Zhou Leyuan, Ni Xinye, Kong Yan, Zeng Haibin, Xu Muchen, Zhou Juying, Wang Qingxin, Liu Cong
Department of Radiation Oncology, Dushu Lake Hospital Affiliated to Soochow University, Suzhou, People's Republic of China.
Department of Radiation Oncology, Affiliated Hospital of Jiangnan University, Wuxi, People's Republic of China.
Phys Med Biol. 2023 Dec 12;68(24). doi: 10.1088/1361-6560/ad0ddc.
Deep learning has shown promise in generating synthetic CT (sCT) from magnetic resonance imaging (MRI). However, the misalignment between MRIs and CTs has not been adequately addressed, leading to reduced prediction accuracy and potential harm to patients due to the generative adversarial network (GAN)hallucination phenomenon. This work proposes a novel approach to mitigate misalignment and improve sCT generation.Our approach has two stages: iterative refinement and knowledge distillation. First, we iteratively refine registration and synthesis by leveraging their complementary nature. In each iteration, we register CT to the sCT from the previous iteration, generating a more aligned deformed CT (dCT). We train a new model on the refined 〈dCT, MRI〉 pairs to enhance synthesis. Second, we distill knowledge by creating a target CT (tCT) that combines sCT and dCT images from the previous iterations. This further improves alignment beyond the individual sCT and dCT images. We train a new model with the 〈tCT, MRI〉 pairs to transfer insights from multiple models into this final knowledgeable model.Our method outperformed conditional GANs on 48 head and neck cancer patients. It reduced hallucinations and improved accuracy in geometry (3% ↑ Dice), intensity (16.7% ↓ MAE), and dosimetry (1% ↑). It also achieved <1% relative dose difference for specific dose volume histogram points.This pioneering approach for addressing misalignment shows promising performance in MRI-to-CT synthesis for MRI-only planning. It could be applied to other modalities like cone beam computed tomography and tasks such as organ contouring.
深度学习在从磁共振成像（MRI）生成合成CT（sCT）方面已显示出前景。然而，MRI与CT之间的不对准问题尚未得到充分解决，这导致预测准确性降低，并因生成对抗网络（GAN）的幻觉现象对患者造成潜在伤害。这项工作提出了一种减轻不对准并改善sCT生成的新方法。我们的方法有两个阶段：迭代细化和知识蒸馏。首先，我们利用配准和合成的互补性质进行迭代细化。在每次迭代中，我们将CT配准到上一次迭代生成的sCT上，生成更对齐的变形CT（dCT）。我们在细化后的〈dCT，MRI〉对上训练一个新模型以增强合成效果。其次，我们通过创建一个结合了来自前几次迭代的sCT和dCT图像的目标CT（tCT）来蒸馏知识。这进一步改善了单个sCT和dCT图像之外的对齐效果。我们使用〈tCT，MRI〉对训练一个新模型，将多个模型的见解转移到这个最终的知识模型中。我们的方法在48例头颈癌患者中优于条件GAN。它减少了幻觉，并提高了几何形状（骰子系数提高3%）、强度（平均绝对误差降低16.7%）和剂量测定（提高1%）方面的准确性。对于特定的剂量体积直方图点，它还实现了<1%的相对剂量差异。这种解决不对准问题的开创性方法在仅基于MRI的规划的MRI到CT合成中显示出了有前景的性能。它可以应用于其他模态，如锥束计算机断层扫描，以及器官轮廓勾画等任务。
