关键词:

text-indent:0pt, "> 人工智能, 乳腺病变, 良恶性, BI?RADS 分类, CAD

Abstract:

Objective To study the application value of CAD machine diagnosis system in ultrasound diagnosis of breast cancer. Methods A total of 5 311 patients with breast lesions who underwent breast biopsy or surgical resections in our hospital were enrolled in the study. With the pathological results as the standards，the diagnosis results by physicians′ image reading were compared with the results by CAD machine diagnosis system as well as the results by combined application of the two approaches. Results （1）The area under the ROC curve of the diagnosis results of the doctor′s image reading was 0.837，and the sensitivity，specificity，and accuracy were 91.65%，75.67%，and 81.98%，respectively.（2）The area under the ROC curve of the CAD system diagnosis results was 0.880，and the sensitivity，specificity，and accuracy are 87.41%，88.64%，and 88.16%.（3）The mis⁃ diagnosed cases by the physicians′ image reading and the CAD system were analyzed to adjust the level of some BI⁃RADS classifications. After the adjustment，the area under the ROC curve with the combined application of the two approaches reached 0.957，and the sensitivity，specificity，and accuracy were 96.91%，88.41%，and 91.56%， respectively.（4）There were significant differences in the ROC curves between the three methods（all P < 0.05）. The physicians′ image reading was significantly different in the specificity and accuracy from the combined application of the two methods（P < 0.05）. The CAD system was significantly in the sensitivity from the combined application of the two methods（P < 0.05）. The CAD system was significantly in the specificity from the combined application of the two methods（P < 0.05）. Conclusion Compared with the physicians′ image reading，the CAD machine diagnosis system alone and the combined application of the two methods after optimizing BI⁃RADS classification are both better at the diagnosis of benign and malignant breast cancer.

Key words: