Radiology · Mammography screening reading support
Transpara
ScreenPoint Medical B.V.
The most rigorously trialled reading aid in screening mammography: across more than 105,000 women in the MASAI randomised trial, AI-supported reading found 29% more cancers at unchanged false-positive rates, cut screen-reading workload by 44% — and, the decisive result, did not let more interval cancers through.
Performance Metrics
Clinical Evidence
No mammography AI has been tested the way Transpara has. The MASAI trial — investigator-led at Lund University, independent of the manufacturer — randomised over 105,000 Swedish women to AI-supported screen reading (Transpara 1.7.0) versus standard double reading, and published its results in three instalments. The interim safety analysis (Lancet Oncology 2023, ~80,000 women) reported 244 versus 203 screen-detected cancers (6.1 vs 5.1 per 1,000) with recall unchanged at 1.5% in both arms and a 44.3% reduction in screen-reading workload. The full-enrolment performance paper (Lancet Digital Health 2025) confirmed a 29% higher cancer detection rate (6.4 vs 5.0 per 1,000) with a 51% rise in in-situ detection and no rise in false positives. The final results (The Lancet 2026) settled the question the trial was designed to answer: sensitivity 80.5% versus 73.8% at identical 98.5% specificity, with an interval-cancer rate 12% lower than double reading — non-inferior on the primary safety endpoint, and with fewer aggressive interval cancers. The wider evidence base points the same way, with honest exceptions. The AITIC paired trial in Córdoba (Nature Medicine 2026; 31,301 women, Transpara 1.7) cut radiologist workload 63.6% and raised cancer detection 15.2% — but its recall rate rose 14.8% and failed the prespecified noninferiority margin, with the inflation concentrated in 2D mammography rather than tomosynthesis. Real-world implementation in the Capital Region of Denmark (Radiology 2024; Transpara 1.7.1, live since November 2021) reduced recall 20.5%, raised the detection rate from 0.70% to 0.82% and cut reading workload 33.5%. A standalone evaluation of 122,969 BreastScreen Norway exams (Radiology 2022) and a retrospective triage simulation in Córdoba (Radiology 2021) round out the independent record; the foundational reader study — AUC 0.840 against 101 radiologists (JNCI 2019) — was manufacturer-authored, as early validation usually is. The structural caveat is version drift. The trial evidence rests on the 1.7.x generation; the currently marketed version, 2.1.0 (FDA-cleared November 2024, adding temporal comparison against prior exams), has no published outcomes of its own yet.
| Study | Design | n | Sensitivity | Specificity | AUC | Published |
|---|---|---|---|---|---|---|
| Lång K, Josefsson V, Larsson AM, et al. (Lund University; independent — MASAI interim safety analysis) | RCTRCT | 80,033 | — | — | — | Lancet Oncology, Aug 2023; 244 vs 203 cancers (6.1 vs 5.1 per 1,000), recall 1.5% in both arms, screen-reading workload −44.3% (Transpara 1.7.0) |
| Hernström V, Josefsson V, Sartor H, et al. (MASAI, full enrolment) | RCTRCT | 105,000 | — | — | — | Lancet Digital Health, Mar 2025; >105,000 women; cancer detection 6.4 vs 5.0 per 1,000 (+29%), in-situ detection +51%, no rise in false positives |
| MASAI trial investigators (Lund University / Region Skåne; final results) | RCTRCT | 105,000 | 80.5% (vs 73.8% double reading) | 98.5% (identical in both arms) | — | The Lancet, Jan 2026; interval-cancer rate 12% lower (non-inferior on the primary safety endpoint), fewer invasive and non-luminal-A interval cancers |
| AITIC trial investigators (Reina Sofía University Hospital, Córdoba; paired noninferiority trial) | ProspectiveProspective | 31,301 | — | — | — | Nature Medicine, 2026; radiologist workload −63.6%, cancer detection +15.2% (Transpara 1.7) — but recall +14.8%, failing the prespecified noninferiority margin, driven by 2D mammography |
| Lauritzen AD, et al. (Capital Region of Denmark; real-world implementation, before/after) | RetrospectiveRetrospective | 119,000 | — | — | — | Radiology, Jun 2024; recall −20.5% (3.09%→2.46%), cancer detection 0.70%→0.82%, false positives −32%, reading workload −33.5% (Transpara 1.7.1, live since Nov 2021) |
| Larsen M, et al. (Cancer Registry of Norway; independent standalone evaluation) | RetrospectiveRetrospective | 122,969 | — | — | — | Radiology, Jun 2022; 122,969 BreastScreen Norway exams scored on Transpara's 1–10 exam scale; high-score thresholds captured most screen-detected and many interval cancers |
| Rodriguez-Ruiz A, Lång K, Gubern-Mérida A, et al. (ScreenPoint-affiliated — manufacturer-authored) | RetrospectiveRetrospective | 2,652 | — | — | 0.840 (vs 0.814 radiologist mean) | JNCI, Sep 2019; multi-reader multi-case study, 9 datasets, 7 countries, 101 radiologists; AI outperformed 61.4% of readers |
Clinical Pulse
Verified clinician sentiment
Verified clinician reviews launching soon · Apply to contribute →
Reviews are signed by clinicians verified against their regulator's public register.
Inside the algorithm
Editorial featureHow Transpara 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
Screening mammograms arrive from compatible 2D and 3D systems.
Transpara receives screening exams as studies from the — full-field digital mammography and, since the 2020 clearance (K193229), digital breast tomosynthesis from compatible systems. Compatibility has widened across successive clearances (Fujifilm FFDM support arrived with K192287 in 2019). Analysis runs in the background; the source images are not modified and no additional imaging is required.
Input
Mammography / DBT DICOM
Inference
Background, per-exam
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 IIb
K241831
Class II
Safety Record
No recalls, field safety notices or FDA safety communications attributable to Transpara were found in indexed public sources as of July 2026 — read as "none found", not an exhaustive audit, since the MAUDE and FDA recall databases could not be queried directly from this environment. The stronger safety evidence is affirmative: MASAI's primary endpoint was designed to detect the harm a reading aid could cause — cancers surfacing between screens — and the AI arm's interval-cancer rate was 12% lower than double reading's. The residual risks are the familiar ones. A low exam score is not a normal-anatomy finding, and in triage configurations it removes a second human read from most screens; the AITIC trial is a reminder that behaviour differs by modality, with recall inflation on 2D mammography that failed the trial's noninferiority margin.
Intended Use & Indications
Transpara is a software-only device used by physicians interpreting screening full-field digital mammography and digital breast tomosynthesis exams from compatible systems. Per the FDA 510(k) record, it identifies regions suspicious for breast cancer — calcification groups and soft-tissue regions — and assesses their likelihood of malignancy, presenting per-region scores together with an exam-level score indicating the likelihood that cancer is present in the exam. Unusually for this catalogue, the broad concurrent-reading indication is essentially the same in the US and the EU; the difference is in use. American sites deploy it as a concurrent reading aid, while several European screening programmes additionally use the exam score to route low-risk screens to single reading — the configuration tested in the MASAI and AITIC trials and running live in the Capital Region of Denmark since November 2021. In every setting the interpreting physician remains the decision-maker.