ScreenPoint Medical

Every product ScreenPoint sells is a model. Transpara Detection scores each mammogram or tomosynthesis exam for the likelihood of cancer and marks suspicious regions, Transpara Density assesses breast density, and Temporal Comparison adjusts findings against up to three prior exams.

The software runs as a processing appliance that returns results in the standard mammography CAD format to the workstations and archives a site already uses, and the user interface is supplied by those workstation vendors. There is no hardware, reading service or workflow platform underneath, so removing the models leaves nothing to sell.

The US label keeps the physician reading every exam, the thresholds are published, and the company also promotes a use in which low risk exams get no human read. Transpara is cleared as a concurrent reading aid (K241831): it is not intended to replace a physician's review, and patient management decisions should not rest on it alone.

The scores and operating points are stated: a 1 to 10 exam score, low, intermediate and elevated risk categories, and sensitivity published at 70, 80 and 97 percent specificity. In the MASAI trial the score routed low and intermediate risk exams to one reader and high risk exams to two, with interval cancers measured afterward.

The company's evidence page now features AITIC, a prospective study in Spain in which low risk exams were excluded from human review, under the heading 'Autonomous AI' for low risk cases, and markets low risk exams as unmarked for expedited review with a 99.97 percent negative predictive value. Nothing published says under which label, in which country or on what conditions a program may run that configuration, or who reviews the excluded exams.

The operating characteristics are published in detail; the model and its training data are not. The score scale, risk categories, the operating points (70, 80 and 97 percent specificity) and the extra high risk threshold are stated, the 510(k) summary names the version (Transpara 2.1.0) and describes what changed from the predicate (detection components updated for mammography and tomosynthesis, temporal comparison added), and the peer reviewed trials name the version used.

The company says it runs routine algorithmic performance updates. What is not described is the architecture beyond deep learning, the training data (sources, countries, equipment, time period, population) or how customers are told when a model changes.

A structurally short chain for the model, with the delivery layer partly named. ScreenPoint develops Transpara in house, founded on its founders' computer aided detection research, and the model a site runs is a cleared, versioned release (2.1.0 in the current US clearance) that changes through the company's own updates. The software returns results to third party workstations and PACS, whose vendors supply the reading interface, and is sold partly through partners.

The company does not say whether processing ever runs in a hosted service, or whose it would be. There is no subprocessor list, and no statement on whether customer images reach model development. The population sharpens that last point: women who attended a screening invitation consented to screening, not to anything else done with their images.

A randomized controlled trial inside a national screening program backs the headline claim. MASAI (NCT04838756), led by Kristina Lång at Lund University, randomized more than 105,000 women at four sites in southwest Sweden between Transpara supported screening and standard double reading.

The clinical safety analysis (The Lancet Oncology, 2023) reported a 44 percent reduction in screen reading workload; a secondary analysis (The Lancet Digital Health, 2025) reported a 29 percent increase in cancer detection (338 cancers among 53,043 participants) with no rise in false positives and more small, node negative invasive cancers; final results (The Lancet, January 2026) reported a 12 percent lower interval cancer rate meeting non-inferiority, 27 percent fewer aggressive interval cancers, sensitivity of 80.5 against 73.8 percent and specificity of 98.5 percent in both arms.

A prospective Spanish study (Nature Medicine, 2026) and a Dutch retrospective comparison with double reading (The Lancet Digital Health, 2025) add to it, and a US randomized trial (PRISM) is under way. The 510(k) summary reports standalone results on 10,207 exams. Limits: the trial ran in one national program, and a UCLA estimate of 30 percent fewer interval cancers is a model, not a trial.

General statements only; nothing says what happens to screening images in the product. ScreenPoint's privacy statement (last updated February 2025) covers website visitors, business contacts and applicants, and says the company also processes personal data on behalf of its customers as a data processor, for example when handling support and complaints about processed images, but that the statement does not apply to that processing.

The 510(k) summary describes processing in a standalone appliance that returns results over DICOM. Not stated: whether images or scores are retained after processing, whether they are de-identified, and whether customer images are used to develop models, which matters for a screening population that agreed to be screened rather than to anything else. Safety engineering, meaning how a missed cancer is handled as a safety event, is not published.

A processor role is stated for customer images, with nothing published behind it. ScreenPoint's privacy statement says the company processes personal data on behalf of its customers as a data processor, for example in supporting hospitals and clinics that use Transpara, and that the statement itself does not cover that processing. No data processing agreement, business associate agreement or HIPAA statement is published, although the company sells in the US with a US support line. Whether support staff can reach identifiable images is not stated either.

Controls are asserted with nothing independent behind them. ScreenPoint's Security Reporting page says its security program is aligned with ISO/IEC 27001 and invites researchers to report vulnerabilities by email under a responsible disclosure program, with no bug bounty. Being aligned with a standard is not certification to it, and a reporting channel is not an audit.

The CE certificate (notified body 0344) and FDA clearances cover the device, not the company's information security, and no SOC 2 report, ISO 27001 certificate, penetration testing cadence or trust portal is named on its security, privacy or about pages. Nothing independent is published.

The US position is identifiable and matches the label. Transpara 2.1.0 was cleared on 25 November 2024 (K241831; 21 CFR 892.2090, computer assisted detection and diagnosis, Class II, product code QDQ) as a concurrent reading aid for physicians interpreting screening full field digital mammography and tomosynthesis from compatible systems, marking suspicious calcification groups and soft tissue regions with scores and an exam score, and adding temporal comparison with prior exams.

It is not intended to replace a physician's review, and patient management decisions should not rest on it alone. The company also lists K232096 and CE certification through notified body 0344, and says not all features are available in every country.

The trial workflows it promotes sit outside that US indication: routing low risk exams to a single reader, as in MASAI, is a screening program decision, and excluding low risk exams from human review, as in the Spanish AITIC study, is a research configuration.

A subgroup evaluation is described with its groups and method; the results by group are not published. ScreenPoint's 510(k) summary for Transpara 2.1.0 (K241831) reports that area under the curve and sensitivity at the three operating points were compared across ethnicity (White non-Hispanic, White Hispanic, Black, Turkish, Asian, other), age above and below 50, lesion size, lesion type, histology, screening against diagnostic exams, and low against high breast density, with the number of cancers in each group, and states that no group showed a deviation and all met the algorithm requirements.

The figures for each group are not given, the requirements are not stated, and the Black and Turkish groups hold only 37 and 27 cancers, so no difference may reflect too few cases to see one. The product page claims Transpara is robust regardless of ethnicity, age, BMI and density; BMI is not among the groups the summary's evaluation covers, and the claim is a statement rather than a result. A triage score that sends low scoring exams to one reader or none needs to be shown not to score dense breasts low for the wrong reason. Groups and counts disclosed, only the conclusion published.

An error rate with its method and denominator is published for the lead product; nothing else stands behind the output. ScreenPoint's own 510(k) summary for Transpara 2.1.0 (K241831, cleared 25 November 2024) reports a standalone test on 10,207 mammography and tomosynthesis exams, 1,350 with biopsy proven cancer and normal exams mostly confirmed by at least a year of follow-up, drawn from centers in seven EU countries and the US and not used in development.

Sensitivity for 2D mammography was 97.4 percent at 70 percent specificity, 95.2 percent at 80 percent and 80.8 percent at 97 percent (the elevated risk threshold); for tomosynthesis 96.9, 95.1 and 78.4 percent. That is a test set result for cancer detection, not performance in a buyer's program, and it does not cover the density product. The MASAI trial also measured interval cancers, the misses that surface between rounds. No warranty, indemnity or remediation commitment is published.

Integration runs through the imaging standard, with the output format named. Transpara runs as a processing appliance that accepts mammograms over DICOM and sends results in the standardized mammography CAD DICOM format to workstations, PACS and RIS; the reading interface is supplied by workstation vendors, and the extra high risk flag reaches the reading worklist.

The company names users in the US (University Radiology Group, University Diagnostic Medical Imaging) and in national screening programs in Europe. No record system integration is claimed, which fits a screening read that ends in the radiology report. A named standard with deployments behind it.

A single delivery model is described, with where it runs implied rather than committed. The 510(k) summary says Transpara results are computed in a standalone processing appliance that receives mammograms over DICOM and returns results to workstations, PACS and RIS, configured through a service interface.

Whether that appliance sits on the customer's premises or in a hosted service, where images rest, and how national programs that require images to stay in country are served is not published. The privacy statement says hosting and other IT providers can receive personal data collected through the website, which does not reach the product. One model, location implied.

There is no published price, though the way the product reaches buyers is easy to trace. ScreenPoint sells Transpara Detection, Density and Temporal Comparison for mammography and tomosynthesis to national screening programs, hospitals, radiology practices and imaging groups in more than 30 countries, directly and through partners, with the reading interface supplied by workstation vendors. The route in is a demo.

Nothing says whether the price is per exam, per site or per program, whether tomosynthesis costs more than 2D, or how volume changes it. The economic case is easy to calculate, since a 44 percent reduction in reading workload converts to radiologist hours at a known rate, which makes the missing unit price matter.

Focused on breast imaging, with validation behind the screening setting and less behind diagnostic use. Transpara covers 2D mammography and tomosynthesis, density assessment and comparison with priors, in national screening programs (Sweden, the Netherlands, Denmark, Norway, Spain), US single reader tomosynthesis practices, hospitals and imaging groups across more than 30 countries, with over 13 million mammograms processed.

Screening is validated in a randomized trial and prospective studies in several countries; the 510(k) test set spans screening and diagnostic exams from seven EU countries and the US. The company also markets Transpara for diagnostic imaging, where the published work is thinner (a retrospective US study on screening to biopsy). The narrow scope is not the limit; the gap between screening and diagnostic validation is. Named coverage with validation behind part of it.