Ibex Medical Analytics

Deep learning on whole slide images is the product. The prostate solution analyzes scanned histopathology images from core needle biopsies stained with hematoxylin and eosin, generating heatmaps that identify tumors potentially missed on initial assessment, with case and slide level alerts. The platform extends the same approach to breast and gastric tissue. There is no manual or rules based version of finding a small cancer a pathologist already looked past.

The cleared indication is one of the most conservative and best constructed in this index: the software functions as a second read safety net running after the pathologist's initial diagnostic assessment, flagging small and rare cancers that may have been overlooked. The pathologist reads first and reaches a diagnosis independently; the AI cannot anchor that judgment because it has not spoken yet.

Output is a heatmap directing attention to suspicious regions rather than a classification, which keeps interpretation with the specialist. Sequencing AI after human review rather than before is a deliberate design choice that eliminates automation bias, and few vendors make it.

Method and output are described concretely, covering input tissue preparation, heatmap generation, and the case and slide level alerting behavior. Validation was published in a peer reviewed venue, with a Lancet Digital Health study cited as supporting evidence alongside FDA submission materials.

The company is candid that its precision and clinical validation studies submitted to the FDA are unpublished, which is an honest disclosure of what remains unavailable rather than an implication that everything is public. Model architecture and performance figures are not published.

Two external certifications do substantive work here and it is worth being precise about what they cover and what they do not. Both require documented data handling, access control and risk management practices, and one of them is built specifically around healthcare information protection, so a buyer knows an outside assessor examined how data is handled rather than taking a marketing statement about security on trust.

That is meaningfully stronger than an assertion and weaker than a disclosure: a certification establishes that controls exist and were tested, and says nothing about what the company is permitted to do with the data those controls protect, which is the distinction this index has recorded elsewhere and which applies squarely here.

No retention period, deletion position or statement on whether customer slides contribute to model development was located, and no hosting arrangement or sub processor list was identified. The lane wide point applies too: whole slide image files routinely embed a photograph of the slide label carrying an accession number and often a patient name, so a laboratory should establish who strips those layers and at what point in the pipeline rather than assuming a de identification step exists. Ask for retention, the training position, the identifier stripping step, and a sub processor list.

Evidence combines peer reviewed publication with routine deployment scale. A Lancet Digital Health study supports the platform's diagnostic performance, and the company describes pathologists worldwide using the system as part of everyday routine for accuracy, quality control, turnaround time, and productivity, characterizing itself as the first and most widely deployed AI platform in pathology.

The gap is that the specific precision and clinical validation studies underpinning the FDA clearance remain unpublished by the company's own acknowledgment, so the strongest performance evidence is not externally reviewable.

No standalone PHI framework was located, but the certification set does substantive work here: HITRUST and ISO 27001 both require documented data handling, access control, and risk management practices, and HITRUST in particular is built around healthcare information protection. That is stronger assurance than a marketing statement about security, though it is not the same as published data retention or model training terms, which were not located.

No explicit HIPAA or BAA commitment was located, though HITRUST certification is closely aligned with HIPAA security requirements and is often accepted by covered entities as evidence of an adequate control environment. A US laboratory buyer should still establish business associate terms directly with this Israel headquartered vendor.

The most complete published certification set of any vendor in this index. The company states HITRUST, Cyber Essentials Plus, ISO 27001, and ISO 13485 certifications, which is notable both for breadth and for spanning two different assurance domains: HITRUST and ISO 27001 cover information security, Cyber Essentials Plus is a UK government backed technical control scheme, and ISO 13485 is the quality management standard for medical device manufacturing. Publishing this set plainly rather than gating it behind a sales conversation sets the bar the rest of this index falls short of.

FDA 510(k) cleared as an in vitro diagnostic medical device for the prostate product, the company's first FDA clearance, covering AI generated heatmaps identifying small and rare missed prostatic cancers. International registrations extend to CE-IVD certification under the IVDR framework plus registration with the UK MHRA, Australia's TGA, and Brazil's ANVISA.

Critically, the company states plainly that its prostate product is FDA cleared and its other solutions are Research Use Only in the United States, so a US buyer should understand the breast and gastric solutions are not available for clinical diagnostic use domestically. Drawing that line publicly is the practice this category requires.

No dedicated governance framework or bias evaluation was located, but ISO 13485 certification imposes a documented quality management system covering design controls and post market surveillance, which is a real governance structure even though it is not AI specific. Multi jurisdictional regulatory registration also implies validation acceptable to several independent authorities. No published subgroup or cross site performance analysis addresses scanner and staining variability.

One disclosure practice here is rare enough to name as the reason for the grade: the company states plainly that its precision and clinical validation studies submitted to the regulator are unpublished. Telling a reader which of your evidence they cannot see is an honest account of what remains unavailable, and it is the opposite of the far commoner posture of implying that everything relevant is public by listing only what is.

A buyer reading that knows exactly where the gap sits and can ask for the studies under agreement rather than assuming they do not exist. Around it, validation is published in a peer reviewed venue, and the method and output are described concretely, covering input tissue preparation, heatmap generation and the case and slide level alerting behaviour, so a pathologist can tell what the system produces and where it intervenes in their workflow.

Held below the top grade because model architecture and performance figures are not published, so the peer reviewed study substantiates the approach without giving an operating characteristic a laboratory could validate against, and no warranty, indemnity or remediation commitment attaches. Heatmap output is worth one practical question of its own, since a heatmap directs attention and a region left unhighlighted is a region a pathologist may examine less closely. Ask for sensitivity by lesion type, the studies under agreement, and what the system does with a slide it cannot assess.

The system operates on scanned whole slide images within pathology laboratory workflows and is described as integrated into pathologists' everyday routine across global deployments, which implies working laboratory information system and slide management integration. No named integrations, standards support, or partner platforms were enumerated in the materials reviewed.

The deepest deployment evidence in this lane, and most of it is verifiable through third parties rather than resting on the vendor's own account. The strongest single credential is a completed national rollout of Galen Prostate across the six university health boards of NHS Wales, which is a whole health system adoption rather than a pilot, and public health systems impose procurement, information governance and clinical safety scrutiny that commercial buyers generally do not.

Alongside it: a live deployment at University of Pittsburgh Medical Center, a network wide rollout with Alverno Laboratories covering roughly 30 hospital laboratories across Illinois and Indiana serving around 8 million people, and Galen Breast implementation through the iCAIRD centre in Glasgow. Interoperability is equally concrete and named rather than asserted.

The platform runs standalone or inside third party environments, with published integrations into Proscia Concentriq, Indica Labs HALO AP, PathPresenter and the Philips IntelliSite Pathology Solution, and connection to scanners, image management systems and laboratory information systems via an open API with DICOM support. Several of those integrations are documented in the partners' own announcements rather than only in Ibex material.

The company describes Galen as the most widely deployed AI technology in pathology, which is a vendor claim, but the named deployments substantiate the shape of it. The one half of this axis left unanswered is residency. No data residency commitment was located, no hosting arrangement or region is named, and nothing states where slide images are processed for customers outside the vendor's home jurisdiction. For a platform delivered as a service into national health systems, that is the question to put in writing.

No pricing is published by the vendor. No list price, no per case or per slide rate, no tier structure and no worked example was located for any of the Galen applications. Third party review material describes a software as a service model charged on annual usage, and reports that some customers have found the cost prohibitive for adopting the technology across all cases rather than selectively.

Both points are recorded here and neither is credited, because they come from an aggregator rather than from Ibex, and this index does not grade on directory or review site claims. They are noted because the second one, if accurate, describes a real procurement pattern worth testing: a per usage model can make selective deployment cheaper than universal deployment, which affects which cases actually get the benefit and is a question a laboratory should ask before signing.

The commercial position is otherwise inferable only in outline. A second revenue line exists through pharmaceutical collaboration, including work with AstraZeneca and Daiichi Sankyo on AI powered biomarker scoring for breast cancer, but no terms attach to it. Scope note: the vendor's own commercial pages were not opened on this pass, so this records that no pricing was located rather than proof that none is published anywhere.

A buyer should ask for the charging basis, whether it is per case, per slide or per seat, and specifically whether the model penalises running the tool across an entire caseload rather than on selected cases.

Three cancer types in routine clinical use, covering prostate, breast, and gastric, with functional coverage extending beyond detection into quality control measures and case review workflow. Geographic coverage is genuinely global, with regulatory registrations across the US, EU, UK, Australia, and Brazil. The important qualifier for US buyers is that only the prostate solution is cleared domestically; the rest is Research Use Only.