Abstract: OBJECTIVE: To analyze the application effects of artificial intelligence (AI) in cytological diagnosis of cervical lesions.METHODS: 2 719 cervical TCT specimens were collected.AI-assisted and manual diagnoses were performed on all specimens to compare their consistency.Histopathological results were used as the gold standard.The accuracy,sensitivity,specificity and area under ROC curve of AI-assisted diagnosis and manual diagnosis were compared.RESULTS: The results of AI-assisted cytological grading diagnosis were basically consistent with results of manual diagnosis.AI-assisted diagnosis of low-grade lesions and inflammation was more accurate than manual radiography (P<0.01).In the diagnosis of high-grade lesions and cancer,the sensitivity of AI diagnosis was 82.1%,higher than 48.3% of manual diagnosis.The specificity of AI diagnosis was 94.3%,slightly lower than 95.0% of manual diagnosis.The area under ROC curve of AIassisted diagnosis (AUC=0.882) was larger than that of manual diagnosis (AUC=0.717),and the difference was statistically significant (P<0.01).CONCLUSION: AI-assisted diagnosis showed high accuracy in the diagnosis of cervical cancer,which should improve the coverage rate and the quality of cervical cancer screening in the general population.

Key words: cervical cancer, artificial intelligence, thinprep cytologic test, the bethesda system, assisted reading