Summary
A study on the Hologic Genius digital diagnostic system (GDDS) for AI-assisted cervical cytology screening identified technical issues during validation. While most errors were manageable and did not compromise patient safety, the study emphasized the importance of standardized slide preparation and workflow planning for successful implementation. Additionally, a review by the College of American Pathologists found that CAP accreditation frameworks are broadly applicable to machine learning-based methods in molecular oncology testing, providing a robust framework for validation and oversight.
Editors: Liron Pantanowitz, MD, PhD, MHA, chair of the Department of Pathology and professor of pathology, University of Pittsburgh Medical Center, and Matthew G. Hanna, MD, vice chair of pathology informatics and associate professor, Department of Pathology, University of Pittsburgh Medical Center.
Technical considerations during validation of Genius digital diagnostic system
April 2026—As pathology laboratories transition to using artificial intelligence-assisted systems to help screen Pap tests, they may encounter not only clinical or workflow challenges while clinically validating those systems but also technical issues. The authors provided cytology labs considering AI-assisted Pap screening with evidence-based, real-world insight into what can go wrong, how often issues may arise, and how to address those problems. They reported on some of the technical issues they encountered while validating the Hologic Genius digital diagnostic system (GDDS) for AI-assisted cervical cytology screening. The authors reviewed 909 archived ThinPrep Pap test slides with biopsy follow-up and systematically documented slide- and instrument-level errors during digital scanning. They found that 21 (2.3 percent) slides showed slide events that prevented initial imaging, most commonly due to quality control errors, cell focus errors, barcode problems, oversaturation, or duplicate scans. Thirteen (1.4 percent) slides could not be scanned and were excluded, largely due to scratched coverslips from long-term storage. Eight problematic slides were successfully rescanned. Importantly, rescanning did not compromise diagnostic accuracy. In addition, 43 imager errors—for example, slide-handling or motor failures—temporarily interrupted scanning but were resolved through user intervention, system reboot, or vendor support. The authors emphasized that minor technical errors should be expected when digitizing large volumes of cytology slides, especially archival material. Therefore, this study sets realistic expectations for system implementation. The results of the study provide quantitative benchmarks for other laboratories relative to error and rescan rates. Laboratories can reference this work to justify policies on rescanning, manual review, and exception handling. Manual glass slide review can remain a fallback. Further, the study showed that use of AI is likely to shift, but not eliminate, the need for human involvement. Cytologists will need to manage rescans and troubleshoot errors, understand system flags and warnings, and plan for downtime and reboot scenarios. It should be noted that not all errors identified in this study were algorithmic—many were mechanical, optical, or preanalytic. Most issues were manageable and did not pose patient safety risks because unscanned slides remained available for manual microscopic review. The authors concluded that with proper, standardized slide preparation, labeling, quality control, and workflow planning, the GDDS can be implemented safely, though continued evaluation of quality and patient safety is needed as AI-assisted cytology is adopted as part of routine practice. They also emphasized that preanalytic quality is critical and reinforced the key digital pathology principle of garbage in equals garbage out.
Harinath L, Harrington S, Matsko J, et al. Technical considerations during validation of the Genius® digital diagnostic system. J Path Inform. 2026;20. dx.doi.org/10.1016/j.jpi.2025.100532
Correspondence: Dr. Lakshmi Harinath at [email protected] or Dr. Liron Pantanowitz at [email protected]