The Usefulness of Bayesian Network in Assessing the Risk of Triple-Negative Breast Cancer

To evaluate a Bayesian network (BN) model learned from epidemiological and clinical information, and various MRI parameters for predicting the risk of triple-negative breast cancer (TNBC).

For this retrospective study, 214 women (mean age ± standard deviation, 50.5±10.6 years) with breast cancer were included between April 2016 and April 2018. All patients underwent MRI, including dynamic contrast-enhanced (DCE)-MRI. The morphologic MRI features, the pattern of the time-signal intensity curve (TIC) and the kinetic parameters were obtained for each lesion. The epidemiological and clinical parameters and those imaging parameters were used to construct BN model to estimate TNBC risk. ROC curves upon probability estimates were used to determine the performance of the BN using area under the ROC curves (A_z), sensitivity, specificity, and accuracy.

A BN model consisted of 16 epidemiological and clinical characteristics, morphologic MRI features, and quantitative DCE-MRI parameters were established. The posttest probability table showed that patients with age <35 years, mass-like lesions, type I TIC, and MaxCon ≥ 0.186 were at the highest risk of TNBC. The constructed BN model had an A_z of 0.663 (95% confidence interval [CI]: 0.654, 0.672), sensitivity of 0.660 (95% CI: 0.644, 0.675), specificity of 0.740 (95% CI: 0.726, 0.753) and accuracy of 0.724 (95% CI: 0.714, 0.733) in classifying TNBC.

The BN model integrating epidemiological and clinical characteristics, morphologic and kinetic MRI parameters provide a noninvasive analytical approach for preoperative prediction of the risk of TNBC.