Application of breast health intelligent detection system based on deep learning technology in breast tumor detection

• DOI: 10.3877/cma.j.issn.1674-0807.2019.01.006
• Publication Date: 2019-02-01
• Authors: Zhangjun Song; Huxia Wang; Jing Zhao; Bin Zhao; Ming Zhou; Liang Xiu-fen; Xiaomin Yang; Pihua Han; Nan Chen; Sai He; Youxi Wang; Yanni Hou; Yongguo Fan
• Keywords: False positive rate; Gold standard (test); Kappa

Objective To investigate the value of MammoWorks™ system based on deep learning technology in breast tumor detection.

Methods We enrolled 448 patients with lesions at X-ray BI-RADS grade 5-6 Shaanxi Provincial Tumor Hospital from January 2015 to April 2017. All underwent operation, complete clinical and pathological data. Additionally, using a random number table, 215 healthy people who had physical examination our hospital same period (X-ray 1) were randomly as control all them no diseases two-year follow-up. The mammographic data subjects analyzed by system. With results 2-year follow-up gold standard, sensitivity, specificity, accuracy, positive predictive value, negative likelihood ratio detection tumors analyzed, well false-positive marks one mammograph. rate comparison was performed χ2 test Fisher exact probability method. mammograph compared nonparametric (Kruskal-Wallis test) Kappa used evaluate consistency between different subgroups.

Results Totally 2 652 photographs 663 females analyzed. A total 284 marked, including 929 true-positive 1 355 false-positive. There 333 cases true-positive, 126 true-negative, 89 115 false-negative. sensitivity 74.3%(333/448), specificity 58.6%(126/215), 78.9%(333/422), 52.3%(126/241), accuracy 69.2%(459/663), 1.80 0.44. 0.50(0.00~0.75). showed significant difference craniocaudal (CC) view mediolateral oblique (MLO) (Kappa=0.278, P 0.050), while efficiency significantly density, lesion type, diameter (χ2=7.985, 15.543, 18.652, P<0.050). presented density (χ2=15.024, P<0.050).

Conclusion Based technology, is helpful auxiliary diagnosis tumors, but its still needs be improved.

Key words: Breast neoplasms; Artificial intelligence; Deep