Deep Convolution Neural Network (DCNN) multi-plane approach to synthetic CT generation from MR images - application in brain proton therapy.

BACKGROUND AND PURPOSE The first aim of this work is to present a novel Deep Convolution Neural Network (DCNN) multi-plane approach and compare it to single plane prediction of syntheticCT (sCT) by using the real CT as ground truth (GT). The second aim is to demonstrate the feasibility of MRI based proton therapy planning for brain, by assessing the range shift error within the clinical acceptance threshold. MATERIALS AND METHODS The image database included 15 MRI/CT pairs of the head. Three DCNNs were trained to estimate, for each voxel, the Hounsfield Unit (HU) value from MRI intensities. Each DCNN gave an estimation in the axial, sagittal and coronal plane respectively. Then, the median HU among the three values was selected to build the sCT. The sCT/CT agreement was evaluated by a Mean Absolute Error (MAE) and mean error (ME), computed within the head contour and on six different tissues. Dice Similarity Coefficients (DSC) were calculated to assess the geometric overlap of bone and air cavities segmentations. A 3-beam proton therapy plan was simulated for each patient. Beam-by-beam range shift (RS) analysis was conducted to assess the proton-stopping power estimation. RS analysis was performed using clinically accepted thresholds of 1) 3.5%+1mm and 2) 2.5% +1.5mm of the total range. RESULTS DCNN multi-plane statistically over performed single plane prediction of sCT (p<0.025). MAE and ME within the head were 54±7 HU and -4±17 HU (mean±standard deviation), respectively. Soft tissues were very close to perfect agreement (11±3 HU in terms of MAE). Segmentation of air and bone regions led to a DSC of 0.92±0.03 and 0.93±0.02, respectively. Proton RS was always below clinical acceptance thresholds, with a relative RS error of 0.14±1.11%. CONCLUSIONS The multi-plane DCNN approach significantly improved the sCT prediction compared to other DCNN methods present in the literature. The method demonstrated to be highly accurate for MRI-only proton planning purposes.