Talking about AI in
Brazilian radiology
Working on a story about assistive AI, radiology reports, CFM 2.454/2026 or data governance? We help with technical and clinical sources.
Quick, verifiable facts
- WHAT IT ISAssistive AI for dictation, structuring and review of radiology reports
- PERFORMANCEMedian of 52 s from editor to signature; 54.6% of reports in under 1 min
- ESTIMATEClose to 10 hours per 100 reports, derived from an external baseline; not a causal measure
- AUTHORSHIPEvery report is reviewed, edited and signed by a physician
- DATAFlow, regions, subprocessors and retention documented per deployment
- EVIDENCERadCommons with cited sources; LaiBench internal and controlled, not external clinical validation
Sources to cite
- METHODOLOGY
Time from opening the editor to signing
Definition of the metric, sample, period and limitations of the historical dataset.
Review the methodology → - BENCHMARK
LaiBench: internal evaluation
Version, case set and limitations of the benchmark. This is not external clinical validation.
View LaiBench → - REFERENCES
RadCommons: sources and classifications
References and versions of the content used to support criteria lookup.
View RadCommons → - RESEARCH
Beyond Templates
An interactive draft on report composition and structuring. This is not external clinical validation.
Read the draft →
Historical data for citation
In a dataset of 5,200 reports finalized between April 9 and May 9, 2026, the median time from opening the editor to signing was 52 seconds. This interval was under one minute for 54.6% of reports. The measurement does not include the service’s total turnaround time and does not demonstrate clinical quality or a causal improvement.
Stories we help with
CFM Resolution 2.454/2026 in the practice of people who report. What changes with assistive AI versus autonomous diagnosis. Data residency and LGPD in healthcare. Honest clinical benchmarking: why averages hide errors. Burnout and workload in radiology.
Available sources: founders and technical team. RadCommons publishes cited sources. LaiBench identifies version, sample, snapshot and clinical limit.
Boilerplate, ready to paste
Laudos.AI is the post-imaging layer of Brazilian radiology: assistive Speech-to-Report for dictation, structuring and review. It was founded in 2025 by Natan Paraíso Ribeiro, CEO/CTO and DPO, and Francisco Akira, clinical lead. REPORT is in production; the other modules have their maturity shown on the public status page. LaiBench v3.10 records 90% fidelity on 120 internal cases and is not external clinical validation. The historical median from editor to signature is 52 seconds across 5,200 reports, from April 9 to May 9, 2026.
Direct questions
Do you share customer data for news stories?
No. Care data and institutional data are confidential. The public numbers are the ones measured in production and published on the site.
Can I test the product for a story?
You can watch a live demo with sample data. Write to us with the story angle and the deadline.
Do you comment on competitors?
We comment on criteria (authorship, auditing, data, public measurement) and leave the comparison to the comparisons page and to the reader.
Source, context and correct use of the brand
How to verify what we publish
Operational numbers should be cited together with the methodology that defines the window and the measured event. RadCommons and LaiBench have their own pages for corpus, source and evaluation. Claims about the product must preserve the central limit: the AI structures and suggests; the radiologist reviews, edits and signs.
Interview requests, high-resolution images and brand-use authorization go through the press contact. Visual material found on the site should not automatically be treated as a kit cleared for republication; credit, cropping and context are confirmed when we handle the request.
Closing a story on deadline?
Write with the story angle and the deadline in the subject line. The reply comes from oi@laudos.ai.