Design And Application Of Transfer Function For 3D Image Volume Rendering

Volume rendering is an important visualization technology for exploring and visual-izing structures of interest in 3D medical images.Transfer functions?TFs?,which decide how the different structures in the 3D images should be displayed by defining a mapping between data value and color/opacity,are particularly important in volume rendering.Un-fortunately,designing an appropriate transfer functions proves difficult in complex 3D images.Therefore,Pat Hanrahan called it one of the top 10 problems in volume visual-ization in his inspiring keynote address at the 1992 Symposium on Volume Visualization.By analyzing the current TFs,we can find that they have some disadvantages such as poor classification ability or complicated interactions.In this paper,many important problems in the design of transfer function are studied,including how to improve the classifica-tion ability of low dimensional transfer function?How to classify the different organs of complex abdominal plain CT images?How to repair the defects of the volume rendering results generated from the improper design of the transfer function?Finally,the improved transfer function is applied to the visualization of clinical medical images,revealing the effectiveness of segmentectomy and diagnosis and treatment of cardiovascular disease.The innovation and organization of this paper are as follows:1.This paper first proposes a separation method of structures with similar attributes in low dimensional TF?LD-TF?space.Different structures with similar attributes have the superposition region in LD-TF space,these structures and non-interesting structures and small fragments will be visualized simultaneously.The LD-TF thus often fails in sep-arating these structures?or their boundaries?and has limited ability to classify different structures in real-world 3D images.Here,according to spatial topological relationship between structures with similar attributes in the 3D image,they are divided into the fol-lowing three classes:?₁—far away??₂—close neighbor and?₃—touch together,and different boundary-separation techniques are applied to them,respectively.The complex task of separating various boundaries in 3D images is then simplified by dividing it into several small separation problems.Spatial connectivity,set operations and combined TFs are applied in?₁??₂and?₃,respectively.The method shows good object classification ability in real-world 3D images while avoiding the complexity of high-dimensional trans-fer functions.2.In abdominal plain CT images,it is a challenge to design an appropriate TF for multi-organ visualization.In this paper,we proposed a multi-atlas based multi-organ clas-sification framework,to accurately visualize multiple organs in abdominal CT images.Specifically,a saliency map of an organ in a 3D image is generated by combining the spatial distribution probability and the attribute-distribution probability.Here,the spatial distribution probability of the organ in the 3D image is determined roughly by register-ing multi-atlas to the 3D image,and then fusing each warped atlas segmentation.The attribute-distribution probability of the organ is generated by computing the organ proba-bility for each voxel of the 3D image,based on the organ-specific intensity-gradient distri-bution histogram?i.e.,a 2D feature space?obtained from multi-atlas.For multiple organs,we will have multiple saliency maps,one for each organ.Finally,to further improve the quality of visualization,we used fuzzy connectedness in the saliency map to assign dif-ferent opacity for each voxel in the CT image.Experimental results have demonstrated the effectiveness of the proposed method in the abdominal CT images.3.In order to generate high-quality boundary,we propose a modified Allen-Cahn equation,which has the motion of mean curvature,to repair the visualized boundaries in respect to recover lost boundary voxels and fill holes.One can observe that above defects of visualized boundaries are usually induced by losing some boundary voxels when classifying boundaries from 3D images.Hence,we will repair indirectly boundaries in the volume rendering by recovering lost boundary voxels based on the known boundary in 3D images.The boundary repairing problem in this paper is modeled as a constrained diffusion of known boundary voxels in a 3D image,and described by a modified Allen-Cahn equation which has the motion of mean curvature.Under the mean curvature flow,the boundary in the repairing region will move faster than that in the non-repairing region under mean curvature flow due to its higher mean curvature.Once the geometric boundary is moved,the voxel near the boundary will be filled by diffusing the information from the nearby region.By the constrained diffusion,visualization quality of boundaries can be greatly improved in the volume rendering.Many experimental results have demonstrated the effectiveness of the proposed method.4.Finally,the transfer function design of volume rendering proposed in this paper is applied to medical imaging.In the resection of lung cancer,our method provides an in-tuitive spatial location relationship between the lung,pulmonary nodules,and pulmonary fissures,to solve the problem of inaccurate cutting caused by the inaccurate location of pulmonary nodules in segmentectomy for clinicians.In the cardiovascular disease,our method provides a visual visualization of the spine,the arteries and the heart,avoiding the risk of stent falling off caused by mismatch of vascular stent and vascular structure in stent implantation.Currently,the framework has already been tested and used in Shanghai Ninth People's Hospital and Shanghai Chest hospital.