Gynecologic Telecytology Using Automated Local Image Selection with Remote Interpretation: The Results of a Phase 2 Prospective Trial.
John H Eichhorn, Barbara A Crothers, David C Wilbur. Massachusetts General Hospital, Boston; Walter Reed Army Medical Center, Washington, DC
Background: Internet-based telecytology would allow for remote interpretation of cervical cytology specimens and has the potential to facilitate regionalization. We have previously shown that remote automated field-of-view (FOV) image selection and transmission can allow for successful interpretation of cervical cytology cases (Am J Clin Pathol 2008;12:686). To test the feasibility of this method in a typical high risk patient population a prospective study was initiated.
Design: Human subjects approvals were obtained. Patients were consented and prospectively enrolled from 2 colposcopy clinics – one collecting Thin Prep cervical cytology specimens and the other SurePath (SP) samples. Slides were scanned by a customized FocalPoint device, the 30 highest risk FOVs were identified, and low resolution images were automatically captured and transmitted to a customized remote reading station. Images were interpreted by a single observer and categorized as normal or abnormal for triage. Interpretations in the test arm were compared to the original glass slide results made as part of each patient's routine care.
Results: 274 patients were enrolled (155 TP, 119 SP). The overall sensitivity of a positive case (ASC-US or greater (+)) triage was 85% with a specificity of 67%. The positive predictive value (PPV) was 44% and the negative predictive value (NPV) was 93%. For LSIL+ and HSIL+, the sensitivities were 79% and 100%, respectively. When examined by specimen type, SP performed better than did TP with sensivity/specificity for ASC-US + of SP being 92%/66% and for TP 75%/68%. The overall PPV/NPV (ASC-US+) for SP and TP were 55%/95% and 34%/93%, respectively. For LSIL+, SP sensitivity was 90% and TP sensivity was 62%. Sensitivity of both specimen types was 100% for HSIL+.
Conclusions: Sensitivity for abnormal SP slides was at or above the level of manual glass slide screening. Specificity, as expected, was lower using remote low resolution static images as observers tend to overcall. The telecytology method performs in a manner that would be acceptable as an initial triage to further on-site glass or remote whole slide image review (for SP 74% of cases were correctly classified and only 3% of cases would have been triaged erroneously to a patient's detriment). TP may not be optimal as the lower cell density on the TP presents fewer cells in each FOV, hence decreasing overall sensitivity; and the FocalPoint device has been optimized for its primary clinical application of screening SP slides.
Tuesday, March 1, 2011 2:15 PM
Platform Session: Section D, Tuesday Afternoon