Characterizing fMRI Activations within Regions of Interest (ROIs) Using 3D Moment Invariants
CVPRW '06 Proceedings of the 2006 Conference on Computer Vision and Pattern Recognition Workshop
Segmentation and classification of breast tumor using dynamic contrast-enhanced MR images
MICCAI'07 Proceedings of the 10th international conference on Medical image computing and computer-assisted intervention
| Hi-index | 0.00 |
This study was designed to characterize the spatio-temporal properties of intratumoral enhancement patterns by using voxel-wise temporal enhancement spectra and morphometry of their spatial distributions in dynamic contrast-enhanced (DCE) breast MRI. Discrete Fourier transformation (DFT) and singular value decomposition (SVD) were used to extract the temporal enhancement features for comparison, generating 4D spectral maps. The spatial variations of DFT and SVD-based eigen spectra within tumor were captured by 3D moment descriptors, respectively. Differentiation between benign and malignant tumors was carried out using least squares support vector machine (LS-SVM) with a radial basis function (RBF) kernel and leave-one-out cross validation was used for performance evaluation. Using DFT, the sensitivity, specificity and area under ROC curve were 84.8%, 64.4% and 0.728. Using SVD, the corresponding values were 100%, 86.7% and 0.935. Combination of SVD and 3D moments yields higher performance in tumor differentiation than that of DFT and 3D moments.