What will AI-assisted radiology reporting change at Mount Sinai—and what is not live yet?
Mount Sinai has signed a seven-year contract covering more than 1.2 million radiology exams a year, but go-live is planned for spring 2027. The announcement describes AI-drafted impressions and quality checks; it reports no measured accuracy, workload or patient outcome.
The short version
- What happened
- Sectra says the seven-year agreement covers all Mount Sinai radiology services—more than 1.2 million exams a year—and is planned to go live in spring 2027.
- Who it affects
- Radiologists could spend less time switching systems, but they will also need time and authority to verify every generated impression and challenge quality flags.
- What remains unproven
- Go-live is planned for spring 2027; no Mount Sinai clinical or workflow outcome has yet been reported.
Keep in mind
Go-live is planned for spring 2027; no Mount Sinai clinical or workflow outcome has yet been reported.
Both reviewed sources are supplier-controlled and commercially interested.
The announcement gives no product cost, implementation denominator by site, model identity or rollout sequence.
No accuracy, omission, laterality, false-alert, override, correction, downtime or cyber-security result is supplied.
The claim of shorter turnaround has no baseline, comparator or target in the announcement.
The 1.2 million annual examinations describe scale, not how many reports will use each AI feature.
Source analysis · 2 linked sources. Primary analysed source date: 4 November 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 10 min The evidence, methods, implications and limitations
The direct answer: the contract could change a very large workflow, but it has not changed care yet
Mount Sinai Health System has agreed to add Sectra Reporting to the imaging platform it already uses for radiology, breast imaging and its enterprise archive. Sectra says the seven-year contract covers every radiology service and more than 1.2 million examinations a year. That denominator makes the decision consequential: even a small change in report preparation, review time or error detection could affect many clinicians and patients. Yet the most important date is spring 2027, when go-live is planned. On 4 November 2026 this is a signed implementation plan, not a live result.
The proposed system brings images, dictation, report structure and AI assistance into one interface. A radiologist would dictate findings, review an automatically drafted impression and see quality checks that flag possible inconsistencies before signing. Mount Sinai says new capabilities can reach radiologists without another system to integrate. Those are plausible workflow advantages, but the announcement offers no before-and-after data. It does not report current turnaround time, expected time saved, report accuracy, override rates or patient outcomes. Readers should therefore separate the verified scale and contract from the vendor's claims about what the product may achieve.[1][2]
What has actually been agreed
Sectra describes Mount Sinai as the first United States health system to enter a partnership with it for radiology reporting. The contract was signed in the second quarter of Sectra's 2026/2027 financial year and runs for seven years. The announcement covers all radiology services rather than a single department or pilot. Mount Sinai includes seven hospitals, more than 400 outpatient practices and over 600 research and clinical laboratories, alongside the Icahn School of Medicine. The source does not break the 1.2 million annual examinations down by hospital, modality, urgency, patient group or interpreting radiologist.
The existing imaging relationship reduces one implementation barrier because Mount Sinai already uses Sectra's enterprise platform. It does not remove the work of configuring report templates, connecting speech recognition, validating AI functions, training staff, testing interfaces and monitoring safety after launch. The release does not state the contract value, implementation cost, migration timetable by site or whether every AI feature will be activated at first go-live. It also does not identify which model produces the impression draft, what data it uses, where inference runs or how model updates will be controlled. Those details determine whether the headline concept becomes a reliable clinical system.[1]
The distinction that matters
Contract, planned workflow and measured outcome are different evidence levels
The announcement verifies scale and intent. It does not yet provide live performance data.
Verified contract
Seven years; more than 1.2 million exams a year
Large planned deployment
Intended workflow
Dictation, structured report, AI impression and quality flags
Features described by the supplier
Planned launch
Spring 2027
Implementation and validation still ahead
Clinical evidence
Accuracy, misses, false alerts, workload and patient outcomes
Not reported
What the proposed reporting workflow does
Sectra's product material describes free-form voice dictation alongside the images, templates and web-based reporting forms. Its one-click AI impression tool is intended to summarise and organise the radiologist's findings into the concluding impression—the short section many referring clinicians read first. The company also says the system can run quality checks for internal contradictions, omissions and laterality errors before sign-off. After signing, the report can be distributed to the referring clinician with links, measurements and images through the wider platform.
Each step needs a different evidence question. Dictation accuracy concerns whether spoken findings become the intended text. Automatic structure concerns whether content reaches the correct section without losing meaning. Impression generation concerns whether the summary accurately reflects every relevant finding and avoids introducing unsupported language. Quality assurance concerns sensitivity, specificity and alert burden: how often a real inconsistency is caught, how often a correct report is interrupted and whether staff begin ignoring warnings. Integration concerns availability, latency, cyber security and recovery when a component fails. The announcement groups these features into one workflow but supplies no measured denominator for any of them at Mount Sinai.[1][2]
Why an AI-drafted impression requires human ownership
A radiology impression is not just a shorter version of dictated text. It prioritises abnormalities, resolves uncertainty and communicates what the referring team should do next. A generated draft may reduce repetitive editing when it is faithful, but it can also omit a small critical finding, overstate a tentative observation or carry a contradiction from the body of the report into a confident conclusion. The public material does not report the model's training data, specialty-specific performance, performance by modality or results for rare urgent findings. It also does not show whether radiologists edit most drafts or accept them with minimal change.
The radiologist remains responsible for the signed report. A safe workflow should make the draft visibly provisional, preserve the dictated source, show what the system changed and require affirmative review rather than passive acceptance. Monitoring should count clinically material edits, generated statements unsupported by findings, missed urgent abnormalities, laterality problems and cases in which staff disable or bypass the feature. Aggregate time saved cannot compensate for a hidden increase in serious reporting errors. Patients and referring clinicians also need a correction pathway when a report changes after release. None of these safeguards is demonstrated by the contract announcement, though they can be specified before launch.[1][2]
What the scale could mean for radiologists and patients
Radiologists often move between an image viewer, worklist, speech recognition, templates and separate AI applications. Fewer interfaces could reduce clicks and the cognitive burden of matching the correct patient, study and report. At more than 1.2 million examinations a year, even modest reductions in avoidable navigation could return meaningful time to image interpretation and communication. The system could also make structured findings easier for referring teams to read. These are hypotheses until Mount Sinai measures them under normal workload, overnight coverage, interruptions and different specialties.
For patients, the hoped-for benefit is a clear, accurate report reaching the treating clinician sooner. The announcement says the integrated approach can shorten turnaround, but it provides no baseline and no target. Faster is not automatically better if the summary loses nuance or an alert distracts from an urgent finding. Evaluation should therefore stratify routine and emergency examinations and should include the time to a clinically actionable report, corrected reports, communication of critical results and downstream delays—not only the time from opening an image to signing text. Patient benefit must be demonstrated in the care pathway rather than inferred from a faster software step.[1]
What Mount Sinai should publish after go-live
A credible evaluation can begin before deployment with a registered baseline. Mount Sinai could report examination volumes by modality and site, current median and tail turnaround times, addendum and correction rates, critical-result communication, radiologist workload and system downtime. During a phased rollout, comparable departments or stepped implementation dates could help distinguish the product's effect from staffing, seasonal demand or wider workflow changes. The denominator should remain examinations and reports, not only users who chose to activate the feature. Results should show how many drafts were offered, accepted unchanged, edited materially or rejected.
Safety reporting should include false-positive and false-negative quality flags, laterality and contradiction errors, generated omissions, incidents reaching patients and near misses caught by radiologists. Usability should include time spent reviewing the draft, alert fatigue and whether junior and senior staff use the tool differently. Performance should be stratified by modality, specialty, language, report complexity and urgent status. Independent review would strengthen credibility because the present evidence comes entirely from the supplier and customer announcement. Publication of negative and neutral findings would be as valuable as positive results, particularly for other hospitals deciding whether integration justifies the implementation burden.[1][2]
Evidence limits, commercial interests and what would change the assessment
Both sources are controlled by Sectra, a publicly listed supplier selling the product. Mount Sinai's vice chair of radiology information technology is quoted in the release, but Mount Sinai did not publish a separate protocol, procurement document or outcome dataset in the material reviewed. The release supplies a contract duration, examination volume and intended go-live date, which are useful verifiable facts. Claims that integration will shorten turnaround, lower complexity or benefit patients remain prospective. The product page cites use across countries and annual report volumes, but it does not provide a controlled comparison or Mount Sinai-specific performance.
The assessment would change after a successful live implementation with transparent denominators. Evidence should show that report accuracy is at least maintained, clinically material omissions and contradictions do not increase, alerts have an acceptable miss and false-alarm rate, and radiologists save review time rather than shifting work into correction. Patient-relevant evidence would include faster communication of urgent findings without more corrected reports or delayed care. Until those results exist, this is a significant health-system technology commitment and a useful marker of where clinical documentation is moving. It is not proof of improved care, not an emergency and not Breaking news.[1][2]
Watch the background, then check the evidence
Illustrative photo · not a video frame
The Royal Society · 44 min 28 sec · Background explainer
Is this the AI future we want? | The Royal Society
YouTube loads only when you choose; playback starts when you press play.
A Royal Society discussion with Madhumita Murgia and biomedical engineer Alison Noble on AI’s potential in healthcare, science and culture, recorded at the 30 June 2026 event.
Evidence check: A discussion of opportunities and risks, including future possibilities. It is not evidence that every application described is available or clinically proven.
Published 2026-07-01. Source checked 14 October 2026. Check captions on the publisher’s video page.
Check the healthcare evidencePhoto: Daniel Sone / National Cancer Institute / Unsplash · Licence
Continue the story
Read next
Continue exploring radiology: compare the finding and its evidence limits.
Qwen3 caught 320 of 335 radiology-report errors—what did it never see?
The model performed strongly on report text from two hospitals and an error-enriched comparison with three radiologists. It never saw the scans, was not used prospectively and did not measure patient outcomes.
Read the next reportExplore the full source and evidence record
What this means for people
- Radiologists could spend less time switching systems, but they will also need time and authority to verify every generated impression and challenge quality flags.
- Referring clinicians may receive clearer or faster reports only if the system preserves nuance and urgent communication; that outcome has not been measured.
- Patients need transparent correction and escalation processes when an AI-assisted report is changed after sign-off.
Global context
Hospitals worldwide are trying to reduce the number of disconnected radiology tools while preserving local reporting rules, privacy and professional accountability. A large United States deployment may influence procurement elsewhere, but results will not automatically transfer to health systems with different languages, staffing, modalities, speech-recognition systems or critical-result procedures.
What the evidence does not yet show
- Go-live is planned for spring 2027; no Mount Sinai clinical or workflow outcome has yet been reported.
- Both reviewed sources are supplier-controlled and commercially interested.
- The announcement gives no product cost, implementation denominator by site, model identity or rollout sequence.
- No accuracy, omission, laterality, false-alert, override, correction, downtime or cyber-security result is supplied.
- The claim of shorter turnaround has no baseline, comparator or target in the announcement.
- The 1.2 million annual examinations describe scale, not how many reports will use each AI feature.
What to watch next
- A public implementation protocol and the exact features enabled at first go-live
- Baseline and post-launch turnaround, correction and critical-result measures
- Draft acceptance, material edit and rejection rates with examination denominators
- False-positive and false-negative quality flags, near misses and serious incidents
- Independent evaluation of workload, report quality and patient-relevant outcomes
Living evidence record
Impact record IAI-1H57ABA
Evidence stage
Announced
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Not yet
Record status
Monitoring
Last checked
4 November 2026
Source trail
2 direct sources across 2 source types.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Related-source reporting disclosure
This record analyses 2 linked source records around the same underlying development. The extra records add method, date or context, but they do not by themselves constitute independent replication of every performance claim or predicted outcome.
Evidence trail
Sources used for this report
Links checked 4 November 2026
Research question, topics and regions
Topics: Radiology · Clinical workflow · AI documentation · Health IT · Patient safety
Regions: United States · North America · Sweden · Europe
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.