"Desired signatures should be mapped to 1.0 and undesired signatures to 0.0. Note that the PVB classification has consistently smaller standard deviations — the Eigen results have noise levels 2-4 times higher despite having 3-valued data to work with instead of the 2-valued data PVB was given.

Table 2 shows similar comparative results for volume measurements made between PPVC and PVB on simulated data, and between PPVC and Mixel on real data. Volume measurements made with PVB are significantly more accurate than those made with PPVC, and the PPVC to PVB improvement is better than the PPVC to Mixel improvement. Table 3 compares noise levels in PVB results and Eigen results. The noise level for the PVB results is about 25% of the level for the Eigen results.

Figures 2 and 5 also show comparative results between PVB and DML. Note that the same artifacts shown in Fig. 10 occur with real data and are reduced by the technique described here.

Models and volume-rendered images, as shown in Figs 11 and 12, benefit from the PVB technique because less incorrect information is introduced into the classified datasets, and so the images and models more accurately depict the objects they are representing. Models and images such as these are particularly sensitive to errors at geometric boundaries because they illustrate the underlying geometries.

Table 4 lists the datasets, the MRI machine they were

TABLE 2 Comparative volume measurement error for four algorithms (PVB, PPVC, Mixel, and Eigen)"






Was this article helpful?

0 0

Post a comment