ABSTRACT
BACKGROUND: Digital cytology (DC) with artificial intelligence (AI) is a new approach. The authors compared DC with liquid-based cytology (LBC) using computer assistance (CAS) in a retrospective, noninterventional study. METHODS: In total, 1994 ThinPrep LBC slides (Hologic), which were previously analyzed in 2020 using an imaging system with CAS in routine cotesting for cytology/human papillomavirus, were reviewed in a blinded mode using the Genius Digital Diagnostics System (Hologic). In 555 cases, a histology result was available. The slides were digitally scanned (volumetric scan) at 14 levels integrated into one. AI algorithms were used to present a gallery of six tiles each (containing objects of interest) in five categories. Six additional tile rows were available, from which the diagnoses were made. All cases with a mismatch between DC and imaging system results were reviewed by an additional cytopathologist. RESULTS: In 86.56% of cases, a complete match between both systems was observed using the same cytology categories. When also considering the histology results, the match was 90.37%. In addition, when a cytology follow-up and/or a retrospective review was applied, the match reached 97.34%. In only 0.65% of cases was a major discrepancy observed (two grades of cytology or a low-grade squamous intraepithelial lesion/high-grade squamous intraepithelial lesion [LSIL/HSIL] shift), and none were identified by DC. Significantly more cases of higher severity (atypical squamous cells cannot exclude high grade [ASC-H], high-grade squamous intraepithelial lesion [HSIL]) were identified with DC, and its negative predictive value was higher. The screening time was significantly shorter with DC. CONCLUSIONS: With the Genius system for DC, the sensitivity for HSIL+/ASC-H and the specificity for LSIL and HSIL were superior to LBC and CAS. Screening time was significantly lower.