Automatic Classification of Breast Tumors Using Features Extracted from Magnetic Resonance Images

Breast cancer is considered as the second leading cause of cancer deaths among women in the United States. Early detection of cancer is crucial in order to reduce its negative effects. Recently, magnetic resonance imaging (MRI) has become an important modality in the detection of breast cancer in daily practice. However, routine breast MRI has a moderate specificity that may increase its false positive rates. Therefore, automated detection techniques of malignancy can provide an important tool for clinicians. In this study, different data classification methods were examined to classify breast tumors screened using contrast enhanced MRI. The used data set included 20 subjects categorized clinically into two groups; benign and malignant tumors. MRI scans were first preprocessed to extract imaging features. Then two classification methods were exploited to differentiate between the two tumor's categories using the extracted features. The used classification methods were K-Nearest Neighbor (KNN), and Linear Discriminant Analysis (LDA). The results show a relatively significant classification accuracy compared with pathological analysis, and also the calculated resubstitution error. In summary, the proposed automatic classification techniques can be used as noninvasive diagnostic tools for breast cancer, with the capability of decreasing false positive errors associated with regular MRI diagnosis.