V. RESULTS

All experiments ran on an SGI Indigo2 Extreme, having one 150 MHz IP22 processor, 96 Mbytes of main memory, and a GU1-Extreme graphics board. To time the execution, the experiments executed the Unix time program to obtain a time estimate. In each instance, the time reported is the amount of user time. This time includes the time to read in the curvilinear data, resample, and create the vector field. The results indicate the new technique offers far superior performance compared to the previous algorithm, and slightly smoother images.

The major improvement of the new algorithm versus the old is execution speed. The old algorithm executes on average for 25 minutes. In comparison, the average time of the new algorithm is 50 seconds. This is an improvement by a factor of thirty. An intermediate stage of the new algorithm (which uses the min-max method of determining which ijk points influence which voxels) runs in 45 seconds. This demonstrates that the performance cost of 3D scan conversion is minimal.

The other improvement is in image quality (Appendix A, Figures 1 - 4). (The following conclusions are not based on the printed images in the Appendix. Those pictures do not clearly demonstrate the differences between the two algorithms, as do the high-resolution computer images.) As demonstrated in the images, the new algorithm generally produces smoother images than the original. This is most likely the result of errors in the original algorithm, caused by the min-max bounding box problem. In Figures 1 and 2, pressure is the value rendered. In Figures 3 and 4, temperature is rendered.

---