Computerized Segmentation and Classification of Breast Lesions Using Perfusion Volume Fractions in Dynamic Contrast-enhanced MRI

Abstract

This study is designed to segment suspicious regions using automatic computerized procedures and to classify kinetic patterns using commercially available three-time-points (3TP) method of computer- aided diagnosis. A novel evaluation method using perfusion volume fractions is introduced for examining meaningful kinetic features in differentiation of benign and malignant breast lesions. Dynamic contrast- enhanced MRI was applied to 24 lesions (12 malignant, 12 benign). Thresholding for suspicious regions, region growing segmentation, hole-filling and 3D morphological erosion and dilation were performed for extracting final lesion volume. The lesion sphericity and center distance of mass to surface area ratio (CDMSAR) were considered in the process of automatic segmentation. The kinetic patterns for each lesion were classified into six classes by the 3TP method. Perfusion volume fraction for each class was calculated in three partitions of whole, rim and core volumes of a lesion. Receiver operating characteristic curve (ROC) analysis was performed using the perfusion volume fractions. When using perfusion volume fractions divided into rim and core lesion volume, the classes having more improved accuracy appeared than using perfusion volume fractions within whole lesion volume. This result indicates that lesion classification using local perfusion volume fractions is helpful in selecting meaningful kinetic patterns for differentiation of benign and malignant lesions.