Deployment of AI-driven automated quality control of whole-slide images in a large tertiary cancer center
Kaitlyn Gelfant, Ali Manzo, Samiha Alam, Md Mostafizur Rahman, K. Hasan Bilal, Rushi Brahmabhatt, Suraj Nayak, Nitin Singhal, Dinesh Joshi, Evangelos Stamelos, Ishtiaque Ahmed, Jonathan Alarcon, Mohit Pasricha, Maria Pirgousis, Peter Ntiamoah, Ahmet Doǧan, Victor E. Reuter, Meera Hameed, Orly Ardon · Journal of Pathology Informatics · 2026
Introduction: Quality control (QC) remains a major challenge in digital histopathology, as artifacts introduced during slide preparation and whole-slide imaging (WSI) can compromise diagnostic accuracy. Automated QC has emerged as a potential solution to the scalability and variability limitations of manual review in digital pathology workflows. However, there is limited evidence describing real-world, enterprise-scale implementation of automated QC systems within high-throughput clinical environments. Materials and methods: We evaluated the clinical implementation of a commercially available, AI-based automated QC platform (AIRAQc) within a large digital pathology lab. Feasibility testing included 60 histopathology slides scanned across 3 WSI platforms. Following enterprise integration, a retrospective operational analysis was performed on 94,995 WSIs generated over 1 month across 10 subspecialty services and multiple stain types. System performance, artifact prevalence, processing latency, reproducibility, and scalability were assessed using structured data exports, statistical analyses, and controlled load-testing scenarios. Results: Artifact detection demonstrated high reproducibility across scanner platforms, with tissue fold detection showing >97% concordance and air bubble detection exceeding 98% concordance. Scanning-related artifacts, including missing tissue and blurred regions, varied by scanner model but were consistently identified. Across the clinical deployment, the mean analysis time was 17 s/image, with no analysis failures observed during load testing. Mean total artifact burden/image was 1.91%, with most images meeting predefined QC thresholds. The system maintained stable performance under sustained high-throughput conditions, supporting daily volumes exceeding 6000 slides without workflow disruption. Conclusions: This study demonstrates the feasible deployment of an AI-based QC framework within a large-scale, multi-vendor clinical digital pathology environment. The QC framework enabled consistent assessment of routine WSIs with low per-image latency and sustained high-throughput scanning without workflow disruption. Consistent application of QC thresholds across a multi-instrument infrastructure reduced reliance on manual review and supports the integration of automated QC as a core component of contemporary digital pathology workflows.