AI-assisted Pap screening improves specificity for atypical glandular cells and cervical neoplasia
The study reports a retrospective evaluation of the Hologic Genius Digital Diagnostics System for interpreting atypical glandular cells in 451 ThinPrep Pap specimens. Cytopathologists using the AI-assisted platform achieved adenocarcinoma detection sensitivities of 98.5% and 95.5%, matching original manual interpretations, while specificity rose to 84.6% and 85.6% compared with 27.7% for the legacy system. Positive predictive values for detecting CIN2/3 and adenocarcinoma also improved with digital AI triage. These findings indicate that integrating artificial intelligence into cervical cytology workflows can substantially reduce false-positive AGC classifications and enhance screening accuracy.
The original study
Assessment of atypical glandular cell interpretation in Pap tests using the Hologic Genius Digital Diagnostics System.
- Authors
- Jones TE, Matsko J, Wu D, Zeng X, Colaizzi A, Brown H, et al.
- Journal
- Cancer cytopathology
- Type
- Journal Article
- PMID
- 42622122
Original abstract
Atypical glandular cells (AGC) are a diagnostic challenge. The aim of this study was to evaluate the efficacy and diagnostic performance of AGC detection on the Hologic Genius Digital Diagnostics System (HGDDS). A retrospective analysis of 451 ThinPrep Pap cases was conducted, including 207 cases of AGC, 27 cases of high-grade squamous intraepithelial lesion (HSIL), 25 cases of low-grade squamous intraepithelial lesion (LSIL), and 192 benign cases. All AGC cases had follow-up histologic diagnoses, with 66 cases subsequently diagnosed as adenocarcinoma. The slides were randomized, scanned, and analyzed by the HGDDS. Patient age and HPV test results were provided to reviewers, an experienced cytologist, who screened the cases, followed by two cytopathologists who independently examined the cases on the HGDDS. Diagnostic concordance between the two cytopathologists indicated strong agreement (κ = 0.829). Sensitivity of AGC on Papanicolaou (Pap) tests for adenocarcinoma detection on HGDDS was 98.5% and 95.5%, respectively, comparable to the original ThinPrep interpretation (OTPI). Specificity for adenocarcinoma detection was significantly higher (84.6% and 85.6%) with the HGDDS than 27.7% with OTPI. Overall, the diagnostic performance for AGC/HSIL interpretation to detect CIN2/3/adenocarcinoma appeared to have improved with HGDDS compared with OTPI, particularly for specificity and positive predictive value (PPV). This is the first study evaluating AGC diagnosis using the HGDDS. The findings demonstrate that the sensitivity of adenocarcinoma detection as AGC on HGDDS is comparable to the ThinPrep Imaging System, but the specificity and PPV are improved. This suggests the potential of artificial intelligence to augment the performance of cervical cancer screening.