| 3D visualization of medical image extracts human organs from 2D image and demonstrates them on computer screen as 3D models,which helps doctor diagnosis.As the basis of 3D visualization,medical image segmentation needs to part the region of organ as the pre-processing procedure.The segmentation performance will directly affect 3D visualization.Medical image segmentation and visualization are related closely to benefit medical diagnosis,surgical simulation and human.anatomy etc.,which attract enormous research interests in medical field.This paper implements target segmentation and 3D visualization of it based on a set of medical images captured by the medical device.The target segmentation has two main steps,standard layer segmentation and neighbor layers segmentation using the result of first step as initial template.3D visualize the target after the segmentation.The medical image segmentation is the preparation of 3D visualization.This paper leverages the target silhouette relevance among neighbor layers of the medical image set.It proposes a three steps optimized segmentation algorithm based on Graph-Cut and VFCCV-Snake.We first choose a high quality image from the input set as the standard layer.Using a modified Graph-Cut method to extract the target silhouette as initial template compared with its neighbor layer.Then,we extract neighbor layer ’s coarse silhouette as the initial setting of VFCCV-Snake to perform the neighbor layer segmentation.Repeat second step until target segmentation finished on every layer.The experiment demonstrates that our algorithm has the advantages of good segmentation and high efficiency with only simple interaction on standard layer.For 3D visualization of medical image,this paper focuses on Ray-Casting algorithm in volume rendering.In order to solve the low speed of volume rendering,we propose a GPU and normal template rendering based accelerating algorithm to improve the efficiency of 3D visualization of medical image.Meanwhile,the cutting and interactive manipulation of 3D target model are also implemented in this paper. |
