Diffusion Weighted Magnetic Resonance Imaging Texture Biomarkers for Breast Cancer Diagnosis

Abstract

Quantification of breast lesion heterogeneity by means of MRI texture contributes in differentiating benign from malignant breast lesions. This study investigates the diagnostic performance of 1st and 2nd order Texture Analysis descriptors on Apparent Diffusion Coefficient (ADC) lesion maps. 78 histologically verified breast lesions (40 benign, 38 malignant) of 67 patients undergoing DW-MRI at 3.0 T, were analyzed. ADC maps were generated for a slice representative of lesion largest diameter. A two-step segmentation approach was applied on high b-value diffusion image, based on Fuzzy C-Means (FCM) clustering and edge-based contouring, for defining the lesion region contour. Lesion contour was transferred to ADC map and subjected to texture analysis by means of twelve first-order and eleven second-order texture features. Logistic Regression Classifier was employed to assess the diagnostic ability of individual features and feature combinations. Diagnostic performance was evaluated by means of the area under Receiver Operating Characteristic curve (Az). The highest classification performance (Az = 0.965 ± 0.024) was achieved by the combined feature subset 25th Percentile (1storder) and Entropy (2ndorder), suggesting the diagnostic significance of accurately quantifying lesion heterogeneity by texture-based feature combinations on ADC maps. Combined 1st and 2nd order texture biomarkers provide accurate spatial information of lesion ADC heterogeneity and holds potential in differentiating benign from malignant breast lesion status.