Research On Registration Of Breast Dynamic Contrast-enhanced MRI

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) of the breastis being used increasingly in the detection and diagnosis of breast cancer thanks to itsminimal invasiveness and to the possibility to visualize functional information. BreastDCE-MRI examination involves the imaging of the breast before and after injection ofcontrast agent. After injection, tumors demonstrate intense enhancement in grey valuecompared to normal tissue. Generally, contrast agent is intravenously injected. However,the side effect of the contrast agent becomes more obvious as the agent accumulates inpatients.In this case, a novel DCE-MRI has been adopted with contrast agent injectedthrough nipple. Nevertheless, it’s difficult to perform image registration, especially forthe area near nipple whose grey value changes greatly. In this thesis, a registrationscheme based on surface registration was proposed, which focused on SPHARM(spherical harmonics) analysis for surface registration. Image preprocessing andinterpolation of volume displacement were integrated to acquire registered image.As the grey value of the area near nipple was very close to the dark backgroundearly after the contrast agent was injected, it was hard to perform image segmentation.Therefore, an edge recovery method based on mathematical morphology was built,which ensured acceptable accuracy of the semi-automatic image segmentation andsurface reconstruction.The surface registration consisted of two steps: registration in orientation andconstruction of point correspondence. In the first step, conformal spherical mapping,regular sampling and SPHARM registration based on greedy search were successivelyapplied to align surface models of pre-contrast and post-contrast images. Afterwards,point-to-point correspondence between surface models was constructed with the sameSPHARM base and different SPHARM coefficients, based on the principle ofSPHARM analysis.In order to perform volume registration, interpolation of Mean Value Coordinateswas utilized to build volume displacement inside the surface model of post-contrastimage. With this primitive displacement field, a linear interpolation method for imagerecovery was designed to perform deformation for almost every pixel in the floatingimage. These two steps of volume interpolation finally achieved the registered image ofrelatively high recognition. The result image not only had the shape and outline of thepre-contrast image, but presented similar grey value information with the post-contrastimage as well. The proposed approach provides a basement for further analysis with DCE curve,which presents more detailed pathological information about the patients’ condition.