Evaluating Diagnostic Concordance of 3-D Virtual Gynecologic Imaging vs. Light Microscopy: Follow-Up Study
Whitney R Wedel, Adam J Horn, Maheswari Mukherjee, Julia Bridge, Mary McGaughey, Amber Donnelly, Elizabeth Lyden, Jane Meza, Stanley J Radio. University of Nebraska Medical Center, Omaha, NE
Background: Virtual microscopy (VM) in cytology has been limited by the inability to focus through 3-dimensional cell clusters with single focal plane (2-dimensional) imaging. We previously established an optimal scanning interval of one micron, and further investigated the optimal number of scanning focal plane levels (3). The goal of this study was to confirm if 3 focal plane levels at 1 micron is comparable to light microscopy (LM) for gynecologic specimens.
Design: Two sets of 96 SurePathTM slides (192 slides total) were scanned at 40X magnification with 3 focal planes at one micron using an iScanCoreo Au scanner (Ventana, AZ, USA), generating two sets of virtual images. The four subsequent sets (two glass, two virtual – 384 slides total) were evaluated by two cytopathologists, two pathology residents, and one cytotechnologist. Bioimagene's Image Viewer (version: 188.8.131.52-RC8) and conventional LM were used to diagnose pre-annotated cells using selected standard squamous epithelial Bethesda diagnostic categories. The slides were evaluated in sets which were presented in variable order.
Results: The intra- and interobserver reliability were evaluated using Kappa statistics. Eight cases (4.2%) of VM slides were designated (by at least one observer) “unable to diagnose” secondary to poor image quality. These cases were included in the final statistics. The overall intraobserver diagnostic concordance between glass and virtual slides was very high (87-97%). Interobserver agreement (in which all five observers gave the same diagnosis) was evaluated, and for all cases was found to be higher for LM (94%) compared to VM (82%). In positive cases (low-grade or high-grade squamous intraepithelial lesion), the interobserver agreement was also higher for LM (76%), and lower for VM (55%). The overall concordance between all observers (LM and VM) was 76%. Concordance of LM and VM with established biopsy diagnosis was lower.
Conclusions: Excellent intraobserver concordance between LM and VM was established. While interobserver agreement was higher for LM, it was still very good using VM. An additional study incorporating more stringent quality control of virtual images may be required to fully establish biopsy:LM:VM diagnostic concordance and method equivalence.
Monday, March 4, 2013 9:30 AM
Poster Session I Stowell-Orbison/Surgical Pathology/Autopsy Awards Poster Session # 61, Monday Morning