Dual-energy computed tomography for improved proton therapy treatment planning

Purpose/Objective:

Cancer treatment with protons requires an accurate prediction of the particle’s range in tissue. In cCurrently clinical practice, computed tomography (CT) images are used to voxelwise translate the CT number into the tissue’s stopping power relative to water (SPR) via heuristic relations (HLUT). However, the general validity of this approach is limited due to the different physical interaction processes of photons and ions. The resulting range uncertainty is taken into account in the treatment plan by adding a safety margin around the tumor, effectively limiting the potential benefits of particle therapy over conventional radiotherapy. The use of dual-energy CT (DECT) allows for a direct derivation of tissue parameters, resulting in a better characterization of the tissue. The potential of a DECT-based,and ultimately allows a patient-individualized range prediction (DirectSPR) has been shown in previous work. In 2015, we were first to introduce DECT scans for routine clinical treatment planning, still using a generic HLUT. Here, we portray the next steps towards the full clinical implementation of DirectSPR, namely its validation and assessment of its clinical benefit.

Material and Methods:

To validate the method for realistic clinical scenarios, its accuracy was investigated in an anthropomorphic head phantom as well as in porcine biological tissue. Furthermore, intra- and inter-patient variabilities in CT-based SPR prediction were investigated in a retrospective analysis of 102 brain-tumor and 25 prostate-cancer patients. The clinical HLUT was then refined by performing a step-wise weighted linear fit of the resulting SPR distribution in different tissue regionsusing the DirectSPR information of the investigated patient cohort. To assess the effect of this refinement on the proton range within the patient, treatment plans were recalculated using the clinical HLUT, the refined HLUT as well as the DirectSPR approach.

Results:

In the complex head geometry, DirectSPR showed an improved accuracy compared to the clinical HLUT approach. For biological tissues in a simple geometry, an accuracy below 0.2% could be achieved. Between clinical HLUT and DirectSPR, mean range differences (±1SD) of (1.2±0.7)% for brain-cancer and (1.7±0.5)% for prostate-tumor patients were determined. By refining the HLUT, they were significantly reduced (p≪0.001, two-sample t-test) below 0.3%. HoweverMoreover, an observed intra-patient soft-tissue diversity of 6% as well as an inter-patient bone diversity of 5%, underline an additional benefit of the DirectSPR approach, as such variabilities cannot be considered by any generic HLUT-based range prediction.

Conclusions:

The clinical feasibility of DirectSPR for proton range prediction as well as its advantage over the HLUT approach has been demonstrated. A retrospective application on patient data allowed for a reduction of systematic deviations found in clinical HLUT. The refined HLUT was implemented at our institution as a step towards the currently ongoing full integration of DirectSPR. The Its higher precision accuracy in range prediction is reflected in a potentially allows a reduction of the safety margin, which is currently under investigation.