Bracco and Avicenna.AI won FDA clearance for a brain aneurysm detector, but the validation record raises a harder question about what that number means outside the lab.
The FDA cleared UPTUNITY | Aneurysm on November 13, 2026. The press release called it an AI-enabled software that âmay improve detection and assessment of brain aneurysms.â The qualifying word is not an accident. Brain aneurysm detection has a gap that tools and regulatory language keep bumping into, and it goes deeper than sensitivity percentages.
About 1 in 50 adults carries an unruptured intracranial aneurysm. Most will never rupture. When they do, roughly 1 in 5 patients dies before leaving the hospital. The survivors often carry long-term disability. Aneurysmal subarachnoid hemorrhage strikes at a rate of 10 to 15 per 100,000 people per year. It is catastrophic, it is fast, and in a meaningful number of cases the aneurysm was sitting on a scan that someone read without catching it.
UPTUNITY | Aneurysm is an answer to part of that problem. It is worth looking at exactly which part.
What the FDA Actually Cleared
The clearance covers two linked products from Bracco Imaging and Avicenna.AI: UPTUNITY | Aneurysm Detection (K262062) and UPTUNITY | Aneurysm Management (K262147). Together they form a system designed to do two things: flag suspected saccular aneurysms greater than 3 mm on head computed tomography angiography (CTA), and then track known aneurysms over time through quantitative measurement comparison.
The detection module runs as a concurrent reading aid. A radiologist reads the scan; the software flags where it thinks an aneurysm is. The clinical decision stays with the radiologist. That is how the approval is worded, and it matters: UPTUNITY is a CADe/CADx system, not an autonomous classifier. It identifies location and size. It does not diagnose.
The management module, branded around Braccoâs MIMPS quantitative measurement platform, is meant to solve a different problem. When a patient returns for a surveillance scan of a known aneurysm, the radiologist has to answer one question: has it changed? That comparison is harder than it sounds. Measurement variability across readers, scanners, and time creates noise that can obscure real growth. MIMPS is designed to standardize those comparisons.
Fabio Tedoldi, Bracco Imagingâs head of global research and development, framed it directly: âFor a patient with an unruptured aneurysm, the follow-up scan always brings the same question: Has it changed?â His point is that precision in detection means nothing if the follow-up system cannot answer that question reliably.
UPTUNITY is slated for a commercial showcase at RSNA 2026 in Chicago, running November 29 through December 3. The partnership behind it was signed in March 2026, focused specifically on AI tools for contrast-enhanced CT.
The Sensitivity Number and What It Does Not Tell You
Braccoâs own validation study reported 94.4% sensitivity for CINA-Aneurysm, the technical name behind the UPTUNITY clearance. That number is the one that travels. It is the figure most likely to appear in a sales deck, in a hospital RFP, and in the system reminder attached to a radiologist workstation.
The problem is that 94.4% sensitivity is a controlled-environment figure. The FDAâs 510(k) review noted that the reader study primary endpoint missed statistical significance, with a p-value of 0.075. That means the trial did not definitively establish that radiologists using the AI outperformed those reading unassisted. This is a threshold issue, not a catastrophic failure. But it is a distinction that tends not to travel as far as the 94.4% headline.
Real-world performance data for similar tools is available. A December 2025 study in the American Journal of Neuroradiology examined 2,534 head CTA scans processed through the Aidoc aneurysm detection algorithm at the University of Washington Medical Center. The researchers found that the AI achieved an overall sensitivity of 70.5% (95% CI: 65.1 to 75.5%). Neuroradiologists in the same institution, over the same cases, achieved a sensitivity of 94.0% (95% CI: 90.7 to 96.3%). Both tools reached comparable specificity: 98.6% for the AI, 98.3% for the radiologists.
When the analysis was narrowed to aneurysms that matched the AIâs intended-use specification, specifically unruptured, saccular aneurysms larger than 5 mm, the AI sensitivity improved to 85.7%. The neuroradiologists still led at 97.1%.
That 23-percentage-point gap in overall sensitivity is not a verdict on UPTUNITY. Aidoc and CINA-Aneurysm are different models trained on different data. But it is an honest benchmark for what happens to AI aneurysm detection when it leaves the validation lab and enters a teaching hospital with heterogeneous cases, rare edge presentations, and clinical urgency pulling attention in multiple directions.
Why Worklist Pressure Changes the Math
The University of Washington study was retrospective and covered cases from 2018 to 2021, a period before AI aneurysm tools were routine. It established a baseline, not a prediction. What happens when radiologists have AI assistance as a concurrent tool? A September 2026 study in the Journal of the American College of Radiology addressed that directly.
Researchers at Northwell Health ran the Aidoc aiOS aneurysm algorithm across 3,856 head CTA examinations prospectively, with the AI running in parallel while radiologists read without seeing its output. After both reads were complete, the results were compared.
The AI identified 55 aneurysms that radiologists had not flagged, a 39% relative increase in detection yield. AI sensitivity was 84.6% versus 71.8% for radiologists. Radiologist positive predictive value was higher: 92.7% compared to 78.2% for the AI. Of 101 AI-only detections, 46 were false positives.
The researchers noted that AI-only detections were concentrated in inpatients, where the AI added 18 aneurysms against 7 false positives. In outpatients, the AI added only 4 true aneurysms but generated more false positives than true detections, suggesting that population prevalence and case mix matter as much as model performance.
Cosimo DePinto, Bracco Americasâ executive vice president for marketing and commercial operations, made this point in the UPTUNITY announcement: âA small incidental aneurysm on a head CTA can easily be overlooked when the work list keeps growing.â He is right. The Northwell study backs it. Radiologist miss rates in that study ran in a range that an AI safety net can meaningfully address, even when the AI is not the more sensitive reader on paper.
There is an uncomfortable paradox here. The AJNR study showed radiologists at 94.0% sensitivity without AI assistance. The Northwell study showed radiologists at 71.8% in a parallel, non-AI-informed read. The difference is not explained by the papers, but worklist size, time pressure, and case mix are plausible contributors. What that suggests is that the value of AI as a concurrent reader may depend less on whether the AI outperforms radiologists in controlled studies and more on whether it catches what gets missed when radiologists are under pressure.
The Detection-to-Follow-Up Gap
FDA clearance for detection creates a secondary problem the clearance does not resolve: what happens after the aneurysm is found?
The Korean Neuroendovascular Society published follow-up imaging guidelines in 2024 based on a modified Delphi consensus among 40 experts. For aneurysms treated by coiling, the recommendations call for initial follow-up imaging within 3 to 6 months, then at 1, 2, 4, and 6 years. High-risk cases, defined as aneurysms greater than 10 mm, those with Grade 2 or higher Raymond-Roy occlusion, or those that ruptured before treatment, warrant imaging every 3 to 5 years for life.
The evidence behind those intervals is sobering. Pooled data from 15 cohorts covering 2,773 patients showed aneurysm recurrence in 11.4% of cases after coiling. Among patients followed over the ISAT trial, the annual bleeding risk after coiling was 0.08%. That is a low annual rate, but across a treated population followed for a decade, it accumulates. ISAT long-term data showed retreatment after coiling in roughly 20% of patients by the 10-year mark.
That retreatment rate depends on detecting the recurrence. Recurrence requires serial imaging. Serial imaging requires consistent measurement. This is exactly the problem the UPTUNITY Management module targets. Whether it delivers measurement consistency that reduces retreatment delays or unnecessary retreatment is a clinical question the FDA clearance does not answer, because the evidence for that outcome was not part of the 510(k) submission.
This pattern is not unique to UPTUNITY. It appears throughout radiology AI clearance. A device wins clearance as a concurrent reader or adjunctive tool, the product goes into clinical use, and the downstream question of whether it changes patient outcomes, not just detection rates, remains open for years. The FDAâs regulatory framework asks whether a device is substantially equivalent to a predicate. It does not require proof that the device improves clinical outcomes. That standard exists for a reason: requiring outcomes evidence for every AI tool would delay clearance indefinitely. But it means the evidence that matters most to patients arrives after, not before, market entry.
The Size and Location Problem
UPTUNITYâs clearance is scoped to saccular aneurysms greater than 3 mm on head CTA. That is a specific population, and not because Bracco chose a generous threshold. It reflects the fundamental limitation of current AI aneurysm detection: small aneurysms are where most misses happen.
In the Lausanne University Hospital reader study published in March 2025, an AI model achieved 74% sensitivity on a 100-scan test set. The researchers found that 93% of the junior radiologistâs misses, without AI assistance, were aneurysms smaller than 3.5 mm. For the senior radiologist, 91% of misses were below 3.1 mm. These are the cases where AI should theoretically help the most. In that study, AI assistance did not significantly improve detection for either reader, though the junior reader moved from 78% to 83% sensitivity.
The 3 mm threshold is simultaneously the most clinically important zone and the zone where current models are least reliable. Aneurysms below 3 mm have low rupture risk. Aneurysms between 3 and 7 mm carry meaningful but hard-to-quantify risk; the ISUIA prospective cohort showed 5-year rupture rates of about 1.2% for aneurysms smaller than 7 mm at the cavernous carotid. It is the 3-to-7 mm range where finding an aneurysm triggers the cascade: discussion with the patient, decision on surveillance versus intervention, serial imaging over years or decades.
Missed aneurysms in that range are not all catastrophic. Most small aneurysms never rupture, and finding them triggers treatment conversations that carry their own risks. But the cases that do rupture, without prior detection or treatment, are the ones that show up in retrospective chart reviews tagged as preventable. AI clearance in this size range, even with imperfect sensitivity, shifts that review in a direction that matters.
What RSNA 2026 Will Actually Show
RSNA 2026 begins nine days from now. Bracco and Avicenna.AI have committed to a commercial showcase. A separate session on December 3, presented by researchers at the University of Louisville, will show results from applying an Aidoc aneurysm algorithm as a secondary reviewer on 631 contrast-enhanced neck CT exams done for routine indications, not dedicated aneurysm screening.
That session found 3 exams flagged positive. Two showed aneurysms under 2 mm that had not been identified in the original report. Both led to report addenda. The third showed a 4.5 to 5 mm aneurysm the radiologist had already caught. CTA confirmed it.
Three cases out of 631 is not a clinical trial. But it shows something that matters for incidental detection: brain aneurysms show up on neck CTs not ordered to find them, and AI running in the background catches some. Whether the finding changes patient management, and in what direction, is what years of follow-up data will eventually answer.
The larger RSNA question for UPTUNITY is commercial deployment. Bracco has 4,000 employees, operations in more than 100 countries, and roughly 2 billion euros in annual revenue. Avicenna.AI has been developing neuro AI tools since 2018 and already holds multiple FDA clearances for the CINA platform, including tools for intracranial hemorrhage and pulmonary embolism. The partnership is backed by real infrastructure. What it lacks is the outcome evidence that hospitals with active quality improvement programs will eventually ask for.
That evidence takes time. It requires tracking patients from AI-assisted detection through surveillance, through intervention decisions, through retreat or watchful waiting. The window is years, not months.
What This Means for Oatmeal Health
The question underneath every AI clearance in radiology is whether earlier and more accurate detection translates into earlier intervention, and whether earlier intervention produces better outcomes. For aneurysms, that chain is complicated by the fact that most aneurysms found incidentally never rupture, and treating them carries real procedural risk.
Lung cancer screening faces the same chain in a different form. Oatmeal Healthâs mission is built on exactly this problem: low-dose CT detects early-stage lung cancer at rates that change survival curves, but detection without systematic follow-up becomes noise. The NLST showed a 20% reduction in lung cancer mortality in screened patients, but the program is only worth that number if patients who screen positive complete the downstream evaluation, the follow-up scans, and the diagnostic workup. When they do not, the detection did not help.
UPTUNITY is new. Its outcome data does not exist yet. What does exist is a clearance, a partnership with real distribution capacity, and a clinical question that is harder than the sensitivity number suggests. The question is not whether AI can find brain aneurysms. The Northwell study shows it can add meaningful detection, particularly in inpatient settings under real worklist pressure. The question is whether finding them earlier, more reliably, and tracking them more consistently translates into fewer ruptured aneurysms, fewer emergency surgeries, and fewer patients who arrive at a subarachnoid hemorrhage without having had the chance to decide what to do with the aneurysm that was sitting there.
That answer will come from follow-up data that takes years to collect. The FDA clearance is the beginning of that chain, not the end of it.
Jonathan Govette is Co-Founder and CEO of Oatmeal Health, a healthcare technology company focused on connecting high-risk patients to life-saving cancer screening. Oatmeal Healthâs LungAI platform is investigational and not FDA cleared. This article reflects publicly available research and is not medical advice.
Key References
Bracco and Avicenna.AI press release, FDA clearance for UPTUNITY | Aneurysm. PR Newswire. November 13, 2026.
Akinpelu BJ et al. Assessing the Accuracy of Artificial Intelligence in Detecting Intracranial Aneurysms in a Clinical Setting Relative to Neuroradiologists. American Journal of Neuroradiology. 2025;46(12):2545-2552.
Stein S et al. AI complements radiologists to improve brain aneurysm detection. Journal of the American College of Radiology. September 25, 2026.
Di Noto T, Jankowski S, Puccinelli F et al. Assessing workflow impact and clinical utility of AI-assisted brain aneurysm detection: a multi-reader study. arXiv:2503.17786. March 2025.
Cho YH, Choi J et al. Imaging follow-up strategy after endovascular treatment of intracranial aneurysms: a literature review and guideline recommendations. Journal of Cerebrovascular and Endovascular Neurosurgery. 2024;26(1):1-10.
Bracco, Avicenna.AI get FDA clearance for brain aneurysm AI software. AuntMinnie. November 13, 2026.
FDA clears new AI software for brain aneurysm detection. Radiology Business. November 13, 2026.
AI can be a safety net for intracranial aneurysms (RSNA 2026 preview). AuntMinnie. November 11, 2026.
Brain Aneurysm Statistics 2026. Gitnux. July 3, 2026.
FDA clears CTA-based AI software for brain aneurysm detection and follow-up. Diagnostic Imaging. November 13, 2026.


