Dynamic Contrast Enhance-Magnetic Resonance Imaging (DCE-MRI) has proved to be a useful tool for diagnosing mass-like breast cancer. For non-mass-like lesions, however, no methods applied on DCE-MRI have shown satisfying results so far. The present paper uses the Independent Component Analysis (ICA) to extract tumor enhancement curves which are more exact than manually or automatically chosen regions of interest (ROIs). By analysing the different tissue types contained in the voxels of the MR image, we can filter out noise and define lesion related enhancement curves. These curves allow a better classification than ROI or segmentation methods. This is illustrated by extracting features from MRI cases and determining the malignancy or benignity by support vector machines (SVMs). Next to this classification by kinetic analysis, ICA is also used to segment tumorous regions. Unlike in standard segmentation methods, we do not regard voxels as a whole but instead focus our analysis on the actual tissue types, and filter out noise. Combining all these achievements we present a complete workflow for classification of malignant and benign lesions providing helpful support for the fight against breast cancer.