Gaël Paul Hammer

Digital Cytology Combined With Artificial Intelligence Compared to Conventional Microscopy for Anal Cytology: A Preliminary Study

ABSTRACTIntroductionRecent studies have shown that digital cytology (DC) coupled with artificial intelligence (AI) algorithms is a valid approach to the diagnosis of cervico‐vaginal lesions using liquid‐based cytology (LBC). We evaluated the use of these methods for anal LBC specimens.MethodsA series of 124 anal LBC slides previously diagnosed by conventional microscopy (CC) were reviewed with a DC/AI system that generated a gallery of images. Diagnoses based on the selected images, according to the 2014 Bethesda System for Reporting Cervical Cytology, were compared to CC.ResultsOverall, CC and DC/AI approaches detected a similar number of abnormal (ASC‐US+) cases (63 and 62 cases, respectively). We observed an exact concordance between CC and DC in 70 (57.9%) cases, corresponding to a moderate agreement between the two approaches (κ = 0.41, p < 0.001). A moderate agreement (κ = 0.48, p < 0.001) was also found when positive cases were stratified into ‘low‐grade’ (ASC‐US, LSIL) and ‘high‐grade’ lesions (ASC‐H, HSIL). The DC/AI system detected more cases of higher severity (ASC‐H, HSIL: 9 and 2 cases, respectively) than CC (3 cases classified as HSIL).ConclusionsThe number of ASC‐US+ cases detected by both systems was similar. The DC/AI system detected more cases of higher severity compared to the CC.