Endoscopy · Real-time computer-aided polyp detection (CADe) for colonoscopy
MAGENTIQ-COLO (ME-APDS)
Magentiq Eye, Ltd.
Magentiq Eye's polyp detector is one of the few colonoscopy AI systems whose pivotal evidence is a randomised trial of the device itself, published in The Lancet Digital Health. Across ten hospitals it found 37 percent more adenomas per colonoscopy. It also removed more tissue that was not adenoma. The trial is the only one, and the manufacturer paid for it.
Performance Metrics
Clinical Evidence
The evidence rests on one randomised controlled trial (RCT) of this device, Maas and colleagues (The Lancet Digital Health, 2024; NCT04640792). Ten hospitals in Europe, the United States and Israel took part, with 31 endoscopists. Patients referred for screening outside a faecal immunochemical test (FIT) programme, or for surveillance, were randomised to CADe-assisted or conventional colonoscopy. A subset had both examinations in tandem. The modified intention-to-treat population was 916 patients: 449 with CADe and 467 without. The trial had two co-primary endpoints. Adenomas per colonoscopy rose from 0.51 to 0.70 (Poisson effect ratio 1.37, 95% confidence interval 1.07 to 1.77, p=0.015). Adenomas per extraction, the check that the extra detections were not padding from non-adenomatous resections, was non-inferior: 0.59 with CADe against 0.66 without (p<0.001 for non-inferiority, 20 percent margin). That is a ratio test, and the counts behind it show more removals overall: 536 extractions in 449 CADe colonoscopies against 360 in 467 conventional ones, so about 0.49 non-adenomatous extractions per colonoscopy with CADe against 0.26 without. The authors read the ratio as no increase in the resection of non-adenomatous lesions. The counts say the number of such removals roughly doubled while the adenoma share of removals fell from 0.66 to 0.59, inside the margin. The FDA summary computes the same endpoint as a per-patient mean, 0.31 with CADe against 0.27 without. In the 127 tandem patients the adenoma miss rate was 19 percent when CADe went first and 36 percent when conventional colonoscopy went first (p=0.024). The FDA 510(k) summary for K223473 reports the same study and adds that adenoma detection rate was significantly higher, more sessile serrated lesions were found, withdrawal was not clinically longer and no intervention-related adverse events occurred in either arm. The manufacturer's press release puts the detection-rate gain at 7 percentage points absolute; that figure is not in the published abstract and is recorded here as the company's number. Two cautions. The gain was not uniform: the FDA summary's subgroup table shows adenomas per colonoscopy of 0.21 in both arms for patients under 50 (73 patients), while every other subgroup favoured CADe. And the trial was funded by Magentiq Eye; no independent RCT of ME-APDS has been published. The same Radboud group's DISCOVERY II trial (Endoscopy, 2024), which found no significant detection gain, tested PENTAX Medical's own DISCOVERY system, a different product, and does not bear on this entry. The FDA standalone tests are consistent across the three clearances. On the 2025 test set of 252 full colonoscopy videos containing 806 polyps the system detected 99.7 percent of histology-verified polyps for at least one frame, and 91.2 percent of all polyps for at least seven consecutive frames. Frame-wise recall, the share of frames in which a visible polyp was boxed, was 76.5 percent. The false-positive rate met the prespecified threshold of 0.033 per frame. Two AurisMarks: FDA clearance, a CE mark and a UK registration, RCT-grade evidence, and no safety signal meet the two-mark bar. The third mark needs a wider regulatory footprint (Canada, Australia or Japan), documented deployment scale and an independent trial; none of the three is on the public record.
| Study | Design | n | Sensitivity | Specificity | AUC | Published |
|---|---|---|---|---|---|---|
| Maas MHJ, Neumann H, Shirin H, Katz LH, Benson AA, Kahloon A, Soons E, Hazzan R, Landsman MJ, Lebwohl B, Lewis SK, Sivanathan V, Ngamruengphong S, Jacob H, Siersema PD (funded by Magentiq Eye) | RCTRCT | 916 | n/a (detection-outcome trial) | n/a | — | Lancet Digit Health 2024;6(3):e157-e165 (PMID 38395537; NCT04640792). Ten hospitals in Europe, the USA and Israel, 31 endoscopists, 952 enrolled, 916 in the modified intention-to-treat analysis. Co-primary endpoints: adenomas per colonoscopy 0.70 vs 0.51 (effect ratio 1.37, 95% CI 1.07–1.77, p=0.015); adenomas per extraction non-inferior, 0.59 vs 0.66 (p<0.001). Adenoma miss rate in 127 tandem patients 19% vs 36% (p=0.024). |
| Magentiq Eye standalone performance test (FDA 510(k) summary K252178) | RetrospectiveRetrospective | 252 | 99.7% polyp-wise recall on histology-verified polyps (one or more frames); 91.2% of all polyps at seven consecutive frames; frame-wise recall 76.5% (95% CI 73.8–79.5) | 0.0334 false positives per frame (prespecified threshold met) | — | FDA 510(k) summary, decision 3 October 2025. Offline test on 252 full colonoscopy videos with 806 polyps (660 small, 111 medium, 35 large; 375 with histology, 302 adenomas); median polyp coverage 81.7%. n is videos, not patients. Algorithm identical to K244023 and K223473. |
| Magentiq Eye standalone performance test (FDA 510(k) summary K223473) | RetrospectiveRetrospective | 172 | 100% to 99.6% polyp-wise recall on histology-verified polyps (one to seven consecutive frames); frame-wise recall 75.7% (95% CI 72.5–78.8) | 0.0328 false positives per frame (prespecified threshold met) | — | FDA 510(k) summary, decision 25 July 2023. Offline test on 172 full colonoscopy videos with 449 polyps (263 with histology, 210 adenomas). Polyp-wise recall on histology-verified polyps stayed between 97.3% and 100% across subgroups by age, sex, race, BMI, US vs non-US site, indication and endoscope make (Olympus, Pentax). Median marker latency 0.166 s. n is videos, not patients. |
Inside the algorithm
Editorial featureHow MAGENTIQ-COLO reaches a verdict.
Five stages — from raw input to verdict — drawn from manufacturer documentation and the public regulatory record.
- INGEST
- NORMALISE
- DETECT
- LOCALISE
- VERDICT
Stage 01 · INGEST
The processor's video feed, split in two.
The system sits beside the colonoscopy cart and takes the digital video output of the endoscopy processor. The endoscopist's own monitor is untouched. A second screen shows the same picture with the software's overlay. Nothing about the scope, the light or the withdrawal changes; the input is the colonoscopy already under way.
Input
Video frames, real time
Inference
On-device
Inside the Auris+ Listing
Five more sections complete this device’s Auris+ Listing.
Decision Ledger
ProPro unlocks a private, cross-vendor log of every case you read on this device — what the AI called, what you concluded, and a one-line reason if you overrode it. Ready for the EU AI Act's deployer logging obligations when they land in 2028.
Clinical Evidence Deep Dive
ProPro unlocks the structured clinical-evidence summary — study count, target patient population, and a tabular accuracy-metrics view drawn from peer-reviewed sources.
Peer-Reviewed Publications
ProPro unlocks the curated peer-reviewed publication list with PubMed cross-links — the citation backbone of every editorial verdict.
Post-Market & Regulatory Conditions
ProPro unlocks the post-market surveillance summary, recall record, and the conditions of approval that bound real-world use.
AI Algorithm Version History
ProPro unlocks the chronological record of algorithm version changes — what changed when, drawn from manufacturer changelogs and regulatory filings.
Regulatory Approvals
Class I
Safety Record
No recall, FDA safety communication or MAUDE adverse-event report naming Magentiq Eye or MAGENTIQ-COLO was found in openFDA on 5 September 2026. The recall and enforcement endpoints return no record for the whole product code QNP. The pivotal trial reported no intervention-related adverse events. The design bounds the risk: the overlay sits on a separate screen, the original video is unaltered, and the endoscopist decides what to inspect and resect. The residual hazards are those of every CADe system. A missing box does not clear a segment, and false alerts can prompt unnecessary polypectomy. On the second, the trial's adenomas-per-extraction ratio stayed within its non-inferiority margin, but extractions per colonoscopy rose from 0.77 to 1.19 and non-adenomatous extractions roughly doubled.
Intended Use & Indications
The cleared indication reads: ME-APDS is intended to be used by endoscopists as an adjunct to the common video colonoscopy procedure (screening and surveillance), aiming to assist the endoscopist in identifying lesions during the colonoscopy procedure by highlighting regions with visual characteristics consistent with different types of mucosal abnormalities that appear in the colonoscopy video. Highlighted regions can be independently assessed by the endoscopist and appropriate action taken according to standard clinical practice. The system is trained on video that may contain regions consistent with polyps, is limited to standard white-light imaging, and is not intended to replace histopathological sampling as the means of diagnosis.