Analysis of Atypical Squamous (Glandular) Cells of Undetermined Significance Smears by Neural Network-Directed Review

Abstract

Objective: The objective of this study was to evaluate the value of neural network–directed review of smears determined to contain atypical squamous (glandular) cells of undetermined significance to identify those cases most likely to be associated with cervical intraepithelial neoplasia. Methods: One hundred sixty smears reported as atypical squamous (glandular) cells of undetermined significance on patients having colposcopy and directed biopsy within 1 year of the smear were identified. The smears were subjected to a neural network–directed review and classified according to findings on this review. The latter findings were related to those obtained on cervical biopsy. Results: One hundred sixty smears originally reported as atypical squamous (glandular) cells of undetermined significance were subjected to neural network–directed review. The smears were upgraded in 20.6% of cases. Ninety-one patients were found to have normal biopsies, and 69 had biopsies reported as abnormal. Of the smears in patients with abnormal biopsies, 37.7% were upgraded, whereas only 7.7% of smears from those with normal biopsies were upgraded (P < .001). Nine patients were found to have cervical intraepithelial neoplasia–3 on biopsy. Six of the nine smears (66.7%) taken on these patients were upgraded. Conclusion: Neural network-directed analysis of smears conventionally diagnosed as atypical squamous (glandular) cells of undetermined significance will reveal findings suggesting a squamous intraepithelial lesion in a significant number of cases. This approach requires further study because it is a relatively cost-effective means of triaging patients with a cytologic diagnosis of atypical squamous (glandular) cells of undetermined significance.