Abstract：Objective To explore the method of measuring the sample quantity of artificial intelligence medical device test set
for quality evaluation, and verify it through experiments. Methods Two kinds of artificial intelligence medical devices for aided
diagnosis were selected to calculate the recall rate (sensitivity), specificity and accuracy with different sample quantity, and
the relationship between sample size and result deviation was analyzed. Results According to the theoretical calculation and
experimental analysis, the sample sizes of the fundus images of diabetic retinopathy and the auxiliary diagnostic software of lung CT
images that met the test requirements were calculated, which were 700 cases and 500 cases respectively. Conclusion In this paper,
the sample quantity estimation method for products with statistical epidemiological diseases is verified, and the sample quantity of
products without statistical epidemiological diseases is analyzed experimentally, and the recommended test sample quantity for two
types of products already on the market are obtained. On the premise of meeting the test requirements, the number of test set samples
should be minimized to save social resources.
2021 年第36 卷 08 期
VOL.36 No.08
专 论
FEATURES
4
孟祥峰，王浩，李佳戈. 人工智能医疗器械质量评价测试集样本量分析[J]. 中国医疗设备, 2021, 36(8): 4-7.
MENG Xiangfeng, WANG Hao, LI Jiage. Analysis of Sample Size of Artificial Intelligence Medical Device Test Set for Quality Evaluation. China Medical Devices, 2021, 36(8): 4-7.
