Abstract: The current limitations of mammography where 75% of lesions detected are benign, and 15-20% of cancers are missed, have created an urgent need to develop novel imaging technologies for more effective early breast cancer detection. Initial work on breast magnetic resonance imaging (MRI) showed little promise due to the considerable degree of overlap between benign and malignant lesions. However, contrast enhanced MR breast imaging, developed in 1986, was able to differentiate between benign and malignant lesions [Heywang86]. A sensitivity and specificity range from 37%-97% [Heywang89, Obdejin96] was initially observed. There remained, however, a need to increase specificity and maintain sensitivity of MR mammography. A combination of systematic quantitative characterization of MR images of breast lesion and statistical tools such as stepwise linear discriminant analysis has been considered a possible strategy towards such increase of the sensitivity and specificity of MR mammography. This dissertation research involved: (1) Evaluating an optimal lesion segmentation method for application to MRI breast. Two methods were compared: temporal correlation and multispectral segmentation method. (2) Extraction of features from the segmented breast lesions. These extracted features were divided into three classes: CLASS 1: Boundary Descriptors, CLASS 2: Enhancement Profile, and CLASS 3: Texture Analysis. (3) Applying stepwise linear discriminant analysis to select the features which are the best classifiers. A training set of 43 patients were used to generate the optimal discriminant equation. The optimal features were then calculated in five new patients and their diagnosis was predicted. (4) Evaluating a fourth MRI-based characterizing feature, the apparent diffusion coefficient (ADC), and its potential to increase the classification accuracy for breast tumors. The statistical significance of the diffusion coefficient subsequently extracted for breast lesion was then evaluated as a potential classifier in addition to the features previously investigated. With the use of these statistical techniques to combine different classes of descriptors of the MR images of the breast lesions, it was established that the sensitivity and specificity of MR mammography could be enhanced significantly. With the use of such strategies, this approach shows significant potential as an important adjuvant modality to the radiologist's diagnosis of breast cancer.