Internet-based gynecologic telecytology with remote automated image selection: results of a first-phase developmental trial.

Abstract

A retrospective set of 191 gynecologic cytology slides with reference interpretations was run on an automated screening device that selects fields of view (FOVs) based on a hierarchical probability of abnormality being present. An interface was developed between the device and a remote server using customized image review software. FOVs were reviewed by 3 cytotechnologists and 3 cytopathologists, and binary triage (unsatisfactory for evaluation/negative for intraepithelial lesion or malignancy [NILM] vs "abnormal" [neither unsatisfactory nor NILM]) and specific interpretations were done. No morphologic training before FOV review was provided. Three or more reviewers agreed on the correct categorization of NILM/unsatisfactory in 89% (85/96) and abnormal in 83% (79/95). Three or more reviewers triaged cases to abnormal as follows: atypical squamous cells of uncertain significance, 83% (5/6); atypical squamous cells, cannot exclude high-grade lesion, 100% (3/3); low-grade squamous intraepithelial lesion (SIL), 83% (52/63); high-grade SIL, 94% (17/18); and atypical glandular cells, 40% (2/5). This procedure may have comparable sensitivity and specificity and possibly could provide effective initial triage to further evaluation. A review of individual cases suggests that further accuracy can be achieved with additional training and experience.