2D-3D Registration Of Medical Image

Registering per-operative datasets to intra-operative image acquisitions can provide up-to-date information at the treatment site, aiding surgical guidance and other interventions. As three-dimensional intra-procedural image acquisition is uncommon in medical institutions, typically only two-dimensional image datasets can be obtained for such purposes. Although these images lack the spatial detail of volumetric data, they have the advantages of faster acquisition time and reduced amount of radiation exposure to both patients and doctors. Ideally, one can recover the advantages of the volumetric data by aligning the intra-operative 2D images with pre-operative volumes.We first discussed digitally reconstructed radiograph(DRR), and produced DRR by casting rays through a CT volume. By adjusting the parameters, we have got good results of two types of DRR images—spine image with no added structures and spine image with soft-tissue overlaid, as was the pre-processing work for image registration.Next, we introduced three kinds of similarity measures, such as mutual information, pattern intensity, and gradient difference; we also present the combined method of simulated annealing and descendent simplicity. Based on the above three similarity measures, we realized registration of two-dimensional images. All this will provide excellent bases for the future research. In this dissertation, we extended the registration of two-dimensional medical images into three dimension, realized the registration of 2D-3D medical images base on three different similarity measures and presented our comparison and evaluation of experiment results. Then we improved on the pattern intensity measure and got some satisfactory results. Experiment results show this improvement will speed up the calculation with the invariance of precision.