Toward MR‐only proton therapy planning for pediatric brain tumors: Synthesis of relative proton stopping power images with multiple sequence MRI and development of an online quality assurance tool

Abstract

Purpose To generate synthetic relative proton stopping power (sRPSP) images from magnetic resonance imaging (MRI) sequence(s) and develop an online quality assurance (QA) tool for sRPSP to facilitate safe integration of magnetic resonance (MR)‐only proton planning into clinical practice. Materials and methods Planning computed tomography (CT) and MR images of 195 pediatric brain tumor patients were utilized (training: 150, testing: 45). Seventeen consistent‐cycle generative adversarial network (ccGAN) models were trained separately using paired CT‐converted RPSP and MRI datasets to transform a subject's MRI into sRPSP. T1‐weighted (T1W), T2‐weighted (T2W), and FLAIR MRI were permutated to form 17 combinations, with or without preprocessing, for determining the optimal training sequence(s). For evaluation, sRPSP images were converted to synthetic CT (sCT) and compared to the real CT in terms of mean absolute error (MAE) in Hounsfield units (HU). For QA, sCT was deformed and compared to a reference template built from training dataset to produce a flag map, highlighting pixels that deviate by >100 HU and fall outside the mean ± standard deviation reference intensity. The gamma intensity analysis (10%/3 mm) of the deformed sCT against the QA template on the intensity difference was investigated as a surrogate of sCT accuracy. Results The sRPSP images generated from a single T1W or T2W sequence outperformed that generated from multi‐MRI sequences in terms of MAE (all p < 0.05). Preprocessing with N4 bias and histogram matching reduced MAE of T2W MRI‐based sCT (54 ± 21 HU vs. 42 ± 13 HU, p = 0.002). The gamma intensity analysis of sCT against the QA template was highly correlated with the MAE of sCT against the real CT in the testing cohort (r = ‐0.89 for T1W sCT; r = ‐0.93 for T2W sCT). Conclusion Accurate sRPSP images can be generated from T1W/T2W MRI for proton planning. A QA tool highlights regions of inaccuracy, flagging problematic cases unsuitable for clinical use.

Validation of an MR-based multimodal method for molecular composition and proton stopping power ratio determination using ex vivo animal tissues and tissue-mimicking phantoms

“…A potential method to mitigate the error due to registration would be the creation of an MR-only method. Current MR-only methods generally rely on creating a synthetic CT by converting MR images to CTlike data (Maspero et al 2017, Aramburu Nunez et al 2020, Maspero et al 2020, Zimmermann et al 2021, Wang et al 2022. While these methods are growing increasingly more realistic and accurate, they generally rely on translating MR intensities to single-energy kVCT HU followed by stoichiometric conversion of synthetic CT HU to SPR.…”

Deep learning in MRI‐guided radiation therapy: A systematic review

“…Their method achieves an excellent MAE of 42 ± 13 HU, but struggles with dosimetric accuracy. 6 Maspero et al achieve a 2%/2 mm gamma pass rate above 99% for proton therapy by averaging predictions from three separate GANs trained on axial,sagittal,and coronal views,respectively. 116 Replacing traditional MC dose calculation methods, Tsekas et algenerate VMAT (volumetric modulated arc therapy) dose distributions in static positions with sCT.…”

“…create the first synthetic relative proton stopping power maps from MRI with a cycleGAN and loss function to take advantage of paired data. Their method achieves an excellent MAE of 42 ± 13 HU, but struggles with dosimetric accuracy 6 . Maspero et al.…”

“…Their method achieves an excellent MAE of 42 ± 13 HU, but struggles with dosimetric accuracy. 6 Maspero et al. achieve a 2%/2 mm gamma pass rate above 99% for proton therapy by averaging predictions from three separate GANs trained on axial, sagittal, and coronal views, respectively.…”

“…In an MRI‐only workflow, sCT avoids registration errors and the radiation exposure associated with traditional CT. 5 In addition, synthetic relative proton stopping power (sRPSP) maps can be generated to directly obtain dosimetric information for proton radiation therapy. 6 The dosimetric uncertainty can be further enhanced with deep learning dose calculation methods, which greatly reduce inference time and could yield lower dosimetric uncertainties compared to traditional Monte Carlo (MC) methods. Synthetic MRI (sMRI), generated from CT, is appealing by combining the speed and dosimetric information of CT with MRI's high soft tissue contrast.…”