ESTRO 2020 Abstract book

S980 ESTRO 2020

(figure 1, right) show an apparent noise reduction and increase in contrast-to-noise ratio with the MAP method.

Figure 2 shows that, in the presence of strong biases (beam hardening on the left panel, and metal artifacts on the right panel), the MAP approach can provide robust estimates of physical parameters.

Conclusion In this pilot study, serial DWI at the MRL showed very good agreement with measurements at a 3 T reference scanner assessed by Bland-Altman analysis. Further patients will be analyzed. These initial data represent an important prerequisite for real-time response adaptive radiotherapy on the basis of anatomical and functional MRI. PO-1686 Electron density and effective atomic number estimation in a maximum a posteriori framework for DECT M. Simard 1 , E. Bär 2 , H. Bouchard 1 1 Université de Montréal, Department of Physics, Montreal, Canada ; 2 University College London, Department of Medical Physics and Biomedical Engineering, London, United Kingdom Purpose or Objective Dual-energy CT (DECT) can accurately estimate physical parameters of tissues, namely the electron density (ED) and effective atomic number (Z med ). In turn, this provides accurate evaluation of radiotherapy-related parameters for radiotherapy treatment planning. Notably, for proton therapy, the proton stopping power (SPR) can be more accurately inferred with DECT than single energy CT, although the noisier images do not allow DECT to reach its full potential to enhance proton therapy. In this work, we propose a DECT tissue characterization methodology that provides robust-to-noise estimates of ED, Z med and SPR. Material and Methods The stoichiometric DECT calibration method (Bourque et al. , Phys. Med. Biol . 59(8) 2014, referred to as least squares or LSQ) is adapted in a maximum a posteriori (MAP) framework. The approach maximizes a likelihood term while ensuring that obtained ED and Z med values lie close to those expected for human tissues, using a kernel density estimator. LSQ and MAP are first compared for accuracy and precision in simulated images with a known ground truth. Qualitative and quantitative performance is then assessed on clinical datasets - prostate and head & neck patients acquired on a Siemens SOMATOM definition Flash dual-source CT with a 100/140Sn kVp couple. Results Simulation results (figure 1, left) show relative error distributions for ED, Z med and SPR. Results illustrate that both methods are generally unbiased on human tissues (<0.4% absolute mean error for ED, Z med and SPR), while the MAP method provides large reduction in the root- mean-square error, which is reduced from 2.8 to 1.7% for SPR. Parameter maps calculated on clinical datasets

Table 1 shows mean values and standard deviation of ED, Z med and SPR in 3 ROIs of the clinical data. This shows that MAP keeps the quantitative accuracy (<1% variation in mean ED and SPR with respect to LSQ), while providing a strong average noise reduction factor of 2.5 for SPR (calculated as the ratio of standard deviations of the methods).

LSQ

LSQ Z med

LSQ

MAP

MAP Z med

MAP

ED

SPR

ED

SPR

7.47 ± 0.96 7.24 ± 1.22 5.55 ± 1.50

1.044 ± 0.022 1.066 ± 0.038 0.947 ± 0.052

1.035 ± 0.007 1.057 ± 0.012 0.902 ± 0.008

7.35 ± 0.36 7.27 ± 0.58 6.03 ± 0.33

1.040 ± 0.011 1.063 ± 0.019 0.929 ± 0.013

Brain 1.037 ± 0.011 Muscle 1.052 ± 0.017 Adipose 0.893 ± 0.013

Conclusion The MAP framework provides accurate and precise estimates of SPR and physical parameters, and strongly reduces the impact of noise as well as artifacts. Using a MAP method will be useful towards more accurate proton therapy treatment planning with DECT. The robustness to noise of the approach can also be used towards dose reduction in DECT scans and for general quantitative imaging with ED and Z med maps.