Assessing the performance of benign and malignant breast lesion classification with bilateral TIC differentiation and other effective features in DCE-MRI.

Abstract

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is increasingly being used to detect and diagnose breast cancer. However, the effective methods based on quantitative feature analysis need to be explored. To investigate a new approach for improving the performance of benign and malignant breast lesion classification by combining bilateral time-intensity curve (TIC) differentiation features with other effective features. Retrospective. In all, 112 DCE-MRI biopsy-proven breast examinations (45 malignant tumors and 67 benign tumors). 3 T MR with a breast coil. Precontrast images were acquired before contrast administration. Subsequently, postcontrast images were acquired approximately every minute directly following administration of the contrast agent (gadopentetate dimeglumine). The postcontrast images contain eight sequences in total. Each sequence contains 78 slices. Accuracy, sensitivity, specificity, and the area under the curve (AUC) of the classification was calculated and compared with the published result. Five-fold crossvalidation was used. Areas under the receiver operating characteristic (ROC) curve (AUCs) of 0.8461, 0.7914, 0.8514, and 0.9058 were achieved using conventional features, unilateral TIC features, bilateral TIC differentiation features, and all the selected features, respectively. In terms of accuracy, the use of only unilateral TIC features or conventional features achieved an accuracy of 0.7321 or 0.8482 (sensitivity of 0.7556 or 0.7111 and specificity of 0.7164 or 0.9403). However, the accuracy increased to 0.9196 (sensitivity of 0.8889 and specificity of 0.9403) when bilateral TIC differentiation features were included. Bilateral differentiation TIC features can serve as a stronger indicator in differentiating benign and malignant breast lesions than the unilateral TIC features computed from one side of the breast only and some conventional features that are commonly used in MR image analysis. 1 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2019;50:465-473.