计算机辅助检测系统在低剂量CT肺癌筛查中非钙化肺结节检出方法的研究

目的：探讨计算机辅助检测系统（computer-aided detection，CAD)提高低剂量CT筛查者非钙化肺结节检出率方法的优化和临床应用价值。方法：抽取100例接受低剂量CT筛查的高危人群进行研究，分别使用3种方法阅读图像。方法A：仅依靠CAD辅助阅片；方法B：根据CAD辅助自动检出结节结果，由影像诊断医师通过分析CAD自动检出的目标结构(结节或非结节结构)，以判定CAD自动检出的目标结果是否为真性非钙化肺结节；方法C：在方法B基础上，影像诊断医师通过薄层横断面图像阅片，分析并记录CAD结合影像医生薄层阅片共同检测结果。记录每种方法检出的每个患者结节总数、结节大小、结节位置和结节密度情况。最终以2名高年资影像诊断主任医师共同拟定的结节作为真结节参照标准。计算三种方法对非钙化肺结节的检出率、假阴性率和检出假结节总数，用ｘ²检验比较三种方法对非钙化肺结节检出率差异。结果：根据参照标准共检出287个真结节。方法A共检出结节总数336个，其中真结节238个，方法B共检出结节总数249个，真结节238个，方法C共检出结节总数285个，真结节数274个。方法C非钙化肺结节检出率95.50%明显高于方法A（ｘ² =23.434，P<0.001）和方法B（ｘ² =23.434，P<0.001）对非钙化结节检出率82.90%；方法C及方法B检出假阳性的非钙化肺结节数（11个）明显少于方法A（98个）。结论：影像医生薄层阅片联合修正CAD结果的筛查方式明显提高LDCT筛查者非钙化肺结节检出率并降低假阳结节检出，可以作为高危人群LDCT筛查肺结节的首选方法。

Study of Improve the Detection of Pulmonary Nodule in Lung Cancer Screening People with Low-dose Computed Tomography(LDCT)by Computer Aided Detection(CAD)

Objective:To study a more wise and effective way of improve the detection of pulmonary nodule in lung cancer screening people with low-dose computed tomography(LDCT)by computer aided detection(CAD).Methods:1 00 subjects diagnosed with lung nodules undergoing the routine low-dose CT screening were selected randomly.Three way was applied to find and analyze these datasets and marked all detected nodules and recorded the results simultaneously.The first way,only CAD was used to find and mark the possible lung nodules;The second way,a radiologist with 7 years working ex-perience was refered to adjust the result of the first way.The last way,the radiologist will read the thin axial image in the CAD workstation concurrently,to find more real pulmonary nodule on the base of the second way.The ground truth of nodules in the datasets was determined by two radiologists with more than 1 5 years working experience.χ2 test was per-formed to test whether there was significant difference between the sensitivities of three reading modes,and statistical signifi-cances were defined as P<0.05.Results:According to the ground truth,these two radiologists found out 287 true pulmo-nary nodules.The first way totally found 336 nodules,while the true pulmonary nodules were only 238.The second way found 249 nodules,the true pulmonary nodules were 238.The last way found 285 nodules,the true pulmonary nodules were 274.The sensitivities of pulmonary nodule detection by the last way (95.50%,P<0.001 )was significantly higher than the sensitivity of pulmonary nodule detection by the first and second way (82.90%).The last and second way found more less false positive pulmonary nodules(1 1 )than that (98)of the first way.Conclusion:The radiologist read the thin axial image with concurrently adjusted CAD result can significantly increase the sensitivities of pulmonary nodule screening in low-dose computed tomography(LDCT).It can be used as the first option to detect pulmonary nodules by CAD in high risk people with low-dose computed tomography(LDCT).