Pramana

A slide scanner whose imaging stands without the models, with learned functions layered on it. The scanners' core is optical and mechanical: volumetric imaging that fuses z-stacks to replicate a microscope's fine focus, covered by the company's granted US patents, modular scan heads and slide handling. On top of that the company describes real-time AI: quality control that detects artifacts such as bubbles, folds, debris and faint tissue and triggers rescans, intelligent tissue detection and region selection, and edge algorithms for feature detection and quantification during the scan. Those are image recognition functions, so they count as learned components. Algorithms from partners that run on the scanner, such as the hematopathology models ARUP is to develop, are the partners' own, not the company's. Remove the models and a working scanner remains, needing an operator to check its output.

Artificial intelligence is a feature layer on a product whose value stands without it.

Autonomous quality control is claimed, and the check it replaces is presented as unnecessary. The scanner detects imaging errors and rescans on its own, and the company says this eliminates the cost and time of manual quality checks on the output before images are used downstream, stating fewer than 0.5 percent imaging errors 99 percent of the time. The models decide how to image, not what the image shows, and pathologists still read the slides. But the company's own Mayo Clinic pilot found, in a manual review of 100 selected slides, two with significant focus or stitching errors that the software had not flagged. Nothing published sets the threshold at which a slide is failed, describes what a laboratory sees for each slide, or says how an image passed in error is caught before diagnosis.

Autonomy claimed with no described control on the errors it misses.

The models are described by function only. The company names what its real-time algorithms do (artifact detection with rescans, tissue detection and region selection, feature detection and quantification at the edge) and publishes its optical method, volumetric imaging with z-stack fusion, backed by granted patents. No model type, training data, error classes by stain or specimen, version or update practice is published for the quality or tissue detection models, and nothing says how a model update reaches installed scanners.

Architecture described in general terms with nothing identified.

A short chain: the models run on the scanner in the customer's laboratory. Pramana's quality and tissue detection models are its own and execute at the edge during scanning, and the company argues against cloud analysis, so in the scanner product images need not leave the laboratory for the AI to run. Partner algorithms can also run on the scanner, with ARUP Laboratories named as developing hematopathology models for that route. What is not stated: whether any image or quality data returns to Pramana or Evident for support or model development, and, for the digitization service, where images are held and by whom.

A structurally short chain with the transfer not needed for the scanner itself.

One named deployment with dated figures and method, and no diagnostic evidence. In a Mayo Clinic pilot of the research use SpectralHT four-head system, 23,916 archival slides from the 1950s to the present were scanned over 30 days by a single operator, about 800 a day, with no cleaning or preparation. The overall acceptability rate was 99.5 percent, with 118 slides of unacceptable quality; focus errors were flagged in 96 slides and stitching errors in 34. Mayo then awarded the company a multi-million slide archival contract through a competitive tender. Moffitt Cancer Center has since selected the scanners. No concordance study shows that diagnoses from these images match glass or other scanners.

A named deployment with dated outcome figures and enough method to test them.

Nothing published describes how patient images are handled. The Pramana site's privacy policy applies only to information collected through the website, and the footer now links Evident's website privacy notice, which covers site visitors and customers' contact data. Neither addresses slide images, label photographs, or data a scanner may send back for support. For the digitization service, in which an institution ships glass slides and receives images, nothing states who holds the images, for how long, whether Pramana or Evident may use them, or how slide labels carrying patient identifiers are handled. No safety engineering is published.

Nothing published on how protected information moves through the products.

No status is stated and no privacy document reaches the products. The only privacy documents are the Pramana and Evident website notices. A scanner running in a customer's laboratory may never give the company access to protected health information, but remote support could, and the digitization service, in which a health system ships patient slides to be imaged, plainly handles it. No business associate position is stated for either route.

No statement of status and no privacy document that reaches the product.

Nothing independent is published. The Pramana site is three pages with privacy, terms of use and cookie settings, and neither the Pramana privacy policy nor Evident's product page for the scanners names a security certification, trust center or penetration testing. Evident's product page says the scanners are designed for secure data management without detail. A scanner on a hospital network that accepts third-party algorithms and may be reached for remote support is a device whose software security matters.

Nothing independent published.

A US pathway in progress, with authorizations elsewhere. Evident's product page states that the scanners are CE marked under the IVDR and licensed by Health Canada, are for research use only in the United States, and are pending FDA 510(k) clearance; the Health Canada licenses for SpectralM and SpectralHT Cubiq were announced in July 2025. The Pramana site carries the US research use only statement on its product section, and the Mayo Clinic pilot used a research use only system. FDA's databases list no Pramana clearance yet. The authorizations cover the scanner as a whole; partner algorithms running on it carry their own status.

A pathway stated and in progress.

One test of the quality models is published, with nothing across slide types. The Mayo Clinic pilot reports the models' flags, and a manual review of 100 selected slides that confirmed every flagged error and found two significant errors the models missed. No results are given by stain, specimen type, section quality or slide age, although the pilot spanned archival slides from seven decades and the scanners are marketed for histology, cytology, hematology and microbiology. Because every downstream analysis model consumes these images, uneven quality control across laboratories or preparations would pass into every analysis built on them. No governance framework or monitoring commitment is published.

Testing published without an evaluation by group.

An error characteristic for the quality models is published, and nothing stands behind it. The company's Mayo Clinic case study states that a manual review of 100 selected slides confirmed every error the software had flagged and found two slides with significant focus or stitching errors it had not flagged, so a buyer has a miss count with its method and denominator, from a small selected sample. The general claim of fewer than 0.5 percent imaging errors 99 percent of the time has no method attached. No warranty, service level or remediation commitment attaches to an image passed in error, and nothing published covers custody, loss or damage of glass slides sent for the digitization service, where archival slides may be irreplaceable.

A published error rate with its method and denominator.

Open image formats and a named image management integration. The scanners output DICOM, OME-TIFF and DZI, and Pramana joined Proscia's Proscia Ready program in 2024, so images can flow into Proscia's Concentriq platform; Evident states the scanners are designed for image management integration. Evident also names a collaboration with Siemens Healthineers at Klinikum Ludwigshafen. No laboratory information system integration is named, so how slide labels, case numbers and scan status reach the laboratory system is not described.

A standard supported with a named system and deployments behind it.

On premises by design for the scanners, with the service route left open. The scanners run their models at the edge inside the customer's laboratory, and the company describes cloud analysis as slow and costly, so images from a purchased scanner rest wherever the laboratory stores them. The digitization service, used for the Mayo Clinic archive, images slides for an institution, but where those images are processed and stored, and for how long, is not stated.

Options stated with residency committed only for the on premises route.

No price is published, and how the scanners are sold can be established. The scanners are sold through Evident in modular configurations, from the two-slide or four-slide desktop SpectralM to SpectralHT systems of two or four heads holding 240 to 960 slides, and the company argues that labs overbuy capacity with single-head scanners. The Mayo Clinic engagement shows a digitization service sold by project. No scanner price, service rate per slide, service contract or storage cost is published.

The sales posture is discoverable, with no price.

Broad specimen coverage named, with validation behind the archival histology case. The company lists anatomic pathology, cytopathology, hematopathology, clinical pathology and microbiology, and specimens including biopsies, frozen sections, liquid based cytology, bone marrow aspirates, blood smears and microbiology slides, in brightfield at up to 80x. The Mayo Clinic pilot supports archival histology at scale; no results are published for cytology, hematology or microbiology slides, and the US use is research only.

Coverage named with validation behind part of it.