Improving generalization in MR‐to‐CT synthesis in radiotherapy by using an augmented cycle generative adversarial network with unpaired data

Abstract

Purpose MR‐to‐CT synthesis is one of the first steps in the establishment of an MRI‐only workflow in radiotherapy. Current MR‐to‐CT synthesis methods in deep learning use unpaired MR and CT training images with a cycle generative adversarial network (CycleGAN) to minimize the effect of misalignment between paired images. However, this approach critically assumes that the underlying interdomain mapping is approximately deterministic and one‐to‐one. In the current study, we use an Augmented CycleGAN (AugCGAN) model to create a robust model that can be applied to different scanners and sequences using unpaired data. Materials and methods This study included T2‐weighted MR and CT pelvic images of 38 patients in treatment position from five different centers. The AugCGAN was trained on 2D transverse slices of 19 patients from three different sites. The network was then used to generate synthetic CT (sCT) images of 19 patients from the two other sites. Mean absolute errors (MAEs) for each patient were evaluated between real and synthetic CT images. Original treatment plans of nine patients were recalculated using sCT images to assess the dose distribution in terms of voxel‐wise dose difference, gamma, and dose–volume histogram analysis. Results The mean MAEs were 59.8 Hounsfield units (italicHU) and 65.8 HU for the first and second test sites, respectively. The maximum dose difference to the target was 1.2% with a gamma pass rate using the 3%, 3 mm criteria above 99%. The average time required to generate a complete sCT image for a patient on our GPU was 8.5 s. Conclusion This study suggests that our unpaired approach achieves good performance in generalization with respect to sCT image generation.

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

“…Efforts have been made to derive synthetic CT (sCT) in lieu of a real CT (rCT)for MR‐only proton therapy planning 3–9 . Recent breakthroughs largely stem from deep learning, specifically generative adversarial networks (GANs), 9–15 which expeditiously transform adult MRI to CT and vice versa. An MR‐only proton therapy planning workflow would not only eliminate the cost, time, and radiation exposure associated with CT acquisition, but also eliminate uncertainties involved in MR‐to‐CT registration, and enable time‐efficient adaptive proton therapy 15,16 …”

Comparison of compressed sensing and controlled aliasing in parallel imaging acceleration for 3D magnetic resonance imaging for radiotherapy preparation

“…This confirms in more detail the results of another CS study with two raters [33] . Radiomics or conversion to pseudo-CT images [34] , [35] from CS accelerated MRI will also require quality control as the texture is different.…”

Synthetic computed tomography generation from multi-sequence magnetic resonance images for nasopharyngeal carcinoma treatment planning via cycle-consistent generative adversarial network

“…Previous research on cycleGAN networks has primarily used a single MR sequence image as the model input, such as T1WI-MR or T2WI-MR images [19][20][21], but no study has determined which MRI sequence is more suitable for generating SCT images using the cycleGAN model. Qi et al [22] revealed that the models using multiple MRI sequences simultaneously for SCT conversion provided better accuracy in the obtained SCT images than those of the single-sequence MRI models.…”