Researchers from the Department of Pathology, Brigham and Women’s Hospital and Harvard Medical School, evaluated how different digital pathology operations influenced digital storage costs, scanner dropouts, and throughput of whole slide imaging (WSI).1 “Our department is initiating a diagnostic digital pathology program, and we needed to compare and validate candidate scanners,” said George Mutter, MD, Professor of Pathology at Harvard Medical School and corresponding author of the study.
They found that digital image fidelity, scanning speed, and storage costs per digital slide varied considerably between scanners, tissue type, tissue area, and scanner settings. “These parameters can greatly affect programmatic digital costs and timeliness of digital slide availability,” said Dr Mutter. “Scanner performance must be defined using the materials and conditions of intended use. Vendor claims may be based on tissue sets and stains that generate very different scan times and file sizes than your laboratory,” he added.
The study was published in the Journal of Pathology Informatics on January 8, 2022.
Study Rationale: Comparing WSI Throughput and Tissue Dropouts
Despite the many advantages of digital technologies, their adoption in routine pathology practice has been relatively slow. Long scan times per slide (i.e., operational throughput) and high digital storage costs (i.e., file size) are important factors hindering the wide clinical implementation of WSI scanners. Sometimes, tissue areas on the slide may be missed during scanning and, hence, omitted in the digital image—this is known as scanner dropout and is another limitation of WSI that affects downstream interpretation.1
However, measuring scanner dropouts, digital storage costs, and operational throughput is often challenging. Accurately measuring these parameters is key to optimizing digital pathology workflows and expanding the clinical adoption of digital technologies in pathology.
To compare scan time, file size, and image completeness among different WSI scanners, the researchers used a randomly selected set of 212 surgical pathology specimens from the Women’s and Perinatal Pathology service at Brigham and Women’s Hospital. Stained glass slides were scanned using a Leica GT450 (standard settings) and a Hamamatsu S210 scanner (default and optimized settings). “The use of a single, well-defined clinical slide set across multiple scanner platforms allowed us to directly compare performance,” Dr Mutter noted. To identify WSI dropouts, the team overlaid whole slide images with photographs of macroscopic glass slides, which were used as reference.
Variations in File Size and Scanning Time
For all scanners, the file size correlated with the scanning time, with longer scanning times resulting in larger digital images. The correlation factor r was 0.828 for GT450, 0.887 for S210 with default settings, and 0.930 for S210 with optimized settings.
Scanning time varied greatly among the scanners. The average scanning time was 93 sec for GT450, 376 sec for S210 with default settings, and 721 sec for S210 with optimized settings. Similarly, the size of digital images differed among the platforms, with average file sizes of 1.4 GB for GT450, 2.5 GB for S210 with default settings, and 3.4 GB for S210 with optimized settings
Differences in Scanning Dropouts
Overall, scanning dropouts were observed in 29.5% (186/631) of successful scans. Dropouts of small dislodged fragments were observed in 22.2% (140/631) of slides, and missed edge domains were seen in 6.2% (39/631) of slides.
“Our image analysis enabled us to detect tissue dropouts in the digital file with high sensitivity. Using this method, we found that tissue dropouts may be more common than can be readily recognized by subjective viewing of digital images by pathologists,” Dr Mutter explained.
Despite the high overall rate of scanning dropouts, their frequency differed significantly among the platforms (p < 0.001; chi‑square test). S210 with optimized settings provided the lowest frequency of scanning dropouts (13.7%; 29/212), followed by the GT450 (34.6%; 73/211) and S210 with default settings (40.4%; 84/208).
Most dropouts involved small parts of the tissue. In 78.5% (146/186) of dropouts, only 2% or less of the tissue on the slide was lost in the digital image. Nonetheless, the scale of tissue dropout varied depending on the scanner used. A larger percentage of tissue was lost in digital images obtained using the S210 with default settings.
Histologically unique tissue dropouts represented only 2.15% (4/186) of all tissue dropouts and occurred in only three slides. Most of the histologically unique tissue dropouts were contaminants from other specimens.
“We have more precisely defined the clinical significance of tissue dropouts created during slide scanning. High sensitivity image analysis detects many tissue dropouts that are clinically insignificant: artifactually contaminating tissue shards, tissue present at extreme slide margins that are difficult to view even under an optical microscope, or non-unique tissue elements represented elsewhere in the slide,” Dr Mutter said. “Dropouts can be remediated prospectively by optimization of scanner settings and changes in histology laboratory practices,” he added.
Future Perspectives
Dr Mutter noted that one important limitation of the study is that it included two scanner types, which represent only a small portion of the commercially available WSI scanners. Future studies are needed to expand the range of scanners characterized for their scanning parameters, including scanning time and tissue dropouts.
“Hardware (scanners) and software (compression, image processing) for digital pathology are changing, leading us to believe that scanner performance and file characteristics will improve over time,” Dr Mutter noted.
Moreover, the team did not examine tissue focusing, which is a critical variable for image interpretation by a pathologist. “Lack of focusing (z-axis) capability in single-layer scanners is a shortfall from optical viewing of glass originals under the microscope. Solutions to this problem are emerging using various combinations of tissue preparation, hardware and software innovations,” Dr Mutter said.
“All digital pathology today requires advance histologic preparation, where scanning and digital costs are additive to current budgets. Direct digital imaging of whole tissues with false-color stains is an exciting possibility that could transform the workflow and costs of digital pathology,” Dr Mutter concluded.
References
- Mutter GL, Milstone DS, Hwang DH, Siegmund S, Bruce A. Measuring Digital Pathology Throughput and Tissue Dropouts. J Pathol Inform. 2022;13:8. doi:10.4103/jpi.jpi_5_21
