Sara Bazzon

Background: Whole‐slide imaging (WSI) is increasingly incorporated into digital cervical cytology workflows; however, interpretive variability in borderline and low‐grade squamous lesions remains a recognized diagnostic challenge. This study evaluated interobserver agreement among cytotechnologists in WSI‐based cervical cytology, focusing on borderline and low‐grade squamous lesions. Methods: Sixty‐eight cervical cytology cases (31 ASC‐US, 19 NILM, 18 LSIL) were independently assessed by 19 cytotechnologists across two WSI‐based evaluation rounds. One participant was excluded due to incomplete evaluations in Round 1. Interobserver agreement was evaluated using Fleiss' kappa, Kendall's W , intraclass correlation coefficient, Krippendorff's alpha, Cronbach's alpha and Spearman rank correlation. Conclusions: Digital assessment using WSI demonstrated moderate interobserver agreement among cytotechnologists, with the greatest variability observed in ASC‐US cases. Repeated evaluation was associated with high internal consistency. Findings are compatible with interpretive variability; WSI‐specific effects cannot be excluded without a direct comparator. Paired rater medians did not differ significantly (4.38 vs. 4.00 min; p = 0.090).

Introduction: High inter- and intraobserver variability is acknowledged in negative for intraepithelial lesion or malignancy (NILM), atypical squamous cells of undetermined significance (ASC-US), and low-grade squamous intraepithelial lesion (LSIL) diagnostic categories in cervical cytology. As digital cytology has emerged, whole-slide images (WSIs) have been compared to conventional light microscopy (LM) to evaluate its feasibility in routine practice. Methods: Six cytopathologists evaluated 20 liquid-based cervical samples diagnosed as ASC-US, LSIL, or NILM with known hrHPV status, both by conventional LM and WSI, equalling 240 evaluations. Interobserver agreement was measured using Fleiss’s kappa, Kendall´s coefficient and interclass correlation (ICC). Results: Most of the samples (93.75%) were assigned to the same original range of categories: NILM 5 (25%), ASC-US 10 (50%), and LSIL 5 (25%). WSI versus LM interobserver agreement varied with Fleiss’s kappa (0.324 versus 0.319), Kendall’s W (0.670 versus 0.572), and ICC (0.585 versus 0.414). Interobserver agreement for the ASC-US category was higher in Kendall’s W (0.435 for WSI versus 0.329 for LM) and in ICC (0.394 versus 0.118). Fleiss’s kappa showed lower interobserver agreement for ASC-US in WSI than LM (0.000 versus 0.039). Conclusion: This study illustrated that despite overall better agreement in WSI, ASC-US results were less favourable in WSI than LM, according to Fleiss’s kappa test, but not other tests.