The GSF Voxel Computational Phantom Family
The rst voxel phantoms were developed at Vanderbilt University and at the GSFNational Research Center for Environment and Health, independently and more or less simultaneously. The phantom from Vanderbilt University was an adult female,14,15 and those from GSF were pediatric phantoms-an 8-week-old baby and a 7-year-old child.6,16-18 At the GSF, the pediatric phantoms, Baby and Child, were followed rst by Golem, the voxel phantom of an adult male,11 and then by nine individual voxel phantoms: three male (Frank, Godwin, and the Visible Human), and six female (Donna, Helga, Irene, Katja, Laura, and Klara), where one of them-Katja-was pregnant.12,19-22
All voxel phantoms constructed so far by our working group were based on computed tomographic (CT) image data of living patients, with the exception of the 8-week-old baby that was constructed from the image data of a dead body. All patients were scanned with a large number of contiguous axial slices. Each slice consists of a matrix of typically 256 × 256 or 512 × 512 pixels (picture elements, in a planar image). The volume elements are the pixels multiplied by the thickness of the slice. The single slice images are stacked, resulting in a three-dimensional (3D) array of voxels. In the primary image data, each pixel has a value that is characteristic of a certain physical property of the respective volume element; in the case of CT images this property is attenuation of x-rays of a speci c radiation quality. The pixel values are the “Houns eld numbers” or “CT numbers” that relate the attenuation property of a pixel to that of water (having the value 0) and range typically from −1000 (for air) to approximately between 500 and 1000 (for bone). Many image-processing software tools convert these Houns eld numbers to nonnegative integer “gray values” by adding a constant value to the Houns eld number of each pixel. Today, typically 12 bits are used to store these values, thus permitting a data range between 0 and 4095. Formerly, when storage needed to be used sparingly, the gray values were often limited to 255, thus reducing the storage requirement to 1 byte per pixel. Clearly, this renormalization was accompanied by a loss of image contrast.