The MR scans of the knee used in this work consist of a set of 30
non-contiguous 2D images, 256x256 pixels, which can be interpolated to a
3D volume. Originally the slices are 4mm thick with a gap of 0.4mm
between adjacent slices. Linear interpolation gives 186 slices and cubic
voxels, 0.704mm
. The patient is positioned in the scanner such that the very end of the
tibia, the joint and approximately 100mm of the femur is imaged.
The femur is comprised of spongy bone surrounded by compact dense bone, as shown in figure 1 . This compact bone does not emit a signal in MR and results in a black ring identifying the bone. The rounded ends of the femur, known as the condyles, do not have the surrounding dense bone, thus often a very subtle change in grey level intensity is the only indication of a boundary. The spongy bone is imaged such that it produces a large distribution of intensities that are not unique to it. For our purposes, the segmentation of the bone comprises only the spongy tissue, so the beginning of the compact bone is used as the boundary of the segmented femur.
N Hill