This study provides a systematic, quantitative evaluation of focus quality in thyroid LBC slides scanned using three different whole-slide scanners under standardized single-plane conditions. Using the variance of the Laplacian as an objective sharpness metric, we demonstrated measurable and statistically significant differences in focus performance among whole-slide imaging systems. One scanner consistently achieved higher median focus values and a greater proportion of in-focus tiles, particularly in SurePath preparations. These findings confirm that measurable differences in focus quality exist between scanners and underscore the importance of objective image quality assessment in digital cytopathology.

Prior studies have reported that SurePath preparations, due to their three-dimensional cell clusters and greater cellular thickness, present more frequent focusing challenges during both microscopy and whole-slide scanning, whereas ThinPrep slides generally produce monolayered, evenly distributed cells that are easier to capture in a single focal plane [11, 12]. Our results corroborate these observations: SurePath slides exhibited greater spatial heterogeneity in focus quality and more pronounced performance gaps among scanners. Two scanners maintained > 98% in-focus rates in most SurePath slides, whereas one scanner showed substantial drops in half of the cases (down to 73.7%), confirming that scanner hardware and focusing algorithms can critically influence digital image quality in three-dimensional preparations.

In ThinPrep slides, differences between scanners were less pronounced. One scanner yielded the highest focus scores in three of six cases (TP2, TP5, TP6), but another performed comparably or better in TP1, TP3, and TP4. The consistently high in-focus rates (> 95%) across all scanners for most ThinPrep slides suggest that ThinPrep preparations are less sensitive to autofocus limitations, likely due to their planar morphology.

The superior performance of one scanner observed in this study, particularly for SurePath preparations, is likely attributable to the multi-focal strategy used in its whole-slide imaging system. This approach employs simultaneous acquisition of information from three distinct focal planes in a single scan pass and algorithmically merges the sharpest pixels from each layer. Functionally, it approximates the benefits of Z-stack or extended-depth-of-field imaging while maintaining the efficiency and data footprint of standard single-plane acquisition. For cytology, where three-dimensional cell clusters are common and large slide volumes must be digitized efficiently, this design may represent a practical compromise between focus quality, scanning speed, and storage requirements.

These findings have important implications for laboratories transitioning to digital cytopathology. Scanner selection should consider the predominant preparation method in use [6]. Institutions processing primarily SurePath samples may particularly benefit from scanners with multi-plane acquisition or focus-enhancement capabilities to ensure effective capture of diagnostically relevant details without compromising storage efficiency.

Laplacian variance has been described as a measure to assess for focus quality of WSIs in histopathology [19], but to our knowledge, no studies have explored the use of Laplacian variance in LBC. Dealing with blurred WSIs is a particularly common problem for cytology and is one of the main reasons why many laboratories would hesitate to adopt digital cytopathology for their routine practice [14]. By classifying WSIs based on their focus quality, using Laplacian variance, laboratories will be able to identify out-of-focus slides using a quantitative measure for quality assurance. Furthermore, this approach may enable the implementation of an AI-based autofocusing system as a quality control step prior to allocation or applying cytology AI software in the future, resulting in improved efficiency [20]. Our study demonstrated that objective metrics such as Laplacian variance can serve as reproducible tools for image quality benchmarking and quality assurance, supplementing traditional visual inspection in LBC [21]. Furthermore, this study could enable real-world data collection from multiple laboratories for independent, comparable evaluation and feedback to manufacturers, to improve their devices.

While continuous Laplacian variance values provide subtle gradations in image sharpness, the binary classification into in-focus vs. out-of-focus tiles is also clinically relevant. Although one scanner demonstrated lower performance in several SurePath cases, all scanners produced more than 70% in-focus tiles in every slide, indicating that overall diagnostic usability was preserved. Performance in focus quality was especially consistent for ThinPrep preparations, with only one slide showing a noticeable decrease in the proportion of in-focus titles (82.9%). However, lower in-focus proportions may have implications for downstream tasks such as automated cell detection or quantitative image analysis, which are sensitive to defocus artifacts.

Strengths of this study include the use of harmonized ROI selection, chroma-based tile filtering, and standardized single-plane scanning conditions to ensure fair comparison across scanners. Furthermore, the focus-assessment workflow was implemented using transparent, script-based procedures in QuPath (for reproducible tile extraction and chroma validation) and Python (for Laplacian-based focus quantification and tile-level data aggregation). By relying on openly available software tools and explicitly defined processing steps, this study provides a reproducible and extensible computational approach that can be readily adopted, verified, and modified by other laboratories seeking to evaluate or optimize focus performance in digital cytology. Because the computation of Laplacian variance itself is computationally lightweight and scales efficiently with tile-based processing, this approach could, in principle, be incorporated into routine quality-control workflows for cytology WSIs. However, practical implementation would require vendor-level integration, harmonization of threshold settings, and validation across diverse cytology specimen types, particularly because total processing time can vary substantially depending on hardware performance and tiling configurations.

Limitations include the relatively small sample size and focus on thyroid cytology slides only. Different specimen types (e.g., effusion cytology, cervical cytology, aspirates from various organs) and different staining methods may present different focusing characteristics. An additional methodological limitation is that all scanners were operated strictly under the manufacturer’s default automatic focusing modes without any manual focus refinement. Autofocus algorithms in certain devices are primarily optimized for histologic sections and may intentionally reduce the number of focus points to accelerate scanning, potentially disadvantaging cytology slides with greater three-dimensional complexity. Therefore, the observed differences in focus quality partly reflect each scanner’s default autofocus behavior under routine single-plane conditions.

Although our analysis demonstrated measurable differences in focus quality across scanner systems under controlled single-plane conditions, these findings should not be interpreted as a definitive ranking of scanner superiority. Performance may vary depending on specimen type, cellularity, staining characteristics, scanning protocols, and software settings. In addition, scanning speed is an important but multifactorial parameter that warrants a dedicated, methodologically optimized study, ideally incorporating standardized queue loads, repeated measurements, and controlled slide characteristics. Therefore, comprehensive validation across a broader range of cytology preparations, stains, and clinical practices will be essential to determine the generalizability of these observations. Additionally, we did not compare single-plane scanning with Z-stack or extended depth-of-field modes directly. Future studies should investigate whether manually optimized focus settings, tri-focal layer merge or other multi-plane approaches improve diagnostic accuracy and computational pathology performance across use cases.