MULTIMODAL IMAGE REGISTRATION OF THE PROSTATE GLAND USING SURFACE-TO-VOLUME FITTING

摘要

This paper demonstrates an automatic surface-to-volume registration method and its application in monomodal and multimodal image matching. Its usage is also extended to automatic ultrasound image segmentation, with the aid of active model algorithm. The objective of this research is to intra-operatively locate the prostate gland under the ultrasound (US) guidance and enhance its view by registering it with the high-contrast MRI information. This will bring benefits to the US-guided prostate biopsy or therapy. The prostate surface, delineated by radiologist from the pre-operative MR images, is fed into the intra-operative US volume and transformed to best fit its counterpart. Due to the similarity between the geometric shape and the US image gradient for the same patient, the genetic algorithm is employed to search for the global optimum of the rigid transformation that will maximize the mean image gradients along the normals on the transformed surface. The registered surface will also provide a good guess of the prostate boundary in the US images, which is naturally positioned as the initial boundary for segmentation. Experiments on patient data show that the registration algorithm works well with rigid-body and small-deformed objects.