Abstract:Objective To analyze performance variation of the same product on different data, compare results on nodules of different
sizes, and test generalizability of computer aided detection algorithms performance for pulmonary nodules. Methods An independent
test set was imported into the pulmonary nodule aided detection software. The software output and the reference standard of all test
sets were matched. The recall rate and accuracy were used as the main detection indexes, and the Dice coefficient was used as the
main segmentation index. The indicators after screening the product output and reference standards were recalculated according to
three methods (Method 1: The output results of all products were matched with the reference standard within a certain size range to
calculate the detection index. Method 2: The output results of a product within a certain size range were matched with the reference
standard within the size range to calculate the detection index. Method 3: The recall rate was calculated by method 1, the output
results of a product within a certain size range were matched with all reference standards, and the accuracy was calculated). After
screening the product output and reference standards, the indexes were recalculated, and the test results of recall rate, accuracy and
Dice coefficient of lesions in different size ranges were compared. Results The result of recall calculated by method 2 was the
lowest, the result of accuracy calculated by method 3 was the highest. There were significant differences in recall and Dice coefficient
with the change of nodule size. Conclusion For computer aided detection products of pulmonary nodules, when the intended use of
the products is related to the size of nodules, the specific test rules have a significant impact on the test results, and it is necessary to
clearly describe them.
孟祥峰,李佳戈,郝烨,王浩. 病灶大小对肺结节辅助检测产品测试结果的影响[J]. 中国医疗设备, 2022, 37(10): 14-17.
MENG Xiangfeng, LI Jiage, HAO Ye, WANG Hao. Effects of Lesion Size on Test Results of Computer Aided Detection Products for Pulmonary Nodule. China Medical Devices, 2022, 37(10): 14-17.