Pseudo-SPR Map Generation from MRI Using U-Net Architecture for Ion Beam Therapy Application

Stopping power ratio (SPR) maps are needed for dose deposition calculations and are typically estimated from single energy CT (SECT) in clinical routine. SECT-based SPR conversion leads to large variability due to the one-to-one relationship assumed by the conversion method. Dual-energy CT (DECT) involving the acquisition of two energy spectra captures both material-specific information and tissue characterization which is essential for an accurate SPR map conversion. The goal of this study is to train a U-Net architecture to generate pseudo-SPR map from MRI (Dixon) using a DECT-converted SPR map. The model performance was validated using Head & Neck cohort of 16 patients with paired MRI and SPR maps. The proposed solution achieved a mean absolute error (MAE) and peak-signal-to-noise-ratio (PSNR) of 19.41 ± 8.67 HU and 58.76 ± 2.17 dB respectively for all test cases. From observation, the sequential incorporation of different Dixon MRI images such as fat-suppressed and water-suppressed yielded an accurate pseudo-SPR map which is comparable to its corresponding target SPR map. Furthermore, bone delineation integrated as additional channel to Dixon MRI sequence demonstrated an enhanced bone identification on predicted pseudo-SPR map. As future direction, we would like to extend this approach to a clinical SPR map which will enable dosimetric analysis of clinical target volume (CTV) to be possible in treatment planning application for ion beam therapy.