Physics Research Laboratory, Department of Radiology and Biomedical Imaging, University of California San Francisco, San Francisco, CA, United States of America. UCSF Physics Research Laboratory, 185 Berry Street, Suite 350, San Francisco, CA 94143-0946, United States of America. Author to whom any correspondence should be addressed.
Phys Med Biol. 2019 Apr 4;64(7):075019. doi: 10.1088/1361-6560/ab0606.
Dedicated brain positron emission tomography (PET) devices can provide higher-resolution images with much lower doses compared to conventional whole-body PET systems, which is important to support PET neuroimaging and particularly useful for the diagnosis of neurodegenerative diseases. However, when a dedicated brain PET scanner does not come with a combined CT or transmission source, there is no direct solution for accurate attenuation and scatter correction, both of which are critical for quantitative PET. To address this problem, we propose joint attenuation and scatter correction (ASC) in image space for non-corrected PET (PET) using deep convolutional neural networks (DCNNs). This approach is a one-step process, distinct from conventional methods that rely on generating attenuation maps first that are then applied to iterative scatter simulation in sinogram space. For training and validation, time-of-flight PET/MR scans and additional helical CTs were performed for 35 subjects (25/10 split for training and test dataset). A DCNN model was proposed and trained to convert PET to DCNN-based ASC PET (PET) directly in image space. For quantitative evaluation, uptake differences between PET and reference CT-based ASC PET (PET) were computed for 116 automated anatomical labels (AALs) across 10 test subjects (1160 regions in total). MR-based ASC PET (PET), a current clinical protocol in PET/MR imaging, was another reference for comparison. Statistical significance was assessed using a paired t test. The performance of PET was comparable to that of PET, in comparison to reference PET. The mean SUV differences (mean ± SD) from PET were 4.0% ± 15.4% (P < 0.001) and -4.2% ± 4.3% (P < 0.001) for PET and PET, respectively. The overall larger variation of PET (15.4%) was prone to the subject with the highest mean difference (48.5% ± 10.4%). The mean difference of PET excluding the subject was substantially improved to -0.8% ± 5.2% (P < 0.001), which was lower than that of PET (-5.07% ± 3.60%, P < 0.001). In conclusion, we demonstrated the feasibility of directly producing PET images corrected for attenuation and scatter using a DCNN (PET) from PET in image space without requiring conventional attenuation map generation and time-consuming scatter correction. Additionally, our DCNN-based method provides a possible alternative to MR-ASC for simultaneous PET/MRI.
利用深度卷积神经网络（DCNN）在图像空间中对未经校正的 PET（PET）进行联合衰减和散射校正（ASC），提出一种新方法，用于解决专用脑 PET 扫描仪在没有组合 CT 或透射源时无法进行准确衰减和散射校正的问题，这对于定量 PET 至关重要。该方法是一种一步法，与传统方法不同，传统方法先生成衰减图，然后将其应用于正弦图空间中的迭代散射模拟。为了训练和验证，对 35 名受试者进行了时间飞行 PET/MR 扫描和额外的螺旋 CT（25/10 用于训练和测试数据集）。提出并训练了一个 DCNN 模型，以便直接在图像空间中从 PET 转换为基于 DCNN 的 ASC PET（PET）。为了进行定量评估，在 10 名测试对象的 116 个自动解剖标签（AAL）上计算了 1160 个区域的 PET 与基于参考 CT 的 ASC PET（PET）之间的摄取差异。基于 MR 的 ASC PET（PET）是 PET/MR 成像中的一种当前临床方案，也是另一种比较参考。使用配对 t 检验评估统计学意义。与参考 PET 相比，PET 的性能与 PET 相当。与 PET 相比，PET 的 SUV 差异平均值（均值±标准差）分别为 4.0%±15.4%（P<0.001）和-4.2%±4.3%（P<0.001）。PET 的整体较大变化（15.4%）容易受到具有最高平均差异的受试者的影响（48.5%±10.4%）。排除该受试者后，PET 的平均差异得到了很大改善，达到-0.8%±5.2%（P<0.001），低于 PET（-5.07%±3.60%，P<0.001）。总之，我们证明了在图像空间中，利用 DCNN（PET）从 PET 直接生成衰减和散射校正的 PET 图像是可行的，而无需生成传统的衰减图和耗时的散射校正。此外，我们基于 DCNN 的方法为同时进行的 PET/MRI 提供了一种替代基于 MR-ASC 的方法。
