Leica Aperio iQC DX

The AI is the product. Aperio iQC DX is software that does one job: a deep learning model checks each scanned slide for six common quality problems and flags the ones that need another look. The FDA's decision summary for the clearance (K253561) describes the model as convolutional neural networks, locked so they do not change in use. Without the model, the dashboard would have nothing to show.

The Aperio GT 450 DX and GT 180 DX scanners it works with are separate products and run without it.

Ask whether the software is licensed per scanner or per site, and whether it can be bought apart from a scanner purchase.

Both what the software does and how far to trust it are published. It flags slides and does nothing else: it never rejects or rescans a slide on its own. A technician reviews every flagged slide in the dashboard and accepts or rejects it, with a comment for each rejection. Scan problems are fixed by rescanning. Preparation problems need the glass slide remade first.

Lab administrators set the detection thresholds for blur, striping, air bubbles and pen marks, and changes apply to new scans only. The user guide tells laboratories not to rely on the software alone and to keep their existing quality sampling, because it may miss some artifacts and flag others wrongly. A pathologist stays responsible for the diagnostic review, and serious incidents go to Leica and the regulator.

Ask what default thresholds ship, how often sampling of unflagged slides catches a miss, and how threshold changes are recorded.

The model, its version and its training data are described. Update practice is not. The FDA's decision summary for K253561 says Aperio iQC DX version 1.0 uses deep learning convolutional neural networks, locked so they do not change in use. They were trained on 12,278 slides (5,158 H&E and 7,127 immunohistochemistry) scanned on the Aperio GT 450, tuned on 2,450 more and tested on 4,752 kept apart. The slides came from outside suppliers or an internal Leica site, with no count of sources.

How a new version reaches laboratories, and how they are told, is not published.

Ask how many laboratories and tissue sources supplied the training slides, and how version changes are announced and validated.

The chain is short. The model is Leica's own, locked and cleared under Leica's name (K253561), so it cannot be swapped for another supplier's model without going back to the FDA. It runs on the laboratory's own server, and the user guide says nothing is sent to Leica in normal operation. The parts it depends on are named: Ubuntu, NVIDIA graphics cards, and Leica's own scanners and SAM DX server.

No software bill of materials is published.

Ask for the software bill of materials, and which parties, if any, can reach the server during service.

The main evidence is the test behind the FDA clearance (K253561, 7 August 2026). It used 4,752 whole slide images kept apart from the training data, with experts marking the true artifacts and a second reviewer and an arbiter checking them.

On the GT 450 DX scanner, the software caught between 93.2 and 99.7 percent of slides with each artifact type. It wrongly flagged between 0.1 and 7.3 percent of clean slides. Air bubbles had the lowest catch rate (883 affected slides). Missing or clipped tissue had the most false flags, with 92.7 percent of clean slides passed correctly (2,568 slides). Results on the GT 180 DX were comparable.

Leica's headline claims come from a real world study with the Institute of Pathology at Heidelberg University: up to 24 percent more artifacts caught that technicians had missed, and up to 69 percent less hands on review time. Neither the product page nor the announcement gives that study's method or slide counts.

Ask for the Heidelberg study in full, and for results from laboratories with your scanners and case mix.

Leica states where data goes and how long it stays. Images and results are kept on the laboratory's server for up to seven days, and event logs for six months. The user guide says nothing is sent to Leica in normal operation, and data is encrypted in transit and at rest. The model is locked and does not learn from the laboratory's slides. The FDA's decision summary says it was trained on de identified slides bought from outside suppliers or prepared at an internal Leica site.

What is missing is a written statement that customer data is never used for training, and what leaves the laboratory during service. Safety engineering, meaning guardrails and how a safety event is handled, is not published beyond reporting serious incidents to Leica and the regulator.

Ask for a written commitment on training use, and what service staff can see on the server.

No business associate status is stated, though Leica's product documents take a clear position on where data goes. The user guide says images and results stay on the laboratory's own server for up to seven days, and that nothing is sent to Leica in normal operation. Slide label images carry patient identifiers, so that server holds protected health information.

Leica's privacy policy describes its scope as its websites, applications and other digital services and products. Its full text renders only in a browser. Leica representatives install the software, and the specifications call for an internet upload connection without saying what it carries.

Ask whether Leica signs a business associate agreement for installation, service and any remote access, and what the internet connection sends.

Leica describes the product's security controls, and the FDA reviewed its cybersecurity documentation as part of the clearance. That review checks the software, not how Leica runs security, and no audit of Leica's own operations is published.

The user guide says users sign in with their Aperio SAM DX credentials, access depends on role, and data is encrypted in transit and at rest. Leica runs a coordinated vulnerability disclosure process and publishes product security advisories. No SOC 2 report, ISO 27001 certificate or penetration test is published on Leica's product security page or the product's own pages.

Ask for any security attestation covering Leica's development and service operations, the product's security disclosure form, and its software bill of materials.

The US position is clear and checkable. The FDA cleared Aperio iQC DX Software, version 1.0, on 7 August 2026 under 510(k) K253561, as a Class II device (product code SIX, 21 CFR 864.3700). The predicate is Leica's Aperio GT 450 DX system.

The indication is an AI aid for finding artifacts in whole slide images of H&E and immunohistochemistry stained tissue that need a further image quality check before pathologists review them. It is not for diagnosis, prognosis or prediction of disease. Leica's marketing matches that use, and Leica calls it the industry's first FDA cleared AI quality control software.

Leica's user guide shows the CE mark under the European IVDR and the UKCA mark.

Ask which software version is installed and how updates are cleared, and for the European declaration of conformity.

The FDA's decision summary for K253561 reports Leica's test results by stain and by scanner, and shows where the software is weaker. On the GT 450 DX, it caught 99.6 percent of air bubbles on H&E slides but 90.6 percent on immunohistochemistry slides. Results on the GT 180 DX were reported separately and were comparable. In a repeatability test, rotated copies of the same slides got the correct call every time, on both scanners and both stain families.

Results are not broken out by tissue type or immunohistochemistry marker. No process for governing model changes is described beyond the model being locked.

Ask for results by tissue type and marker, and how a new model version is validated before release.

An error rate with its method and counts is published, and nothing stands behind it. The FDA's decision summary for K253561 (7 August 2026), a clearance Leica announces as its own, reports a test set result on 4,752 slides kept apart from training, checked against expert annotation with a second reviewer and an arbiter. It covers H&E and immunohistochemistry slides on the Aperio GT 450 DX and GT 180 DX.

On the GT 450 DX, the software caught 93.2 to 99.7 percent of slides with each artifact and passed 92.7 to 99.9 percent of clean slides correctly. For air bubbles, it caught 93.2 percent of 883 affected slides in a set of 2,134. These are test results, not performance in use.

Leica's user guide warns that the software may miss artifacts or flag clean slides, and the warranty sits in its standard terms and conditions. No guarantee, indemnity or remedy tied to a missed artifact is published.

Ask what Leica commits to when a missed artifact reaches a pathologist, and whether the warranty covers detection performance.

Integration is limited to Leica's own systems. The software reads scanned images from a network share fed by Aperio GT 450 DX or GT 180 DX scanners. It signs users in through Leica's Aperio SAM DX server, and shows results on the scanner console and in its own dashboard. The specifications mention DICOM access without naming a system.

No connection to a laboratory information system or a third party image management system is published.

Ask whether flags can reach your laboratory information system or image management system, and by what standard.

The software runs inside the laboratory. It installs on a physical or virtual server the laboratory provides, running Ubuntu 24.04, next to its Aperio scanners and SAM DX server. Images and results stay on that server for up to seven days, and the user guide says nothing is sent to Leica in normal operation.

One path is left open. The specifications call for a 100 Mbps internet upload link, while the user guide says the system works on the local network. What the connection carries is not stated.

Ask what the internet connection is used for, and whether the software runs with it switched off.

No price is published, but how the product is sold can be worked out. Aperio iQC DX is software for laboratories that scan on Leica's Aperio GT 450 DX or GT 180 DX. It runs on a server the laboratory provides, physical or virtual, with one to four NVIDIA L4 graphics cards. Leica representatives install it and train staff.

The product page routes to a sales contact. No license term or unit of charge is stated.

Ask whether it is priced per scanner, per slide or per site, what server hardware is required, and whether it can be added to an existing scanner fleet.

The setting is narrow and stated precisely. The software checks image quality in clinical histology laboratories, before a pathologist reviews the slides. It covers H&E and immunohistochemistry stained, formalin fixed, paraffin embedded tissue scanned on the Aperio GT 450 DX or GT 180 DX. It looks for six problems: out of focus areas, image striping, missing tissue, clipped tissue, air bubbles and pen marks.

The test behind the FDA clearance (K253561) validated each problem type on both scanners and both stain families, with results reported separately. Leica's user guide lists known limits for each type, such as faint staining, and warns that results may be worse on unusual tissue, stains or other scanners. Tissue folds are not among the problems it is cleared to find.

Ask for results on your own tissue types and immunohistochemistry markers, and how to handle problems outside the six.