Weijie Chen MS, Presenter: Nothing to Disclose

Maryellen Lissak Giger PhD, Abstract Co-Author: Nothing to Disclose

Gillian Maclaine Newstead MD, Abstract Co-Author: Nothing to Disclose

Ulrich Bick MD, Abstract Co-Author: Nothing to Disclose

Li Lan, Abstract Co-Author: Nothing to Disclose

The purpose of this study is to develop an automated method for identifying characteristic signal-time curves in contrast-enhanced MR lesions of the breast using bias field correction as a preprocessing step, since clinical MR images are often corrupted by intensity inhomogeneity arising from imperfections of the surface coil.

We developed a fuzzy c-means (FCM) based algorithm that estimates the bias field in breast MR images. In our method, a multiplier field term that models the intensity variation is incorporated into the FCM objective function, which is minimized iteratively. In each iteration, the bias field is estimated based on the current tissue class centroids and the membership values of the voxels, and then smoothed by an iterative low-pass filter. After the images are corrected with the estimated bias field, we apply a standard fuzzy c-means algorithm to the serial 3-D lesions that are outlined by an experienced radiologist. The algorithm finds fuzzy cluster centers (i.e., signal-time curves) and assigns membership values to each voxel. The signal-time curve with maximum initial enhancement is selected as the characteristic curve of the lesion and the thresholded membership map is the identified region of the rapid enhancement. Four features are then extracted from the characteristic curves and are used for classification with linear discriminant analysis (LDA). The approach was applied to the analysis of 121 lesions (77 malignant and 44 benign). The differentiation performance of LDA output in leave-one-out cross evaluation was assessed using ROC analysis.

The merged features from FCM-identified signal-time curves yielded an Az value of 0.80, whereas the merged features from the signal-time curve obtained by averaging over the entire lesion yielded an Az value of 0.60 (p < 0.00001).

Our automated method significantly improved the performance of signal-time curves in the task of distinguishing between malignant and benign lesions.

M.L.G.,L.L.,U.B.: M.G., U.B., and L.L. are shareholders of R2 company.

Chen, W, Giger, M, Newstead, G, Bick, U, Lan, L, Computerized Analysis of Contrast-enhanced MR Images of the Breast: Automated Bias Field Correction and Identification of Characteristic Signal-time Curves.  Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL. http://archive.rsna.org/2004/4406162.html