Imaging: fractures and bone
Software that marks suspected fractures, dislocations and joint effusions on radiographs, or flags vertebral compression fractures on CT scans done for other reasons, for the reading clinician.
Category overviewHide category overview
These tools read radiographs for fractures, dislocations and joint effusions, or read CT scans done for other reasons for vertebral compression fractures, and mark findings for the reading clinician. FDA created the main device type, radiological computer-assisted detection and diagnosis software for fracture, when it granted the De Novo request for OsteoDetect on May 24, 2018; later fracture tools were cleared by 510(k) under that classification.
Most evidence comes from reader studies, in which clinicians read enriched case sets with and without the software. A 2026 meta-analysis pooling 17 studies found AI assistance raised clinicians' sensitivity from 73% to 87%, with specificity of 94% unaided and 95% aided, and the largest gains among junior clinicians. All 16 studies in a meta-analysis of foot and ankle fracture detection were retrospective. Randomized trials in routine care have begun: a trial at three hospitals in Austria and Germany finished enrolling in April 2026 without published results, and a U.K. cluster randomized trial began recruiting in January 2026. Open questions include whether flagged findings change treatment, recalls for missed fractures and time in the emergency department.
AZmed Rayvolve
Software that marks suspected fractures on adult and pediatric limb and pelvis radiographs as a reading aid.
Full reviewHide full review: AZmed Rayvolve
Rayvolve analyzes radiographs of the limbs, hip and pelvis in patients aged 2 and older and highlights suspected fractures for the reading clinician, according to its FDA summary. A study with an AZmed author describes a wider suite that also reads chest radiographs and estimates bone age.
Why this grade. Published studies measure accuracy only. The reader study behind its FDA clearance, 24 readers and 186 cases, was published by Cleveland researchers with company co-authors; aided sensitivity rose from 0.865 to 0.955. The Nice pediatric emergency study has posted no results, and no study of its effect on care was found.
Caution. In an independent study of 1,500 radiography sets at Lariboisière Hospital in Paris, Rayvolve's specificity was 70.4%, against 88.6% for adults in its FDA summary. Sensitivity in that study, 92.6%, matched the other products tested.
The evidence
The 2024 FDA summary reported pediatric sensitivity of 0.9611 and specificity of 0.8597 on 3,016 radiographs and noted the device did not perform as well in children as in adults. In a pediatric validation of 2,634 radiography sets at Bicêtre Hospital near Paris, sensitivity was 95.7% and specificity 91.2%; sensitivity was lower in children under 5. In a Reims comparison of 366 children, Rayvolve and BoneView had similar areas under the curve (0.874 and 0.862).
In the Lariboisière study, Rayvolve's accuracy was 71.0%, against 88.8% for BoneView. In a Munich registry, its sensitivity was 79.48% against reports that counted old as well as acute fractures. In Munich, radiologists' accuracy did not change with Rayvolve, though reading was faster. An evaluation of 258,373 radiographs with an AZmed employee as corresponding author reported 97.4% sensitivity.
- Used by
- Lenval Children's Hospital in Nice tested it with residents in pediatric emergencies in a prospective study completed in February 2024.
- FDA
- FDA cleared Rayvolve for adults by 510(k) on June 2, 2022 (K220164), extended it to patients aged 2 and older on July 17, 2024 (K240845), and cleared a further version on May 12, 2026 (K260378).
- Randomized trial
- None published
- Key study
- Academic Radiology, 2024, reader study behind the FDA clearance, 24 readers and 186 cases
- Limits
- All studies measured accuracy, many with AZmed employees as authors, and independent studies report lower specificity or sensitivity than the company's figures. No study measured missed fractures, treatment or time in routine care.
- Cost
- No published price found. checked
- Specialties
- Pediatrics, Emergency medicine, Radiology
Gleamer BoneView
Software that marks suspected fractures, dislocations and effusions on trauma radiographs as a concurrent reading aid.
Full reviewHide full review: Gleamer BoneView
BoneView analyzes limb, pelvis, rib and spine radiographs and highlights suspected fractures for the reading clinician, according to its FDA summary; the version used in a trial in Austria and Germany also flags dislocations, joint effusions and focal bone lesions.
Why this grade. Nonrandomized studies have measured its effect in care: in an Italian before-and-after study, missed fractures fell nonsignificantly after orthopedists began using it on night shifts, and at Bærum Hospital in Norway, a radiographer-supervised workflow led to one premature discharge among 1,248 patients. The FAIR randomized trial of 1,667 encounters finished in April 2026 without published results.
The evidence
In the Tel Aviv study, mean emergency stays for patients without fractures fell by 8 minutes after deployment; 8,253 such visits without fractures were compared with 6,864 visits with fractures as a control; stays over four hours fell from 11.3% to 8.2% and early revisits did not rise. In Milan, fracture recalls were 0.19% of 13,813 patients before and 0.17% of 14,842 after. At Bærum, radiographers changed the pathway the AI suggested in 20% of 1,248 patients, and one patient was discharged prematurely.
In a reader study of 480 examinations with Gleamer co-authors, the study reported in its 510(k) summary, 24 readers' per-patient sensitivity rose from 64.8% to 75.2%. A systematic review of eight reader studies found higher sensitivity, mixed specificity and high risk of bias. In Munich, two radiologists read 1,037 studies with and without three tools; accuracy did not change, but reading time fell.
Independent figures vary. At Lariboisière Hospital in Paris, sensitivity was 91.3% and specificity 90.5% on 1,500 radiography sets. In a Munich registry, sensitivity was 75.57% against reports that counted old as well as acute fractures; in a Norwegian four-hospital study, specificity was 71%.
- Used by
- Bærum Hospital in Norway introduced it in a radiographer-supervised emergency workflow in September 2023, researchers reported. Tel Aviv Sourasky Medical Center deployed a fracture detection tool, which its study names as BoneView in its keywords, in its orthopedic emergency department in October 2023.
- FDA
- FDA cleared BoneView for adults by 510(k) on March 1, 2022 (K212365), with Imagen's FractureDetect as predicate, and cleared version 1.1 extending some regions to patients aged 2 to 21 on March 2, 2023 (K222176).
- Randomized trial
- None published
- Key study
- Journal of Imaging, 2026, before-and-after study of orthopedists' night-shift use in Milan, 13,813 patients before and 14,842 after
- Limits
- No randomized trial has reported. The care studies were single-center and nonrandomized, and the Tel Aviv paper names BoneView only in its keywords. Most reader studies used enriched case sets, with Gleamer staff among authors of several.
- Cost
- No published price found. checked
- Specialties
- Emergency medicine, Orthopedic surgery, Radiology
Imagen OsteoDetect and FractureDetect
The first FDA-authorized AI software for fracture detection, marking suspected fractures on adult radiographs.
Full reviewHide full review: Imagen OsteoDetect and FractureDetect
OsteoDetect marks suspected distal radius fractures on adult wrist radiographs during review, according to FDA; its successor FractureDetect covers 12 adult study types from ankle to wrist, including hip, pelvis and femur, according to its 510(k) summary.
Why this grade. Its evidence is accuracy data reviewed by FDA: the De Novo decision summary reported a reader study of 24 clinicians in which aided area under the curve rose from 0.840 to 0.889, and published company studies are reader studies; no study of the product's effect on patient care was found.
The evidence
In the De Novo review, standalone sensitivity was 0.921 and specificity 0.902 on 1,000 wrist images; in the reader study of 200 cases, aided sensitivity was 0.803 against 0.747 unaided. The FractureDetect summary reported standalone sensitivity of 0.951 and specificity of 0.893 on 11,970 radiographs, and reader area under the curve rising from 0.912 to 0.952.
In the company's founding study, emergency medicine clinicians reading wrist radiographs raised sensitivity from 80.8% to 91.5% with the model. In a company reader study of 24 clinicians across 12 anatomic regions, the fracture miss rate for clinicians with limited imaging training fell from 20% to 9%, similar to aided radiologists' 10%. Imagen employees wrote or co-wrote these studies, and no independent validation was found in PubMed.
- FDA
- FDA granted OsteoDetect's De Novo request on May 24, 2018 (DEN180005), creating the device type radiological computer-assisted detection and diagnosis software for fracture, and cleared FractureDetect by 510(k) on July 30, 2020 (K193417). De Novo covers OsteoDetect (DEN180005, 2018); its successor FractureDetect, covering 12 adult study types, was cleared by 510(k) (K193417) on July 30, 2020.
- Randomized trial
- None published
- Key study
- FDA De Novo decision summary for DEN180005 (granted May 24, 2018), reader study of 24 clinicians and 200 cases
- Limits
- All published studies were run by the company on retrospective, enriched case sets read under test conditions. No independent accuracy study and no study of diagnoses, treatment or workflow in routine care was found.
- Cost
- No published price found. checked
- Specialties
- Emergency medicine, Radiology
Nanox AI HealthVCF
Software that flags suspected vertebral compression fractures on chest and abdominal CT done for other reasons.
Full reviewHide full review: Nanox AI HealthVCF
HealthVCF analyzes chest and abdominal CT scans and flags those suggesting a vertebral compression fracture so that clinicians can prioritize bone health follow-up, according to its FDA summary. A related product, HealthOST, labels vertebrae and measures height loss and bone density, its summary states.
Why this grade. One study measured what followed its use in care: at a Danish hospital, among 538 patients with confirmed fractures flagged by the software, 78 were referred for bone density scans and 25 newly diagnosed with osteoporosis within 6 months; there was no comparison group and no randomized trial.
Caution. Independent studies found lower sensitivity than the 90.20% in its 510(k) summary: 68% for moderate or severe fractures in the Danish study, whose authors concluded it was not generalizable to that population, and 65% in an Australian study of 1,696 scans.
The evidence
In Denmark, 1,543 of 10,012 scans were flagged and 630 were confirmed on review; specificity was 0.91, and 18% of the cohort died within 6 months. The authors said the tool should be tied to a fracture liaison service. In Sydney, specificity was 92%, and the software found 88 fractures the radiologist had not reported, one for every 19 scans.
In São Paulo, among 899 scans, sensitivity for moderate to severe fractures was 73.8% and specificity 92.7%; a general radiologist had not reported fractures in 62 of 145 flagged scans. In Toronto, HealthOST at a 20% height-loss threshold had 92.0% sensitivity but a positive predictive value of 16.5%. A study with Clalit Health Services, with company co-authors, combined fracture and bone density outputs to predict fractures.
- Used by
- Zealand University Hospital in Denmark ran it on about 10,000 CT scans in 2022, researchers reported.
- FDA
- FDA cleared HealthVCF, from Zebra Medical Vision, as notification and prioritization software by 510(k) on May 12, 2020 (K192901), and cleared HealthOST on April 22, 2022 (K213944).
- Randomized trial
- None published
- Key study
- Calcified Tissue International, 2024, single-hospital descriptive study, 10,012 CT scans
- Limits
- The care study was descriptive, at one hospital, without a control group; radiologists had already reported 91% of the confirmed fractures, so it could not show that flagging changed treatment beyond usual reporting. The accuracy studies used different fracture thresholds and reference readers.
- Cost
- England's National Institute for Health and Care Excellence put Zebra-Med's yearly license fee for HealthVCF at £38,000 to £90,000, depending on a site's scan volume, in a July 2021 briefing. checked
- Specialties
- Endocrinology, Radiology
Reviews of the published evidence, not medical advice or an endorsement of any product.