Use of T2-weighted Images in Computer-aided Diagnosis for Breast MRI

Purpose: To evaluate the use of lesion features derived from T2-weighted images in computer-aided classification of benign and malignant masses in dynamic contrast-enhanced (DCE) magnetic resonance imaging studies of the breast. Methods and Materials: Fat-suppressed DCE-T1-weighted and non-fat-suppressed T2-weighted images of 50 patients containing 64 masses (28 benign and 36 malignant) were acquired. All masses were segmented in the DCE-T1-weighted images using a semi-automatic method. Twenty-two morphological and kinetic DCE-T1-weighted features proposed in literature were computed for each mass. Additionally, 14 quantitative T2-weighted features were defined and evaluated. These features are mainly based on the intensity characteristics within the lesion or within the enclosing shell of normal tissue. The intensity values were normalized with respect to the mean intensity of different reference regions to account for the qualitative and inhomogeneous nature of MR images. Feature selection was performed to determine the best set consisting of not more than five DCE-T1-weighted features and the best set combining not more than a total of five DCE-T1-weighted and T2-weighted features. Subsequently, a linear classification function was estimated by Fisher discriminant analysis. Results: In a leave-one-out evaluation, including T2-weighted features improved the classification performance from an area-under-the-ROC-curve of 0.94 to 0.99, yet a test for statistical significance failed with p=0.064. The selected T2-weighted features reflect the existence of bright structures in the surrounding tissue (caused by e.g. edema) and dark areas within the lesion. Conclusion: The obtained results encourage the use of quantitative information from T2-weighted images in CAD for breast MRI