Digital Pathology and AI: Beyond the Microscope in Cancer Diagnosis

The pathologist is the physician who examines tissue under a microscope and determines whether cancer is present, what type it is, and what grade. That decision initiates the entire chain of oncologic treatment. Yet classical pathology is inherently demanding: a single prostate biopsy can produce dozens of glass slides, each containing thousands of cells in a cubic millimeter, and a tiny missed focus can change the diagnosis entirely. It is precisely at this point that digital pathology and artificial intelligence (AI) meet.

Digitalization First: Whole Slide Imaging (WSI)

For AI to enter pathology, the glass slide must first be converted into a digital image. Whole slide imaging (WSI) scans a slide at high resolution to produce a gigapixel-scale digital file. That file can then be examined remotely (telepathology) or fed as input to an algorithm. In the United States, the FDA cleared the first WSI system for primary diagnosis in 2017, opening the way for this transformation. Digitalization on its own already creates value — it speeds up archiving, second opinions, and consultations. But the real leap comes with the AI layer added on top of these digital slides.

A Turning Point: Paige Prostate

The milestone of digital pathology AI is Paige Prostate. In 2021, Paige Prostate received FDA de novo marketing authorization, becoming the first FDA-authorized artificial intelligence product in digital pathology. The system flags areas suspicious for cancer on prostate biopsy slides, acting as an assisting second reader for the pathologist.

The clinical data are striking. With Paige Prostate assistance, pathologists' sensitivity for correctly identifying cancer rose from approximately 88.7% to 96.6% — corresponding to roughly a 70% reduction in false negatives (missed cancers) and about a 24% reduction in false positives. Cancer detection in slide images increased by an average of 7.3% compared with unaided review. The key point here is that the system does not replace the pathologist; it raises performance by highlighting small foci that might be missed. In 2025, the Paige Prostate product family also received IVDR certification in Europe.

Going Platform: PathAI and AISight Dx

The second major trend in the field is the move from single-task algorithms to end-to-end digital pathology platforms. PathAI's AISight Dx platform is a leading example: on 30 June 2025, it received FDA 510(k) clearance for primary diagnosis use in clinical settings. The system is cleared for use with the Hamamatsu NanoZoomer S360MD and Leica Aperio GT 450 DX scanners.

AISight Dx is a cloud-based image management system that combines slide management, simultaneous multi-slide navigation, live collaboration, and AI integration in a single platform. A noteworthy regulatory detail is that the clearance includes a Predetermined Change Control Plan (PCCP), which allows the manufacturer to deploy certain updates — such as new scanners, displays, or file formats — without filing a new 510(k) each time. In other words, it grants regulatory flexibility to a "living" piece of software. PathAI is also developing algorithms for automated digital scoring of biomarkers such as HER2 (AIM-HER2), a field with high inter-observer variability that directly affects treatment selection in breast cancer.

The Evidence Base: Pathologist + AI

Beyond individual products, independent studies also support the value of AI-assisted pathology. For example, in gastric cancer diagnosis, deep-learning-assisted pathologists' area under the ROC curve (ROC-AUC) increased from 0.863 to 0.911 compared with unaided reading, a statistically significant difference. In some narrowly defined tasks — such as detecting signet-ring-cell carcinoma — models have reported AUC values approaching 0.99 on test sets. These results show that, in well-framed and well-validated tasks, AI provides measurable additive value to pathologic diagnostic performance.

The biomarker-prediction dimension of digital pathology is also advancing rapidly: a growing literature shows that weakly supervised deep learning can predict molecular features (e.g., microsatellite instability) directly from routine H&E-stained slides. In the future, this could provide a pre-screening layer for some expensive and time-consuming molecular tests.

Obstacles and Honest Limits

The obstacles in front of digital pathology are both technical and operational. Scanner and laboratory variability: variations in staining protocol, scanner brand, or section thickness can reduce model performance in another center, making multi-center validation critical. Infrastructure cost: WSI scanners, massive storage requirements, and workflow integration represent a serious investment, which has slowed the spread of digital pathology compared with radiology. Interpretability: the ability of an algorithm to show the pathologist why it called something "cancer" (e.g., heatmaps) matters for both trust and legal accountability.

For these reasons, almost all approved systems today are positioned within a decision-support framework, with the pathologist making the final call. Autonomous pathologic diagnosis is not — regulatorily or clinically — the reality of today.

Conclusion

Digital pathology and AI have reached real maturity in cancer diagnosis: Paige Prostate, the first FDA-cleared pathology AI that meaningfully reduces false negatives; AISight Dx, a platform extending all the way to primary diagnosis; and automation in biomarker scoring. The value proposition is clear — a partner that does not tire the pathologist, highlights what could be missed, and provides objective measurements. The microscope is still in the pathologist's hand; but now there is an unblinking digital eye beside it.