Riverain Technologies

The software is the product. ClearRead applies deep learning vessel suppression to make nodules visible that vessels would otherwise obscure, paired with detection, and the company sells no scanner or services layer. The vessel suppression approach is itself the technical thesis: rather than only classifying findings, the model alters the image so both the algorithm and the human read better, which is a distinct mechanism from pure detection overlays.

Notable for the concurrent read model specifically. Most computer aided detection is cleared as a second read, running after the radiologist's initial interpretation to avoid biasing it; ClearRead was cleared to support concurrent reading, meaning output is available alongside the initial read.

That is a deliberate design tradeoff, faster workflow against greater exposure to automation bias, and it sits at the opposite end from Ibex, whose second read sequencing structurally eliminates that bias. The radiologist retains the diagnosis in both cases, but buyers should understand the difference, because concurrent presentation changes how much the tool can influence the human.

The clearance documentation is inspectable and reasonably detailed, including performance estimates broken out by nodule type, residual analysis showing little performance impact across slice thickness values, and disclosure of the micro nodule filtering option with its site configurable threshold capped at 5mm to keep behavior within labeled performance. Acquisition normalization is described as handling scans across manufacturers and protocols without per device tuning. Training data scale and composition are not published, and there is no third party head to head benchmarking, so transparency is strong in the regulatory record and lighter in marketing materials.

The architectural point that matters here is the concurrent read design, and it belongs on this axis as well as on autonomy because it changes what the vendor holds. This product presents findings to the radiologist alongside the study during the primary read, rather than filtering or triaging a worklist beforehand, which means the vendor receives studies while they are being actively reported rather than afterwards from an archive.

What is held is therefore live work in progress rather than historical records, and that makes retention terms and access controls worth reading closely rather than assuming: an unreported study in a vendor's possession is a different object from a completed examination, because the finding has not yet been adjudicated and the patient has not yet been told anything.

Nothing was retrieved on retention of studies or derived outputs, de identification, training use, or deletion at contract end, and no model, hosting arrangement or sub processor list was named. The training question is unaddressed as it is almost everywhere in this lane, and a company developing detection for this modality has an evident interest in the studies flowing through its deployments, with nothing stating whether customer images contribute or whether a site can decline.

Operating in a single jurisdiction removes the layer that complicates several peers, so the answers should be simpler to obtain. Ask what is retained after a read, whether studies train models, and what is deleted at termination.

The core evidence is a multi reader multi case clinical trial supporting the original clearance, reporting a 29 percent reduction in missed actionable nodules, which measures the right thing for a detection aid: change in reader performance, not standalone algorithm accuracy.

An independent study led by University of Washington School of Medicine radiology researchers examined the vessel suppression product, and the company reports commercial traction including expansion across eight countries and roughly 50 new customers in 2024. The evidence base is real but thinner and older than newer entrants with dozens of publications, and outcome level data is absent.

Converted from Not Rated. No stewardship terms were located.

Nothing was retrieved on retention of studies or derived outputs, de identification, model training use, or deletion at contract end.

The architectural point that matters here is the concurrent read design, and it belongs on this axis as well as on autonomy. This product presents findings to the radiologist alongside the study during the primary read rather than filtering or triaging the worklist beforehand. This index treats that as a deliberate contrast with second read and autonomous approaches, and it has a data consequence: the vendor receives studies while they are being actively reported, not afterwards from an archive. What is held is therefore live work in progress rather than historical records, which makes retention terms and access controls worth reading closely rather than assuming.

The training question is unaddressed as it is almost everywhere in this pass. A company developing chest radiograph detection has an evident interest in the studies flowing through its deployments, and nothing states whether customer images contribute to model development or whether a site can decline.

Being domestic removes the jurisdictional layer that complicates several peers, so the answers here should be simpler to obtain.

Ask what is retained after a read, whether studies train models, and what is deleted at termination.

Converted from Not Rated. No published position was located, and unlike most records in this pass there is no jurisdictional complication to explain it.

No business associate agreement, addendum, role statement, subcontractor flow down, breach notification timetable or review cadence was retrieved.

The role is unambiguous and domestic. The hospital or imaging provider is the covered entity, and a vendor receiving chest studies and returning findings processes protected health information on its behalf as a business associate with direct liability under the rule. The company is United States headquartered selling into United States radiology, so the questions this index has had to raise about contracting entities, cross border processing and reconciling two regimes simply do not arise here. That narrows the gap to a straightforward publishing choice.

Agreements plainly exist, since the product is deployed in United States hospitals and no radiology department contracts without one. None is public, so a buyer cannot see the terms before entering a sales process, and this index's route to a higher grade is publishing the instrument.

The concurrent read design means the vendor receives studies during active reporting rather than after the fact, which makes retention and access terms worth reading closely rather than assuming.

Ask for the agreement, the subcontractor terms, and what is retained after a read completes.

Converted from Not Rated, and the prior note draws the right line, which this index has now drawn against six different credential families.

No SOC 2 of either type, no ISO 27001, no HITRUST, no trust centre and no penetration testing statement was retrieved.

What the submission references is ISO 13485 with participation in the medical device single audit programme. Both are real and both are quality management: they govern whether the software is designed, manufactured and maintained under control, and the audit programme lets one inspection satisfy several regulators. Neither examines encryption, access control, code security, monitoring or incident response. This index applied the identical reasoning to a dental imaging vendor listing the same two standards among its security credentials, and the reading holds here.

The concurrent read architecture is worth noting because it shapes what a breach would mean rather than how likely one is. This product presents findings to the radiologist at the same time as the study rather than filtering the worklist, so it sits inside the primary read rather than beside it, and the images it holds are the images being actively reported.

The company is United States headquartered and selling into United States radiology, so no cross border complication explains the absence.

Ask what independent security assessment exists and under which standard.

Genuinely cleared and maintained over time, though this is an older generation of authorization than most vendors indexed here. ClearRead CT was cleared in 2016 as the first device supporting concurrent reading with detection across solid, subsolid, and ground glass nodules, with subsequent clearances including ClearRead CT Compare, K221612, and a 2025 clearance extending to coronary artery calcium scoring on ungated non contrast scans.

Available for sale in the United States, Canada, and Europe. Solid multi clearance record; graded B rather than A because the portfolio is narrow and the foundational clearance predates the current generation of imaging AI standards such as Predetermined Change Control Plans.

The clearance record includes performance analysis across slice thickness and across scans from multiple manufacturers and acquisition protocols, which addresses technical generalization. What is absent is any governance framework, post market monitoring commitment, or demographic subgroup analysis. Technical robustness tested, population level performance unexamined publicly.

The clearance documentation is inspectable and detailed, and one design choice inside it deserves naming as a benchmark. Performance estimates are broken out by finding type rather than aggregated, and a residual analysis shows little performance impact across slice thickness values, which addresses the acquisition variable a radiology department would actually vary.

The choice worth naming is the micro nodule filtering option: it is site configurable and the threshold is capped so that behaviour stays within labelled performance. A capped configurable is a materially better design than either a fixed setting or an open one, because it gives a department the flexibility it needs while making it impossible to operate the device outside the range in which its performance was established.

Most configurable products in this index let a site set a threshold anywhere and publish nothing about what happens at the extremes, so the operating point that was validated and the one in use can silently diverge. Acquisition normalisation across manufacturers and protocols without per device tuning is also described.

Held below the top grade because training data scale and composition are unpublished, no third party head to head benchmarking exists, and no warranty, indemnity or remediation commitment attaches, so transparency is strong in the regulatory record and lighter everywhere else. Ask for training composition, and performance at the configurable threshold's cap.

Deployment friction is low by design. The software is described as installing into the clinical environment without new hardware and without customized tuning to specific devices or protocols, processing CT scans across a wide range of manufacturers and acquisition protocols, and the company markets it as an enterprise ready solution across a healthcare network. This is PACS and imaging workflow integration; no EHR integration is claimed or applicable.

Described as installing into the existing clinical environment without new hardware, which implies on premise or near premise deployment and therefore likely keeps imaging data inside the institution, an advantage over cloud only competitors for residency sensitive buyers. Specific tenancy, hosting, and residency terms are not published, so this is inferred from deployment description rather than stated.

No pricing is published. The company discloses growth figures including 147 percent growth in 2024, expansion across eight countries, and nearly 50 new customers added, which speaks to trajectory but not cost. Nothing on per study, per site, or licensing structure.

Deliberately narrow: cardiothoracic imaging, centered on lung nodule detection in chest CT and radiography, with a recent extension into coronary artery calcium scoring that enables opportunistic cardiac screening from scans acquired for other reasons. That opportunistic screening angle is a genuine workflow advantage, but coverage does not extend beyond the chest, and the vendor does not compete on breadth with multi finding platforms covering neuro, musculoskeletal, and emergency triage.