Whole Slide Imaging Validation Using Cervical Biopsies Yields Significant Interobserver Variability for Low Grade Dysplasias
Susan L Haley, Michael J Thrall. The Methodist Hospital, Houston, TX; Weill Cornell Medical College of Cornell University, New York, NY
Background: Whole slide imaging (WSI) involves scanning glass slides to produce digital images, which may be viewed remotely. Validation requirements have not been adequately addressed, and there are no standard guidelines for validating WSI for diagnostic use in the laboratory. Furthermore, few studies address the utility of WSI in challenging specimens such as cervical biopsies for dysplasia, when high resolution is particularly critical for accurate diagnosis.
Design: Fifty (50) cases with cervical biopsies (103 total specimens) were imaged at 20x magnification using BioImagene iScan Coreo Au scanners. Two examiners then blindly and independently evaluated the WSI using image-viewing software (Virtuoso). The examiners were aware of prior Pap test results for all cases. The histologic diagnoses were then compared to the original glass slide diagnoses. All diagnoses were stratified as: Negative, HPV/CIN 1, CIN 2, and CIN 3.
Results: One hundred and three (103) scanned specimens were collected. Of these, 1 slide had not been scanned and 3 were lacking coverslips or were overstained. Of 200 WSI diagnoses made, no specimen originally diagnosed as high grade dysplasia was called negative. There were 33 minor discrepancies and 14 major (calling a positive finding “negative” or two categorical differences). Seven cases (6.8%) could have had different clinical treatment based on WSI interpretation: 5 cases of HPV/CIN 1 were upgraded to CIN 2/3 and 2 cases of CIN 2/3 were downgraded to HPV/CIN 1 by at least one observer. The overall diagnostic accuracy was 83.5%, using the original glass-slide diagnosis as the gold standard.
|Original Diagnosis||Reviewer 1||Reviewer 2|
|Negative||HPV/CIN 1||CIN 2/3||Negative||HPV/CIN 1||CIN 2/3||N/A|