[论文解读] Deep learning based CT-to-CBCT deformable image registration for autosegmentation in head and neck adaptive radiation therapy
本研究提出了一种深度学习框架,可联合生成合成CT(sCT)并执行从CBCT到pCT的可变形图像配准,以实现在头颈部自适应放疗中的自动分割。通过结合CycleGAN与5级级联Voxelmorph网络,该方法实现了高精度的轮廓传播,平均Dice分数达0.83,95% Hausdorff距离为2.01 mm,其精度与效率均优于标准方法。
The purpose of this study is to develop a deep learning based method that can automatically generate segmentations on cone-beam CT (CBCT) for head and neck online adaptive radiation therapy (ART), where expert-drawn contours in planning CT (pCT) can serve as prior knowledge. Due to lots of artifacts and truncations on CBCT, we propose to utilize a learning based deformable image registration method and contour propagation to get updated contours on CBCT. Our method takes CBCT and pCT as inputs, and output deformation vector field and synthetic CT (sCT) at the same time by jointly training a CycleGAN model and 5-cascaded Voxelmorph model together.The CycleGAN serves to generate sCT from CBCT, while the 5-cascaded Voxelmorph serves to warp pCT to sCT's anatommy. The segmentation results were compared to Elastix, Voxelmorph and 5-cascaded Voxelmorph on 18 structures including left brachial plexus, right brachial plexus, brainstem, oral cavity, middle pharyngeal constrictor, superior pharyngeal constrictor, inferior pharyngeal constrictor, esophagus, nodal gross tumor volume, larynx, mandible, left masseter, right masseter, left parotid gland, right parotid gland, left submandibular gland, right submandibular gland, and spinal cord. Results show that our proposed method can achieve average Dice similarity coefficients and 95% Hausdorff distance of 0.83 and 2.01mm. As compared to other methods, our method has shown better accuracy to Voxelmorph and 5-cascaded Voxelmorph, and comparable accuracy to Elastix but much higher efficiency. The proposed method can rapidly and simultaneously generate sCT with correct CT numbers and propagate contours from pCT to CBCT for online ART re-planning.
研究动机与目标
- 为解决在线自适应放疗(ART)中CBCT图像上人工勾画轮廓不准确且耗时的问题。
- 开发一种快速且准确的方法,将专家绘制的计划CT(pCT)轮廓传播至存在伪影和截断的锥形束CT(CBCT)。
- 利用深度学习从CBCT生成具有准确Hounsfield单位值的合成CT(sCT)。
- 实现实时、自动化的轮廓勾画,以支持头颈部癌放疗中的在线再计划。
提出的方法
- 训练CycleGAN将CBCT图像转换为具有准确Hounsfield单位值的合成CT(sCT)。
- 采用5级级联Voxelmorph网络执行从pCT到生成sCT的可变形图像配准。
- 联合训练CycleGAN与Voxelmorph网络,以确保sCT生成与解剖结构形变之间的一致性。
- 通过将Voxelmorph获得的形变场应用于原始pCT轮廓,实现轮廓传播。
- 该框架利用端到端学习,同时提升配准精度与sCT质量。
- 采用Dice与95% Hausdorff距离指标,在18个头颈部结构上对方法进行评估。
实验结果
研究问题
- RQ1基于深度学习的方法能否联合生成高精度的合成CT并实现从CBCT到pCT的可靠可变形配准?
- RQ2与Elastix和标准Voxelmorph等传统方法相比,所提出方法在分割精度与计算效率方面表现如何?
- RQ3CycleGAN与级联Voxelmorph的联合训练是否能提升在存在伪影与截断的CBCT图像上的轮廓传播性能?
- RQ4生成的sCT中是否能保持准确的Hounsfield单位值以满足临床使用需求?
- RQ5该框架能否实现头颈部放疗中快速、自动化的轮廓勾画,以支持在线自适应再计划?
主要发现
- 所提出方法在18个头颈部结构上的平均Dice相似系数达到0.83,表明分割精度较高。
- 95% Hausdorff距离为2.01 mm,表明尽管存在图像伪影,边界对齐仍非常精确。
- 在Dice与Hausdorff距离指标上,该方法优于标准Voxelmorph与5级级联Voxelmorph。
- 其精度与Elastix相当,但计算效率显著更高。
- CycleGAN与级联Voxelmorph的联合训练提升了sCT生成与轮廓传播的一致性与质量。
- 该方法实现了快速、自动化且临床可行的轮廓传播,适用于在线自适应放疗。
更好的研究,从现在开始
从阅读论文到最终审阅,大幅缩短您的研究时间。
无需绑定信用卡
本解读由 AI 生成,并经人工编辑审核。