EN
Back to the archive

The encyclopedia · R&D & Science · Technical decision · 2016–2021

MSKCC automated radiotherapy planning so it didn't depend on one expert

MSKCC automated the manual tuning of radiotherapy treatment plans with constrained optimization, helping over 5,000 patients.

Memorial Sloan Kettering Cancer Center

the move

About half of all cancer patients get radiotherapy, and a successful treatment needs beams customized to spare normal organs while killing the tumor. That planning is labor-intensive and demands an expert planner for every patient.

Memorial Sloan Kettering Cancer Center developed and applied advanced optimization tools — hierarchical constrained optimization, convex approximations and Lagrangian methods — to automate treatment-plan tuning. The approach turns clinical priorities into an optimization that yields reliable, consistent, high-quality plans.

The result is a streamlined workflow where the quality of care no longer hinges on one expert's available time, making urgent treatment faster and more reproducible.

why it works

  • It converted an expert's manual, case-by-case tuning into a repeatable algorithm.
  • The optimization encoded clinical priorities directly, so plans reflected what mattered, not just what fit.
  • Removing the human chokepoint made high-quality plans available faster for patients in severe pain or urgent need.
  • Because it's automated and structure-based, it can scale to resource-constrained settings.
the payoffEncode the scarce expert's tuning into an algorithminspired

what transfers

When the bottleneck is a scarce expert doing the same delicate tuning by hand every time, encode their judgement into an optimizer and the skill stops being a chokepoint.

what came after

The automated techniques have been the foundation of high-quality treatments and have positively impacted over 5,000 patients, including many who might otherwise have needed longer hospital stays or unnecessary surgery. The work was a 2021 INFORMS Edelman finalist and published open access in the INFORMS Journal on Applied Analytics.

references

spotted an error? The archive wants to know.

same kind of clever