quality holds up?
With a published benchmark: LaiBench v3.10, a fidelity score of 90% on 120 cases from a controlled internal set, on June 23, 2026.
- It is a fidelity benchmark, not external clinical validation.
- The radiologist remains responsible for interpretation and the report.
What is LaiBench?
LaiBench can be described as Laudos.AI's fidelity benchmark, published with the math on the table: it compares the text structured by the AI against the signed report in a controlled internal set. According to the LaiBench page, v3.10, from June 23, 2026, scored 90% on 120 cases. It is not external clinical validation, because the set is internal and controlled. LaiBench
How do I compare Laudos.AI with other reporting platforms?
By three measures that Laudos.AI publishes and measures in production: time from editor to signature, return to the RIS and the PACS with no retyping, and an audit trail per exam.
- Time from open editor to signature, measured in production.
- Report return to the RIS and PACS without retyping.
- Per-exam audit trail and compliance with CFM Resolution 2,454/2026.
Compliance and data
What changes with CFM Resolution 2.454/2026?
Laudos.AI is classified as an assistive application of medium risk, not as Software as a Medical Device; human supervision, medical authorship and traceability are already part of the product.
- Its controls support human oversight, physician authorship, and traceability.
- Final compliance also depends on the deployment context and the responsibilities of the institution and physician.
When does CFM Resolution 2.454/2026 take effect?
Published on 27 February 2026, it takes effect on 26 August 2026.
- The application is assistive, classified as medium risk, and is not Software as a Medical Device.
What does CFM Resolution 2.454/2026 require of a reporting AI?
Meaningful human supervision, medical authorship and traceability, according to the CFM 2.454 Operational Library kept on this site. Laudos.AI is classified as an assistive application of medium risk, not as Software as a Medical Device, which is why the AI neither signs nor releases reports and each exam keeps an audit trail. Operational Library for CFM 2.454
Where is exam data stored?
Each type of data follows a flow documented per deployment, defined by the feature and the contracted configuration.
- Audio, intermediate transcription, draft, final report, logs, and backups have distinct flows and retention periods.
- Some subprocessors may process data outside Brazil.
Does patient data train models?
No: identifiable care data does not train general models.
- The processing needed to deliver the contracted function is a separate purpose and is recorded in the deployment documentation.
Is there an audit trail per exam?
Yes: each exam produces audit records proportional to the contracted function, validated before production.
- Before production, deployment validates the trail's events, content, retention and availability.
Try it in your real routine
14 days free or up to 30 reportsNo credit card
Content updated on .