Diagnostic Efficacy of All Series of Dynamic Contrast Enhanced Breast MR Images Using Gradient Vector Flow (GVF) Segmentation and Novel Border Feature Extraction for Differentiation Between Malignant and Benign Breast Lesions

Background/Objective: To discriminate between malignant and benign breast lesions; conventionally, the first series of Breast Subtraction Dynamic Contrast-Enhanced Magnetic Resonance Imaging (BS DCE-MRI) images are used for quantitative analysis. In this study, we investigated whether using all series of these images could provide us with more diagnostic information. Patients and Methods: This study included 60 histopathologically proven lesions. The steps of this study were as follows: selecting the regions of interest (ROI), segmentation using Gradient Vector Flow (GVF) snake for the first time, defining new feature sets, using artificial neural network (ANN) for optimal feature set selection, evaluation using receiver operating characteristic (ROC) analysis. Results: The results showed GVF snake method correctly segmented 95.3% of breast lesion borders at the overlap threshold of 0.4. The first classifier which used the optimal feature set extracted only from the first series of BS DCE-MRI images achieved an area under the curve (AUC) of 0.82, specificity of 60% at sensitivity of 81%. The second classifier which used the same optimal feature set but was extracted from all five series of these images achieved an AUC of 0.90, specificity of 79% at sensitivity of 81%. Conclusion: The result of GVF snake segmentation showed that it could make an accurate segmentation in the borders of breast lesions. According to this study, using all five series of BS DCE-MRI images could provide us with more diagnostic information about the breast lesion and could improve the performance of breast lesion classifiers in comparison with using the first series alone.