Application of Ultra-Fast Dose Calculation in Real-Time Interactive Treatment Panning
P Ziegenhein*, C Kamerling, H Heinrich, U Oelfke, German Cancer Research Center (DKFZ), Heidelberg, GermanySU-C-213AB-1 Sunday 1:30:00 PM - 2:15:00 PM Room: 213AB
We are investigating a new treatment planning paradigm: realtime planning via interactive dose shaping (IDS). IDS is a universal two-step process: First, a desired local dose variation is imposed for a 3D dose distribution by adjusting the correspondent fluence amplitudes. This dose modification inevitably causes unwanted dose changes outside the considered local dose area. Thus, second, a recovery strategy is applied to compensate for these unintended changes, so that the initial modification only has an influence on the desired region of the plan. The planning process could be carried out as a sequence of this Dose Modification and Recovery (DMR) steps. For implementing DMR strategies in realtime an ultra-fast local dose calculation within a few milliseconds is needed for evaluating the incrementally modified dose distribution.
The fast dose calculation algorithm is based on a pencil beam convolution technique using a set of measured pencil beam kernels whose convolutions with the beam fluence are carried out in a discretized spatial domain. This leads to three key features for our algorithm: i) A spatial-limited modification of fluence amplitudes requested by the DMR strategy can be considered as a relatively small incremental term to the convolution in space domain ii) The discrete convolution consists of simple multiply-add operations which are highly optimized on modern CPU and GPU hardware. iii) The trade-off accuracy/runtime can be adapted selectively for each fluence region of interest by controlling the radius of the finite pencil beam kernel.
The runtime of dose updates over a fluence area of 20x20 mm is about 10 ms on a Intel i7 CPU with an accuracy of 1% in the high dose region.
It could be shown that the proposed method of local dose update calculation is fast and accurate enough for our interactive realtime dose shaping planning approach.