aetherAI

Two kinds of product under one name. aetherSlide is a digital pathology image management and viewing system, and its FDA clearance is for that function, so a laboratory could run it with no model at all. Beside it the company sells models of its own: aetherAI Hema, bone marrow smear analysis software, and aetherAI Endo, computer-aided polyp detection for colonoscopy, plus research applications for slide quality control, region recognition, cell counting and IHC quantification. The bone marrow and whole slide models are the company's own work, described in its published research.

The model is the engine of core modules while the platform carries value without it.

Assistance is implied by the product names, and no review mechanism is described. aetherAI Endo is sold as computer-aided detection, and the research applications are said to give reference support for image research and data analysis. For aetherAI Hema, which produces a bone marrow differential count, nothing published says who confirms the count, whether each classified cell is shown for review, what confidence level routes a smear to a manual count, or what happens to cells the software cannot classify. That matters because in the company's own validation study the research model from which the product derives classified 57.7 percent of blasts correctly, and blasts are the cell class whose count most often changes a diagnosis.

Oversight implied without a described control.

The approach is published, with no product version or update practice. The company's bone marrow model is described in a peer reviewed paper (International Journal of Medical Informatics, 2025) as a Mask R-CNN network that detects nucleated cells and sorts them into 15 classes, trained on 542 slides and 597,222 expert-annotated cells stained with Liu's or Wright-Giemsa stain; the paper states that aetherAI and National Taiwan University Hospital own the product derived from the study. Its whole slide method, a deep network trained on entire gigapixel slide images, is described in Nature Communications research on lymph node metastasis detection and, the company says, covered by two US patents. No product version, change history or retraining practice is published for Hema, Endo or the research applications.

Approach named without the version and update discipline behind it.

An in-house build, with the hosting path unstated. The company develops its own models, described in its published research and developed with named hospitals, so no outside model provider sits between a slide and a result. aetherWeb, the education platform, is described as cloud based without a host named, and nothing states where Hema or the research applications run, whether customer images reach aetherAI for support or model development, or which subprocessors apply.

A structurally short chain at the model layer, with hosting and data terms unstated.

Published multi-site research on the bone marrow model, short of independent validation of the product. In the International Journal of Medical Informatics (2025), the model classified 88.1 percent of 200,639 cells correctly in a retrospective validation on 254 patients at three Taiwanese centers and one US laboratory, with recall by class ranging from 0.97 for erythroblasts to 0.577 for blasts, and differential count percentages correlated with manual counts (rho above 0.8) for most cell classes. The corresponding author is at National Taiwan University Hospital, aetherAI employees are coauthors, and the paper describes the product as derived from the study rather than naming a version. The lymph node metastasis model in Nature Communications is research, not a product on sale. aetherSlide deployments are named at Chang Gung Memorial Hospital, National Taiwan University Hospital, the university hospitals of Tübingen, Frankfurt and RWTH Aachen, and National Cancer Center Hospital East in Japan, among others; those record adoption, not outcomes.

Published research short of independent validation.

Nothing published describes how patient images or results are handled. The only privacy policy covers visitors to the company website, defining the service as the website, and does not address slides, smears or colonoscopy video processed by the products. Nothing states whether aetherAI receives customer images for support or model development, what it retains, who de-identifies images, or how customers are told about model updates. Whole slide files embed a photo of the slide label by default, which often carries the accession number and patient name. No safety engineering is published either.

Nothing published on how protected information moves through the products.

No position is published under HIPAA or any equivalent regime. The only privacy policy covers the company website. The company names US partners, including UPMC, and its validation study used slides from a US laboratory, yet nothing says whether aetherAI signs business associate agreements, whether its products process protected health information outside the customer's site, or what data protection terms apply for customers in Taiwan, Japan or Europe. Whether an agreement is needed turns on whether images leave the customer's environment, including for remote support from Taiwan, and nothing published answers that.

No status stated and no privacy document reaching the products.

An information security certification is named without the artifact. The company's about page lists ISO/IEC 27001 among its certificates, and its timeline records the certification as a milestone, alongside ISO 13485 for its quality system and Taiwan medical device licenses. ISO/IEC 27001 is an audit of how the company runs security; ISO 13485 is not.

No certificate, scope or expiry is published, there is no trust center, and the site carries no penetration testing or vulnerability disclosure statement. The same list includes a D-U-N-S registration, which is a business identifier rather than a security credential.

A platform clearance, and non-US authorizations for the AI with no US claim for any model. aetherSlide, the image management and viewing system, was cleared by FDA on 30 May 2024 (K233126, product code QKQ), and the company also lists Taiwan approval and CE marking for it. That clearance covers the viewing platform, not an algorithm. aetherAI Hema is listed as TFDA approved (License 007251) and CE marked, and aetherAI Endo as TFDA approved (License 007508); neither is claimed to hold FDA clearance, and the research applications are framed as support for image research. The company states CE marking without saying whether under the older directive or the IVDR. FDA's databases list no other aetherAI device.

Non-US authorizations named for the AI, with the marketing matching them, and a US clearance for the platform only.

Nothing is published on how the products behave across patient groups, and no governance framework or monitoring commitment is named. The published research gives a partial view: the bone marrow study reports results for Liu's and Wright-Giemsa stained slides separately, a condition that changes how cells look, but it evaluates the research model from which the product was derived rather than a named product version, and its validation data come mostly from Taiwan with one US laboratory. No results by site, age or diagnosis group are presented, and none for the colonoscopy product. Slide appearance varies with each laboratory's stain, processing and scanner, so a model built largely on one country's slides can lose accuracy elsewhere in ways that look like ordinary difficult cases.

No evaluation of the product by group published.

A clinician checks the output, and nothing stands behind it. aetherAI Hema produces a bone marrow differential count, and the published method locates each cell it classifies, so a hematopathologist can inspect individual calls, but no warranty, correction commitment or remediation route attaches to a wrong count. The error figures in the company's published study (accuracy 0.881, blast recall 0.577) are for the research model from which the product derives, not a named product version, so they do not stand as the product's error rate. No error figure is published for the colonoscopy product or the research applications.

Review of a decision support output with nothing attached to it.

Integration with imaging equipment is described, with no laboratory or record system named. aetherSlide manages and displays scanned slides, and the company describes its software as integrating with healthcare institutions' existing imaging equipment. With Techman Robot and Hamamatsu Photonics it offers an automated arrangement in which a robot loads slides into a NanoZoomer scanner and images pass to aetherAI software. No laboratory information system, record system, standard or interface is named, so a buyer cannot check how cases, orders or results move between aetherSlide and the systems around it.

Integration claimed with no record system named.

Hosting and location are left open. aetherWeb, the education platform, is described as cloud based; the clinical software is described as integrating with equipment inside the customer's institution; and the automated scanning arrangement sits in the laboratory. Which of these applies to Hema and the research applications, where any hosted component runs, and what is retained after analysis are not stated. The company operates from Taiwan and sells into Japan and Germany and names US partners, so where analysis runs and where support staff can see images are separate questions from where the customer is.

Hosting implied rather than committed.

No price is published, and the company's finances are partly public. aetherAI has traded on Taiwan's Emerging Stock Board since 26 November 2024 (code 7803), applied on 21 November 2025 to list on the Taiwan Innovation Board, and names Quanta Computer, CDIB Capital, Cathay Venture and Taiwan Business Bank VC among its investors. The platform, the algorithm products and biopharma services are all sold through a sales contact, and nothing published describes the unit of charge for any of them.

Sales posture discoverable, no price published.

Three fields, with published validation behind one. aetherSlide serves anatomic pathology slide management and review, aetherAI Hema analyzes bone marrow smears for hematopathology, and aetherAI Endo detects polyps in colonoscopy; research applications cover slide quality control, cell counting and IHC quantification. The bone marrow model has published validation at three Taiwanese centers and one US laboratory, and aetherSlide is named at hospitals in Taiwan, Japan and Germany. No validation is published for Endo or the research applications, and the lymph node metastasis model is research rather than a product.

Coverage named with validation behind part of it.