Hologic Genius Cervical AI
Genius Cervical AI is Hologic's AI for cervical cancer screening. Cytology laboratories buy it as part of the Genius Digital Diagnostics System, which scans Pap test slides so staff can review them on screen instead of under a microscope.
The AI looks through the tens of thousands of cells on each slide and pulls the ones most likely to matter into a gallery. A cytologist reviews the gallery and makes the call. In the study behind its FDA authorization, reviewers using it caught 82 percent of high grade lesions, against 74 percent by microscope, while flagging more normal slides as abnormal.
The FDA authorized the system in January 2024 as a new type of device, and Hologic says it is also CE marked in Europe. It reads only Hologic's own ThinPrep Pap test slides. Images stay on a server in the laboratory.
Hologic, based in Marlborough, Massachusetts, was taken private in April 2026 by funds managed by Blackstone and TPG. Its breast imaging AI has a separate record. No price is published.
Capability Axes
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
The AI is the reason to buy the system, but it is not the whole system. Genius Cervical AI is a deep learning model that Hologic says analyzes every cell on a scanned Pap test slide. It builds a gallery of the cells most likely to be abnormal and estimates the squamous cell count used to judge whether a sample is adequate.
It runs inside the Genius Digital Diagnostics System, which also includes the slide scanner, an image server and review stations. Those parts would still scan, store and display slides without the AI, and in Europe the same scanner now images tissue slides too. The FDA authorized the system and the algorithm together, under De Novo DEN210035.
The oversight rules come with the FDA authorization (De Novo DEN210035). The AI shows a gallery of 30 images from each slide, with more available on request. A qualified cytologist or pathologist must review the entire gallery before giving an interpretation. Where the diagnosis is uncertain, the reviewer examines the glass slide under a microscope. No slide is signed out by the AI alone. The instructions for use also set a workload limit of 200 case reviews in a workday of at least eight hours when no case needs full manual review, and no more than 100 CLIA slide equivalents a day.
What is not published is how a laboratory catches an abnormality the gallery did not show. In the authorization study, reviewers using the system still missed about 18 percent of high grade readings.
The approach is described. The model's version and training data are not. The FDA's decision summary for De Novo DEN210035 identifies the algorithm as a convolutional neural network. It reads whole slide images holding 5,000 to 100,000 cells and ranks objects of interest into the gallery. The system also estimates the squamous cell count.
Hologic calls the model deep learning, trained with a robust and standardized data set, without the size, source or makeup of that data. The instructions for use, revised in August 2026, point to operator manuals for system software versions 1.1.x and 1.2.x. No version identifier for the model, release notes or notice of model updates is published.
The chain is short. The model is Hologic's own, authorized under Hologic's name as part of the device (De Novo DEN210035), so it cannot be swapped for another supplier's model without returning to the FDA. It runs on the laboratory's own server. The server manual describes no cloud processing, and no outside model provider is named.
Third party software on the server, such as the antivirus tools Hologic has tested, is named in the manual. No software bill of materials or list of parties involved in remote support is published.
The main evidence is the multi site study behind the FDA authorization (De Novo DEN210035, summarized by the FDA). Four US laboratories reviewed 1,994 archived ThinPrep slides, deliberately enriched with abnormal cases, three ways. They read them by microscope, with Hologic's earlier ThinPrep Imaging System, and with Genius. An expert panel set the reference result.
Reviewers using Genius caught 81.5 percent of high grade lesions or worse (699 of 858 readings), against 74.0 percent by microscope. They also flagged more normal slides, as specificity at that level fell from 97.2 to 94.8 percent. For cancer, detection was the same either way, at 66.7 percent. At the lowest threshold, atypical cells or worse, the gain of 1.6 points was not statistically significant.
Hologic also cites published studies from laboratories using the system, including Cleveland Clinic, the University of Michigan, ProPath and CytoMol in Germany, for review times 46 to 60 percent shorter. Its pages give no detection figures from routine screening.
Hologic describes where the data sits, not what happens to it beyond the laboratory's own use. Slide images and the AI's galleries are held on the laboratory's Image Management Server as a short term cache. The laboratory manages archiving, retention and deletion under its own record rules, and deleted galleries cannot be recovered. Hologic's privacy notice says patient data is not among the device data it collects.
Nothing published says whether laboratory images are ever used to train or monitor the model, or how the training data was de identified. Hologic says only that the model was trained on a robust and standardized data set. Safety engineering, meaning guardrails and how a safety event is handled, is not published.
No business associate status is stated. Hologic's privacy notice, last revised in March 2022, covers its medical technology services, including technical support. It lists device log data from diagnostic systems among what it collects, but not patient data.
The Genius system keeps slide images and case data on the laboratory's own server, and the laboratory is responsible for retention and archiving. Hologic installs the system and describes remote support among its features. Nothing published says whether support staff can see patient images, or whether an agreement covers that access.
A recognized certification is named by the manufacturer, but not for this product. Hologic states in the security disclosure for its breast imaging AI, dated June 2025, that it holds ISO 27001 certification. The certificate's scope and date are not published, and the Genius product pages carry no security disclosure or certification.
The server manual sets security requirements for the laboratory, including a firewall, tested antivirus software, TLS 1.2 or 1.3, and hardware disk encryption rather than software encryption. It also publishes a cybersecurity assessment of the server on Windows Server 2016 with seven findings, four of them severe and all four on SMB settings, alongside Hologic's recommended mitigations. The instructions for use say critical security updates are applied immediately after validation and other patches at scheduled maintenance. Hologic publishes a coordinated vulnerability disclosure policy.
The US position is clear and checkable. The FDA granted De Novo authorization DEN210035 on 31 January 2024 to the Genius Digital Diagnostics System with the Genius Cervical AI algorithm. That decision created a new Class II device type (product code QYV, 21 CFR 864.3900).
The indication covers creating and viewing digital images of ThinPrep Pap test slides. The AI's objects of interest are shown in a gallery for cytologists and pathologists to review, as an aid to cervical cancer screening under the Bethesda System. Hologic's marketing stays inside that use, though its pages call the system FDA cleared where the decision was a De Novo grant.
In Europe, Hologic states the system is CE marked. Its November 2025 extension to tissue imaging was marked under the IVDR and is not available in the United States.
The FDA's decision summary for De Novo DEN210035 reports results laboratory by laboratory and tests of scanner consistency, from Hologic's own studies. Nothing is reported by patient group.
Scanning the same slides on three imagers gave the same kind of flagged cells 96 percent of the time between instruments and 99 percent on one instrument (999 pairs each). Across the four study laboratories, results varied. For high grade lesions, reviewers using Genius did better than by microscope at three sites and worse at one, 77.5 against 87.4 percent (231 readings). Hologic says the model's performance generalizes across healthcare settings.
No results by age, HPV status or other patient groups are published, and no process for governing model changes is described.
Review steps exist, and nothing stands behind the AI's output. A trained reviewer must look at the whole gallery, can examine the glass slide, and makes every interpretation, so responsibility stays with the laboratory. The accuracy figures in the FDA's decision summary for De Novo DEN210035 are what reviewers achieved using the system, not the algorithm's own miss rate.
The current server manual mentions a warranty that use outside its instructions may void, without stating its terms. No indemnity, guarantee or correction route for a missed lesion is published.
Integration is claimed, not named. Hologic's November 2025 announcement lists laboratory information system readiness among the system's features, without naming any system or standard.
Input is closed. The AI reads only ThinPrep slides scanned on Hologic's own Digital Imager. Images and galleries move between the imager, the server and review stations, which can sit at other sites on the laboratory's network. Any archive is set up by the laboratory. How the interpretation reaches the laboratory information system or a patient record is not described.
The system runs in the laboratory, and the data stays there. The Image Management Server sits on the laboratory's network as a short term store for scanned slides, offered as hardware with software or as software only. Review Stations connect to it locally or remotely, and any longer term archive is installed and run by the laboratory. Hologic's server manual describes no cloud processing or Hologic hosting.
Remote access is left open. Laboratories that connect remote review stations are responsible for their own network security, and how Hologic's remote support reaches the server is not described.
No price is published, but how the product is sold can be worked out. Hologic sells Genius to laboratories as a system: the Digital Imager, the Image Management Server, Review Stations and the Cervical AI algorithm. The server comes either as hardware with software or as software only, and Hologic installs and configures the system itself.
The system reads only ThinPrep Pap test slides, so a laboratory buying it is also committing to Hologic's slide preparation and stain. The product pages route to a sales contact, and no license term or unit of charge is stated.
The intended setting is narrow and stated precisely. Genius is for cervical cancer screening in clinical laboratories. It reads ThinPrep Pap test slides prepared on the ThinPrep 2000, 5000 or Genesis processors and stained with ThinPrep stain. Trained cytologists and pathologists report the results in the Bethesda System categories.
The authorization study (De Novo DEN210035) validated that use at four US laboratories, with results for each category from atypical cells to cancer. Hologic cites laboratories in the United States and Germany using it. The thinnest categories in the study were glandular abnormalities and cancer, with about 20 slides each, and results varied by laboratory.
Other cytology specimens are outside the product. The tissue slide imaging Hologic added in Europe in November 2025 is not available in the United States.
Alternatives to Hologic Genius Cervical AI
Other products that do the same kind of work as Hologic Genius Cervical AI, and what each one does.
Techcyte sells Fusion, a digital pathology platform that human, veterinary and environmental laboratories use for both anatomic and clinical pathology.
Aperio iQC DX is Leica Biosystems' AI quality check for digital pathology slides. Clinical laboratories that scan slides on Leica's Aperio GT scanners use it to catch bad scans before a pathologist sees them.
Pramana makes whole slide imaging scanners that use real time AI during acquisition: quality control that detects artifacts such as bubbles, folds, debris and faint tissue and rescans automatically, tissue detection and region selection, and edge algorithms for feature detection while the…
Tribun Health sells CaloPix, software that pathology laboratories use to manage cases and review scanned slides on screen.
PathAI sells AISight, a cloud image management system that anatomic pathology laboratories use to manage cases, read slides on screen and run AI, along with its own algorithms and a menu of partner algorithms.
Aiosyn sells AI that analyzes images of stained tissue, to pathology laboratories and to drug companies running animal studies.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
Not published
|
Quote based. Sold to laboratories as a system: Genius Digital Imager, Image Management Server (hardware with software, or software only), Review Stations and the Genius Cervical AI algorithm. No unit of charge published. | Not published. Slide images stay on the laboratory's own server; whether an agreement covers Hologic's installation and remote support is not stated. | Not published. Hologic installs and configures the system. | Vendor Published |
No price is published for Genius Cervical AI or the Genius Digital Diagnostics System, and the product pages route to a sales contact.
Hologic sells it to laboratories as a system of scanner, image server, review stations and AI, and installs it itself. The server comes as hardware with software or as software only.
The system reads only ThinPrep Pap test slides, so a laboratory is also committing to Hologic's slide preparation and stain. No unit of charge is stated, whether per slide, per instrument or per site.