Realtime Extended Focus vs. Conventional Z-Stacking: Comparative Scanner Validation in Cytopathology

Cytopathology demands multiplane whole slide imaging for three-dimensional cell clusters, yet conventional z-stacking imposes disproportionate scan time and storage costs. Realtime Extended Focus (Realtime EF) merges focal planes in a single pass, potentially resolving this tradeoff. We benchmarked the Vieworks LH210 (Realtime EF) against 3DHistech FLASH using pathologist evaluation and dual automated quality control (QC) algorithms.Twelve cytopathology cases (eight organ sites; conventional smear and liquid-based cytology) were scanned at 40× under four conditions: 1-layer and 3-layer acquisition on each platform. Three blinded raters scored 144 randomised images across six quality domains on a 3-point Likert scale (Friedman test; Fleiss' κ). Automated QC used two parallel pipelines: GrandQC (EfficientNet-B0 UNet++ segmentation; 7-class artifact mapping) and a Laplacian variance focus metric (512×512 tile analysis). Scan time, file size, and quality-adjusted efficiency ratios were recorded.No platform differences emerged across all six domains for any specimen type (all post-hoc p > 0.05; Fleiss' κ 0.61-0.78). GrandQC detected lower artifact burden in LH210 images (18.1% vs. 29.7%; p = 0.005), especially out-of-focus regions (0.42% vs. 2.16%). Critically, GrandQC misclassified 3D cytology clusters as pen markings-domain shift from histopathology training-highlighting the need for cytology-specific model development. Laplacian and GrandQC were weakly correlated (r = −0.24), confirming complementary quality dimensions. LH210 achieved equivalent image quality at one-third the scan time (3.0 vs. 14.4 min) and file size (0.21 vs. 0.62 GB).LH210 Realtime EF delivers validated cytomorphological quality with threefold efficiency gains. Dual automated QC confirms objective artifact advantages while revealing a critical gap: histopathology-trained AI models require cytology-specific adaptation before routine deployment.

Learning Objectives:

1. Compare image quality and scanning efficiency of Realtime EF versus z-stacking across diverse cytopathology specimen types.
2. Assess GrandQC deep learning and Laplacian variance as dual automated QC tools for WSI benchmarking in digital cytopathology.
3. Reveal readiness gaps in histopathology-trained AI models for cytology and define a cytology-specific QC development agenda.