TK
← Go back

Software that plans X-ray treatment and works out the dose

Where
Xstrahl
2012–2020
Founding engineer → CTO → COO

Hospitals plan every radiation treatment in detail before a beam is turned on, but in lab research much of it was not planned at all: researchers visually aimed the X-ray beam and turned it on. Our planning software became the most widely used planning system for preclinical research and helped make real planning the standard in the field.

I led the partnership with Kitware, one of 3D Slicer’s lead developers, that built the planner alongside our own developers, including a major GUI overhaul from KWWidgets to the Qt framework.

1 beam: the skin where it enters gets the most dose2 beams: the dose starts to gather where they cross3 beams: the tumor gets the most, every path far less (green)tumor1 Outline the tumoron the 3D scan2 Choose where thebeams come from3 Work out the doseeverywhere, on a GPU4 Send the plan straightto the machine
A plan on one slice of the scan, one beam at a time. With low-energy (kV) X-rays, one beam puts the most dose on the skin where it enters. Add beams from different angles and each path carries less, while the dose adds up where they cross, inside the tumor.

Hospital-grade planning, simple enough for biologists

Planning means outlining the tumor on the 3D scan and choosing where the beams come from, then checking how much radiation every part of the body gets. Hospitals have trained specialists for this, while in a lab the biologists do it themselves, so it had to feel simple. Our planner was built on the open-source 3D Slicer platform.

Dose math 100 times faster on an NVIDIA GPU

The dose math uses a GPU dose engine from Johns Hopkins, built on superposition-convolution, a standard method in clinical planning. On an NVIDIA GPU it ran over 100 times faster than a leading commercial planner on ordinary processors (in their tests on an NVIDIA GeForce GTX 280, under a second against 94 seconds). On the machine it runs on the control computer’s NVIDIA Tesla card, so a typical plan’s dose comes back in under a minute and a researcher can try a change and see the result.

A finished plan goes straight to the robot and the X-ray source. Plans carry over automatically when several subjects are scanned together and when the same subject comes back for more treatments.