VOLUMETRIC VISUALIZATION OF ANATOMY FOR TREATMENT PLANNING

Purpose: Delineation of volumes of interest for three-dimensional (3D) treatment planning is usually performed by contouring on two-dimensio nal sections. We explore the usage of segmentation-free volumetric ren dering of the three-dimensional image data set for tumor and normal ti ssue visualization. Methods and Materials: Standard treatment planning computed tomography (CT) studies, with typically 5 to 10 mm slice thi ckness, and spiral CT studies with 3 mm slice thickness were used. The data were visualized using locally developed volume-rendering softwar e. Similar to the method of Drebin et al., CT voxels are automatically assigned an opacity and other visual properties (e.g., color) based o n a probabilistic classification into tissue types. Using volumetric c ompositing, a projection into the opacity-weighted volume is produced. Depth cueing, perspective, and gradient-based shading are incorporate d to achieve realistic images. Unlike surface-rendered displays, no ha nd segmentation is required to produce detailed renditions of skin, mu scle, or bony anatomy. By suitable manipulation of the opacity map, ti ssue classes can be made transparent, revealing muscle, vessels, or bo ne, for example. Manually supervised tissue masking allows irrelevant tissues overlying tumors or other structures of interest to be removed . Results: Very high-quality renditions are produced in from 5 s to 1 min on midrange computer workstations. In the pelvis, an anteroposteri or (AP) volume rendered view from a typical planning CT scan clearly s hows the skin and bony anatomy. A muscle opacity map permits clear vis ualization of the superficial thigh muscles, femoral veins, and arteri es. Lymph nodes are seen in the femoral triangle. When overlying muscl e and bone are cut away, the prostate, seminal vessels, bladder, and r ectum are seen in 3D perspective. Similar results are obtained for tho rax and for head and neck scans. Conclusion: Volumetric visualization of anatomy is useful in treatment planning, because 3D views can be ge nerated without the need for segmentation. When relationships among an atomical structures, rather than geometric models of them, are importa nt, volume rendering presents advantages. The presented algorithm is r eadily adaptable to distributed parallel implementation on a network o f heterogeneous workstations.