Abstract
Background
Whole-slide imaging (WSI) systems differ in their tissue detection algorithms, which can alter the scanned area, scan time, and file size. In high-volume labs, these differences translate into tangible workflow and cost implications. Digital pathology workflows require high-resolution digitization of glass slides that can be achieved by using specialized WSI systems. Whole-slide scanners vary in technical features that affect magnification, throughput, image quality, and resulting file formats and sizes. Variations in scan area can profoundly impact operational efficiency. Scan area, determined by scanner-specific tissue detection algorithms, directly influences scan time, which in turn affects workflow and staff planning as well as file size, a major expense in storage and data management. This study compares tissue detection patterns across four commercial whole-slide scanner models to evaluate their effects on these metrics, using classical computer vision to establish perimeter-based benchmarks.Design
260 routine diagnostic glass slides (balanced by hematoxylin and eosin-stained and immunohistochemistry slide types and biopsy/resection tissue types) were scanned on 8 whole-slide scanners representing 4 different commercial manufacturers (designated scanner models A-D). A classical computer vision pipeline was used to delineate the minimal tissue perimeter on each slide, which served as the reference area. Scanner area, scan time, and file size were extracted from the WSI metadata. Absolute and relative area differences were calculated, and linear regression quantified the relationship between area and downstream metrics. One-way ANOVA was used to test the differences between scanner models, stratified by slide and tissue types.Results
All 260 slides were successfully scanned, yielding 1040 WSIs. Scanner models A and B modestly overestimated tissue area with a median of 76 mm2 (0-481 mm2), whereas model D underestimated tissue area by a median of 196 mm2 (61-584 mm2). Scan time increased linearly with scan area (slope 0.08-0.13 s mm-2 for models A-C; 0.75 s mm-2 for model D). File size scaled with scan area (1.05-1.59 MB mm-2 for model A and B; 3.51-3.86 MB mm-2 for model C and D). Stratified ANOVA confirmed significant model differences in all strata (p Conclusion
Tissue detection algorithms vary significantly across scanner models, affecting scan area estimates and downstream performance. Whereas not the sole determinant of throughput, scan area detection is a foundational parameter that impacts time and storage costs. In high-throughput digital pathology environments, understanding these algorithmic differences is critical for informed scanner selection recommendations and workflow optimization.