Imaging: stroke and emergencies
Software that reads emergency imaging for suspected stroke, hemorrhage and other urgent findings and alerts radiologists and treatment teams.
These tools read emergency CT scans for suspected large vessel occlusion stroke, intracranial hemorrhage, pulmonary embolism and other urgent findings and alert radiologists and treating teams. The FDA created their device type, radiological computer-assisted triage and notification software, when it granted Viz.ai's De Novo request for ContaCT on Feb. 13, 2018; Aidoc's BriefCase and RapidAI's Rapid LVO were later cleared under the same regulation. Such a device does not remove cases from a reading queue and works in parallel with the standard of care, according to the FDA.
In a 2020 briefing on AI for brain CT, the National Institute for Health and Care Excellence summarized 11 studies, six reported only as abstracts lacking methodological detail. Accuracy differs between products cleared for the same task: on the same 4,081 head CTs at one U.S. academic institution, hemorrhage sensitivity was 94.4% for Aidoc and 59.5% for Viz.ai. In a randomized trial, door-to-groin times were shorter with Viz LVO, but functional outcomes did not differ.
Aidoc
Aidoc's triage software runs in the background on CT and X-ray studies, flags suspected acute findings such as intracranial hemorrhage, pulmonary embolism and cervical-spine fracture, and moves flagged studies up the radiologist's worklist. According to the company, it analyzes imaging automatically to prioritize findings, activate care teams and facilitate patient follow-up.
Why this grade. Prospective two-phase studies in clinical use found that Aidoc's hemorrhage triage did not improve radiologists' accuracy or report turnaround on 9,954 head CTs and its pulmonary embolism triage shortened wait times for positive scans at the University of Alabama at Birmingham; its only randomized study is a 2020 conference paper, not a journal article.
Caution. An independent retrospective study of the clinically deployed cervical-spine fracture module at the University of Wisconsin-Madison found sensitivity of 54.9% (67 of 122 fractures) and a positive predictive value of 38.7% across 1,904 CT scans, compared with 91.7% sensitivity in the module's 510(k) summary.
The evidence
In a randomized study at the University of Rochester, reported in a 2020 conference paper, Aidoc's hemorrhage flags were switched off at random, with 50% probability, for AI-positive scans among 620 head CTs over 14 days; flagged scans were reported in a mean of 73 minutes versus 132 minutes for unflagged ones. The two-phase hemorrhage study of 9,954 examinations in 7,371 patients, which AuntMinnie identified as using Aidoc, found no improvement in radiologists' diagnostic performance or report turnaround. At the University of Alabama at Birmingham, mean wait time for embolism-positive scans fell from 21.5 to 11.3 minutes, with no significant change in accuracy, miss rate or report turnaround.
A retrospective before-after study of 587 patients with hemorrhage at a level 1 trauma center in Israel found 30-day mortality of 27.7% before and 17.5% after implementation. In a head-to-head study of 4,081 head CTs at one U.S. academic institution, Aidoc's hemorrhage sensitivity was 94.4% and specificity 97.4%.
- Used by
- Aidoc says it is used at more than 1,600 medical centers worldwide. Its hemorrhage tool ran in the emergency department of a level 1 trauma center in Israel, where researchers compared outcomes before and after its introduction.
- FDA
- Cleared by 510(k) as radiological computer-assisted triage and notification software. The first clearance, for hemorrhage triage, came Aug. 1, 2018 (K180647); cervical-spine fracture triage followed May 31, 2019 (K190896). A Jan. 7, 2026, clearance for its CARE foundation model (K252970) covers 11 acute findings on CT of the chest, abdomen or pelvis in patients aged 18 and older.
- Limits
- The randomized study lasted 14 days at one hospital and measured only report time, and the mortality difference comes from a nonrandomized before-after comparison. Accuracy has differed by module and by site.
RapidAI
RapidAI's stroke software analyzes noncontrast CT, CT angiography and CT perfusion. Rapid LVO flags suspected large vessel occlusions and notifies the care team, and the perfusion module estimates infarct core and tissue at risk.
Why this grade. In a retrospective before-after study of Rapid LVO in use at a U.S. comprehensive stroke center, CT angiography report turnaround fell from 30.58 to 22 minutes but treatment times did not improve; no randomized trial has tested the software, which RapidAI says was used to select patients by perfusion imaging in the DEFUSE 3 thrombectomy trial.
The evidence
DEFUSE 3, a randomized trial of thrombectomy for stroke at 6 to 16 hours in patients selected by perfusion imaging, was stopped early for efficacy: 45% of patients were functionally independent at 90 days with thrombectomy versus 17% with medical therapy. In the before-after study of Rapid LVO, the tool's sensitivity was 0.96 and specificity 0.85, and door-to-needle and door-to-puncture times and rates of severe disability or death did not differ significantly.
A retrospective study of the company's earlier RAPID CTA software at a regional hospital in Australia found sensitivity of 0.94 and specificity of 0.76 for intracranial occlusions. Rapid LVO's 510(k) summary reported sensitivity of 0.970 and specificity of 0.956. In a head-to-head study of 263 CT angiograms, Rapid LVO's sensitivity was 0.90, specificity 0.86 and positive predictive value 0.45; the study was funded by Canon Medical, and two authors owned stock in the maker of the competing product.
- Used by
- RapidAI says it is used in more than 2,500 hospitals in more than 100 countries. The company lists DEFUSE 3 among trials run with its software.
- FDA
- The FDA cleared Rapid LVO by 510(k) on July 9, 2020 (K200941) as computer-assisted triage and notification software. A 510(k) clearance on Dec. 27, 2018 (K182130) covers RAPID CT and magnetic resonance perfusion to aid selection of patients with known internal carotid or proximal middle cerebral artery occlusion for thrombectomy.
- Limits
- The evidence on Rapid LVO's effect in care is one single-center before-after study, and the link between DEFUSE 3 and the software rests on the company's statement. The Rapid LVO studies are retrospective, and two, including the before-after study, were written by authors with ties to a competitor.
Viz.ai
Viz LVO analyzes CT angiograms for suspected large vessel occlusion stroke and sends an alert with the images to the stroke team's phones, in parallel with the radiologist's read. It is part of a suite of software for suspected neurovascular disease that includes a hemorrhage module and team messaging.
Why this grade. A stepped-wedge cluster randomized trial at four comprehensive stroke centers in the Houston area found that switching on Viz LVO with secure messaging was associated with an 11.2-minute reduction in door-to-groin time among 243 patients treated with thrombectomy, with no difference in functional outcome.
Caution. A head-to-head study of 4,081 head CTs at one U.S. academic institution found the Viz ICH hemorrhage module had a sensitivity of 59.5%, compared with 93% in its 510(k) summary. The caution concerns the hemorrhage module, not Viz LVO.
The evidence
In the Houston trial, the time from CT to the start of thrombectomy fell by 9.8 minutes; times to thrombolysis, length of stay and functional independence did not differ. A prospective study of 1,822 stroke-code CT angiograms at a U.S. academic stroke center found sensitivity of 93.8% for internal carotid artery terminus and M1 occlusions, falling to 74.6% when M2 occlusions were included, with specificity of 91.1%; two senior authors had received research funding from Viz.ai.
A retrospective study of 1,167 CT angiograms in Haifa, Israel, found sensitivity of 0.81. Before-after studies reported door-to-groin time for directly arriving patients falling from 127 to 86 minutes at a U.S. comprehensive stroke center and transfer time falling from 132.5 to 110 minutes at a spoke hospital in a U.S. hub-and-spoke network.
- Used by
- Viz.ai says more than 2,000 hospitals use its software. Four comprehensive stroke centers in the Houston area used Viz LVO in a randomized trial from January 2021 to February 2022.
- FDA
- The FDA granted the De Novo request for ContaCT, now Viz LVO, on Feb. 13, 2018 (DEN170073), creating the device type radiological computer-assisted triage and notification software; such a device operates in parallel with the standard of care. The hemorrhage module, Viz ICH, was cleared by 510(k) on March 18, 2020 (K193658).
- Limits
- The trial's primary outcome was a treatment time; functional independence, an exploratory outcome, did not differ. The other outcome studies are small before-after comparisons, and several accuracy studies had authors with Viz.ai funding or advisory roles.
- Cost
- No list price is published. The Centers for Medicare & Medicaid Services set a maximum new technology add-on payment of $1,040 per hospital case involving ContaCT for fiscal 2021.
Grades: A, randomized evidence of benefit; B, evidence from clinical use; C, accuracy studies only; D, little or no independent evidence. How the grades work. Reviews of the published evidence, not medical advice or an endorsement of any product.