The Problem
A radiotherapy treatment planning system used collapsed-cone convolution to calculate dose distributions, but the calculation was too slow for the iterative planning workflows clinicians needed, and accuracy checks against physical measurements showed worst-case errors as high as 10% — concentrated at tissue discontinuities such as bone-to-soft-tissue and tissue-to-air boundaries, where the physics is hardest to model correctly.
Our Approach
We profiled the dose engine to find where computation was being spent and restructured the kernel loops and data layout, bringing the calculation to roughly a third of its original runtime without changing the underlying physics model. In parallel, we tracked down the source of the measurement-to-calculation discrepancy by examining dose distributions pixel by pixel at tissue discontinuities, isolating the artifact to how the collapsed-cone kernel handled density gradients across boundaries, and corrected it.
Results
Facing a similar challenge?
Let's discuss how we can help you achieve similar performance gains.
Request an assessment