A mapmaker drawing a route through a city doesn't sketch every brick in every wall. Streets, intersections, and distances are enough. The same goes for calculating radiation dose in breast tomography: you don't need every cell, just a faithful overall picture.
Previously, for each patient, a digital model of the breast was built from millions of tiny cubes (voxels), and the computer would churn through them for hours. The calculation took about 27 hours. That's like planning a trip across town by first studying the floor plan of every apartment in every building along the way.
It turned out that the cube size in the imaging plane affects the dose more than its thickness. Making the cubes 2.5–3 times larger in width and height changes the dose by 5–9%. But increasing only the slice thickness by the same 2.5–3 times shifts the dose by just 1–3%. So they reduced the resolution unevenly: more along the thickness, less in the plane. A final voxel size of 0.30 × 0.30 × 3.00 mm³ kept dose accuracy within 3.6–4.4% for all four samples. The number of cubes dropped by a factor of 20–31.
The total computation time shrank from 1600 minutes to 72 minutes. That's like a mapmaker swapping a detailed atlas for a tourist map and still not getting lost.
Previously, it was unclear whether such a coarse breast model could be used for synchrotron phase-contrast CT. Now it's clear: yes. The method will be useful in clinics where dose needs to be estimated quickly, before the main scan. The rough map turned out to be just as good as the detailed one, but 22 times faster.
The original phantom of one breast contained up to 95 million voxels, and after optimization only about 3 million remained. Yet the difference in the calculated dose was less than half a percent of the target radiation level.