Pathologist Tomáš Jirásek from Liberec describes how they moved from “tallying on paper” to digital pathology and artificial intelligence. After a year to a year and a half of routine slide scanning, the first algorithms are showing more accurate measurements, faster evaluation, and new opportunities for collaboration with clinicians. What already works, where are the limits, and what comes next?
Why digitize pathology
The impetus was a simple yet key task: reliably counting tumor proliferative activity without markers and manual tally marks. The department therefore implemented a Ventana DP200 (Roche) scanner with the Navify platform – in practice a “microscope under the hood” that converts slides, including large macrosections, into diagnostically usable digital images. The environment is integrated into the hospital network, so the specimens can be accessed from any computer.
The digital image is often combined with the classic microscope, which brings the greatest benefit. Advantages include precise measurement of distances and resection margins, as well as easy sharing at seminars and multidisciplinary boards. In practice, correlating histology with MRI for prostate cancer screening has proven useful, as has collaboration with interventional radiologists and use in neurosurgery, gastroenterology, and during bronchoscopic lung biopsies.
When AI steps in: what it can already assess
Navify sends the scanned images to the cloud, where they are analyzed by artificial intelligence algorithms. The first set focuses mainly on breast carcinoma: assessment of HER2 (immunohistochemically and via in situ hybridization), estrogen and progesterone receptors, and the Ki‑67 proliferation index; for the lung, PD‑L1 (TPS) scoring is available in non‑small cell lung carcinoma. The outputs take the form of heatmaps and quantifications that confirm or refine the expected finding and have a direct therapeutic impact (e.g., indication of anti‑HER2 or checkpoint inhibitors). A practical benefit is that there is no longer any need to manually count thousands of nuclei.
Certification is crucial: for a result to be IVD, the entire “chain” from the immunostainer through the chemistry and the scanner to the algorithm must be approved in IVD mode; some solutions currently operate only as RUO. In preparation are reporting of HER2 low/ultra‑low status and extending the use of Ki‑67 beyond the breast. The team is also working with the manufacturer to refine the reports and integrate them into hospital and laboratory systems. Looking ahead, AI is expected to expand to additional predictive markers, such as ALK, ROS1, MMR, alternative PD‑L1 assessments, or selected chromogenic ISH applications (e.g., EBV, kappa/lambda).
Third parties in Navify: IBEX for breast and prostate
A recent development is opening the platform to external algorithms as well; the first to be added were solutions from the Israeli company IBEX. In breast carcinoma, the AI generates heatmaps of invasive foci, distinguishes ductal and lobular types, marks DCIS and atypical ductal hyperplasia, lymphovascular invasion, TILs, necrosis, adenosis, inflammation, and microcalcifications. The missing pieces of the puzzle are, for now, automatic Nottingham/Elston‑Ellis grading and a consolidated report covering all of a patient’s current samples.
The prostate algorithm works like a “traffic light”: red denotes a high probability of tumor, yellow uncertainty, and green a negative finding, which helps avoid missing small foci. It displays a map of Gleason patterns 3–5, measures tumor length in biopsy cores, and highlights perineural invasion, PIN, and inflammation. Still to be implemented are reporting of cribriform pattern 4, an overall Gleason score, and use for TURP or radical prostatectomies; several modules are also awaiting transition from RUO to IVD. The takeaway from practice is clear: AI does not replace the pathologist, but a pathologist with AI will be faster and more accurate than a pathologist without it; the key is to further fine‑tune the working environment and integration with information systems.
