Preliminary Evaluation of Data Mining on Non-Masslike Enhancement of Breast Lesions on MRI

TAN Hong-na,SU Yi,LI Rui-min,CHEN Ying,WANG Pei-hua,TANG Feng,MAO Jian,SHEN Xi-gang,QIAN Min,GU Ya-jia
DOI: https://doi.org/10.3760/cma.j.issn.1005-1201.2009.05.002
2009-01-01
Abstract:Objective To evaluate the diagnostic values of the breast imaging reporting and data system-MRI (BI-RADS-MRI)description about non-masslike enhancement by data mining. Methods Fifty-five patients with non-masslike enhancement lesions showed on breast contrast-enhanced MRI were evaluated using two data mining algorithms (Logistic regression and decision tree) and 10-fold cross-validation methods. Results There were 28 malignant and 27 benign lesions. The most frequent findings of the malignant lesions were clustered ring enhancement and clumped enhancement [ 12 and 4 lesions, respectively; 84. 2% (16/19) in decision trees, partial regression coefficients in Logistic model were 2. 128 and 1.723, respectively], whereas homogenous, stippled, reticular internal and linear ductal enhancement were the most frequent findings in benign lesions [ 4、9、1 and 7 lesions, respectively; 72. 4% (21/29) in decision tree, partial regression coefficients in Logistic model were 0.357 (homogenous), 1. 861 (stippled) and 18. 870( reticular), respectively]. 10-fold cross-validation indicated that decision tree (C5.0) achieved an accuracy of 69.3% with a sensitivity of 66.7% and a specificity of 71.7% in comparison to the Logistic regression model with an accuracy of 57. 0%, a sensitivity of 43.3% and a specificity of 71.7%. Conclusions The diagnosis efficacy of non-masslike enhancement interpretation according to BI-RADS-MRI is not high. It is very important to find more potential features of non-masslike enhancement to improve the diagnosis accuracy.
What problem does this paper attempt to address?