Introduction With the endometrial cell test is widely used in the primary screening of high-risk population of endometrial cancer. It revealed the shortage of cytopathology experts and the imbalance of resources distribution.We introduced convolutional neural network into the screening and diagnosis of endometrial cancer and established a set of AI-assisted medical imaging analysis system that can automatically identify and differentiate benign and malignant endometrial cell mass. But the homogeneity in different hospitals has not been defined for clinical application.

Methods A retrospective study was conducted to select endometrial fluid-based cytological pathological sections from the First Affiliated Hospital of Xi ‘an Jiaotong University (Jiaotong University Group) and Xi ‘an Daxing Hospital (Daxing Group) from September 2021 to May 2023 due to abnormal vaginal bleeding or uterine abnormalities indicated by ultrasound, with 100 cases each. The results were reported by the AI-assisted medical imaging analysis system of the same model in the two hospitals.The accuracy, sensitivity and specificity were analyzed based on the pathological results of the patient‘s endometrial tissue as the gold standard.

Results The diagnostic accuracy of the AI-assisted endometrial cell medical imaging analysis system in the Jiaotong group and the Daxing group was 93.0% and 89.0%, respectively. The sensitivity was 87.8% and 82.1%, respectively, and the specificity was 96.6% and 91.7%, respectively. There was no significant difference in diagnostic accuracy, sensitivity, and specificity between the two systems (P>0.05).

Conclusion/Implications The AI-assisted endometrial cell medical imaging analysis system shows homogenization in the diagnostic accuracy, sensitivity, and specificity when used in different medical institutions.