Identification of breast contour for nipple segmentation in breast magnetic resonance images

Abstract

The contributions of this study are twofold. First, a method was proposed to identify and segment the nipple in breast MR images. Second, a curve-fitting method was used to simulate the breast contour, allowing the breast to retain its original smooth shape.

Fully automated segmentation of whole breast using dynamic programming in dynamic contrast enhanced MR images

“…As mentioned in the introduction section, very few automated methods have been developed to process fat‐suppressed MR images in DCE‐MRI . Moreover, most of these studies did not provide sufficient details for us to compare the performance of our method and the referenced methods.…”

“…Hence, the sophisticated methods designed for nonfat‐suppressed MR scans could not be directly applied to the fat‐suppressed MR scans. To the best of our knowledge, few fully automated methods with sufficient evaluation of segmentation performance have been reported to process fat‐suppressed MR images in DCE‐MRI …”

“…The manual trace and the automatic trace agreed on 94.95% of the breast segmentation in their study. Gwo et al . proposed a method for chest wall detection and breast contour identification through curve fitting based on, respectively, only 4 and 9 T1‐weighted fat‐suppressed MR scans (160 slices per MR scan), and they achieved the lowest root‐mean‐square error of 1.1 pixels on average, the lowest centroid difference of 0.36 pixels, and the maximal Dice coefficient of 88.1%.…”

“…To the best of our knowledge, few fully automated methods with sufficient evaluation of segmentation performance have been reported to process fat-suppressed MR images in DCE-MRI. 13,[31][32][33][34][35] Yang et al 13 recently reported a new quantitative image analysis method, which was the first to combine BPE features in the pre-and the postcontrast MR scans for improving breast cancer diagnosis. In their previous study 31 , they obtained average overlap rates of 76.4% and 75.5% for benign and malignant lesions segmentation, respectively.…”

“…The manual trace and the automatic trace agreed on 94.95% of the breast segmentation in their study. Gwo et al 33,34 proposed a method for chest wall detection and breast contour identification through curve fitting based on, respectively, only 4 and 9 T1-weighted fat-suppressed MR scans (160 slices per MR scan), and they achieved the lowest root-meansquare error of 1.1 pixels on average, the lowest centroid difference of 0.36 pixels, and the maximal Dice coefficient of 88.1%. Ribes et al 35 presented an automatic segmentation method using a denoizing step and a Markov random field statistical model, and they achieved Jaccard coefficients of 76.9%, 75.6%, and 69.4% for adipose, glandular, and muscle and skin components, respectively, on 10 T1-weighted fatsuppressed breast MR images (144 slices per MR scan).…”