Study On Breast Lesions Computer-Aided Diagnosis Between Malignant And Benign Based On DCE-MRI

Breast cancer severely threats female's health. Early detection and diagnosis of breast lesions are important to formulate correct treatment for increasing patients'survival rate. The breast computer-aided technology (CAD) based on dynamic contrast-enhanced magnetic resonance (DCE-MRI) can provide a good platform. This paper systematically iinvestigates the overall framework of breast CAD, consisting of lesion segmentation according to DCE-MRI features; extraction and analysis of the dynamic enhancement features, texture features and morphological features; and discrimination between malignant and benign by the classifier design after feature seletion. The main work of this paper is as followed:First, Patient cases collation and classification are implementedbased on clinical knowledge of breast disease. And we compared segmentation results of OTSU method and Level Set method for breast lesions. It turns out that OTSU method is more suitable for real-time breast lesion CAD system.As for the complexity characteristics of breast lesions on DCE-MRI, we focus on the characteristics extraction and analysis. Because the generation of time intensity curve (TIC) is not descripted clearly in previous studies, this paper presents a region-based generation method which reduces image deformation caused by respiration or other movement and provides more accurate and reliable dynamic infomation. Then, we analyses classification performance of morphology, texture and dynamic enhancement features, and the results show that dynamic enhancement features performed superiorly. Moreover, after dimension reduction for redundant features, combination of nine features (smoothness, compactness, entropy etc.) performed excellently.As for the classifier selection, we compare four classifiers which are frequently-used in previous study, including SVM classifier, ANN classifier, KNN classifier, and Regression Tree classifier. The result shows that the SVM classifier performs best during the discrimination between benign and malignant breast lesions with the nine feature vector. It is useful for developing breast CAD system and improving doctor's work efficiencyFinally, we frame the DCE-MRI based breast CAD platform for future reasearch.