Author: Diana Cardona, Casey C. Heirman, William Jeck, Kyle J. Lafata, Xiang Li, Lauren M. Neldner, Jeffrey S. Nelson, Megan K. Russ, Ehsan Samei 👨🔬
Affiliation: Duke University, Department of Radiation Oncology, Duke University, Department of Pathology, Duke University, Clinical Imaging Physics Group, Department of Radiology, Duke University Health System 🌍
Purpose: Medical physicists traditionally support radiation-based medicine, but their expertise is translatable to image-based fields like pathology. As pathology transitions to digital practices, physicists’ skills in image analysis can aid in optimization and harmonization of image quality in slide scanners. Variability in slide preparation, imaging, post-processing, and display remains a critical challenge. This study evaluates differences in artifacts and sharpness in slides scanned on two whole slide imaging (WSI) scanners used interchangeably within one department.
Methods: Five H&E-stained mouse buccal tissue slides were scanned using two Leica Aperio WSI systems: GT450 and AT2. The AT2 allows user-defined adjustments in number of in-focus regions, whereas the GT450 is fully automated for high-throughput scanning. A machine learning algorithm trained to detect lymphocytes segmented all lymphocytes in each image. Sharpness was estimated using a method adapted from a rib sharpness metric for chest radiographs. Line ROIs sampled the transition from lymphocyte to extra-nuclear space to create edge profiles, with sharpness quantified as the angle of inclination (AOI) of a fitted sigmoid curve. Images were also visually inspected for stitching artifacts, with lymphocyte sharpness evaluated on either side of artifacts.
Results: At the image level, lymphocytes in AT2 images were sharper on average (AOI = 89.1°) than GT450 images (88.5°). In GT450 images, some neighboring stitched image tiles exhibited noticeable difference in sharpness, with AOI dropping from 88.7° to 86.7° in one representative case. These artifacts displayed misregistration and reduced contrast between tiles. AT2 images did not exhibit significant sharpness differences in neighboring tiles.
Conclusion: This study highlights scanner-related variability in sharpness and artifacts, raising concerns about the reliability of WSI systems for detection tasks. Blurring, contrast loss, and tissue misregistration in GT450 images underscore the need for physicists with expertise in image metrology to harmonize image quality across scanners to ensure diagnostic accuracy.