| 王源,杨雪倩,郝志鑫,等.基于深度渐进学习重建和有序子集期望最大化重建算法的肝脏与胰腺18 F-FDG PET/MR图像质量比较研究[J].中华放射医学与防护杂志,2026,46(8):795-803.Wang Yuan,Yang Xueqian,Hao Zhixin,et al.Comparing the effects of the deep progressive reconstruction and ordered subsets expectation maximization algorithms on the quality of the 18F-FDG PET/MR images of liver and pancreas[J].Chin J Radiol Med Prot,2026,46(8):795-803 |
| 基于深度渐进学习重建和有序子集期望最大化重建算法的肝脏与胰腺18 F-FDG PET/MR图像质量比较研究 |
| Comparing the effects of the deep progressive reconstruction and ordered subsets expectation maximization algorithms on the quality of the 18F-FDG PET/MR images of liver and pancreas |
| 投稿时间：2026-02-08 |
| DOI：10.3760/cma.j.cn112271-20260208-00045 |
| 中文关键词: 18F-FDG PET/MR|深度渐进学习重建|等效低剂量|肝脏|胰腺 |
| 英文关键词:18F-FDG PET/MR|Deep progressive reconstruction|Equivalent low-dose simulation|Liver|Pancreas |
| 基金项目:国家自然科学基金(U24A20758);北京市自然科学基金(L242062);国家重点研发计划(2024YFC2419400);中国医学科学院医学与健康科技创新工程项目(CIFMS-2024-I2M-ZD-001,CIFMS-2023-I2M-2-002,CIFMS-2021-I2M-1-002,CIFMS-2021-I2M-1-003,CIFMS-2021-I2M-1-025);中央高水平医院临床科研专项项目(2022-PUMCH-D-001,2022-PUMCH-D-002) |
|
| 摘要点击次数: 1903 |
| 全文下载次数: 4 |
| 中文摘要: |
| 目的 比较深度渐进学习重建(DPR)与有序子集期望最大化(OSEM)重建算法在一体化正电子发射断层显像/磁共振(PET/MR)系统中对肝脏与胰腺18F-FDG PET图像质量的影响,并评估DPR在提升图像质量、抑制噪声及等效低剂量图像重建中的潜在价值。方法 回顾性纳入23例接受腹部18F-FDG PET/MR检查的患者。PET图像分别采用标准OSEM算法与不同强度(1~3)的DPR算法进行重建,并通过缩短采集时间模拟等效1/2、1/3及1/4的低剂量条件。图像质量评估包括定性与定量分析。定性评估由两位经验丰富的核医学科医师在双盲条件下采用5分制Likert量表进行评分;定量分析则在肝脏与胰腺区域勾画感兴趣体积(VOI),计算信噪比(SNR)和噪声指标。结果 主观评价显示,DPR各强度组在整体图像质量、图像清晰度、伪影水平、噪声水平和肝脏均匀性方面的评分均显著高于OSEM组,且两位评估者间一致性良好。定量分析显示,与OSEM相比,DPR重建图像在肝脏与胰腺VOI内具有更高的SNR和更低的噪声水平。其中,DPR强度1在有效抑制噪声的同时较好地保留了组织结构细节,表现出较优的图像质量与噪声控制平衡。在等效低剂量模拟实验中,采集时间缩短至1/2的DPR重建图像,即等效1/2低剂量条件下的DPR图像,其图像质量仍显著优于全时间OSEM重建图像。结论 与传统OSEM重建算法相比,DPR可显著改善肝脏与胰腺18F-FDG PET/MR图像质量,并在缩短采集时间或等效低剂量条件下保持良好的成像稳定性。其中,DPR强度1综合表现较优,具有作为腹部PET/MR临床标准化重建参数的潜在应用价值。 |
| 英文摘要: |
| Objective To compare the effects of deep progressive reconstruction (DPR) and ordered subsets expectation maximization (OSEM) algorithms on the quality of 18F-fluorodeoxyglucose positron emission tomography (18F-FDG PET) images of liver and pancreas acquired using an integrated positron emission tomography/magnetic resonance (PET/MR) imaging system, and to assess the potential value of DPR in improving image quality, suppressing noise, and reconstructing equivalent low-dose PET scans. Methods A retrospective study was conducted on 23 patients who underwent abdominal 18F-FDG PET/MRI scans. The resulting PET images were reconstructed using the standard OSEM algorithm and the DPR algorithms of levels 1-3. Then, equivalent low-dose conditions corresponding to 1/2, 1/3, and 1/4 of the standard dose were simulated by reducing acquisition time. The PET images were subjected to qualitative and quantitative quality evaluations. Regarding qualitative evaluation, the images were scored using the 5-point Likert scale under double-blind conditions by two experienced nuclear medicine physicians independently. For quantitative evaluation, volumes of interest (VOIs) were delineated in the liver and pancreas zones, and their signal-to-noise ratio (SNR) and noise index were calculated. Results Qualitative (subjective) evaluation indicated that compared to those reconstructed using OSEM, images reconstructed using the DPR algorithms at all levels exhibited significantly higher scores in terms of overall image quality, image sharpness, artifact suppression, noise suppression, and liver uniformity, with high inter-observer consistency. Quantitative evaluation revealed that compared to OSEM-reconstructed images, DPR-reconstructed images showed higher SNR and lower noise in the liver and pancreas VOIs. Among the DPR algorithm of three levels, level-1 DPR better preserved the details of tissues and structures in images while effectively suppressing noise, achieving a good balance between image quality and noise control. The equivalent low-dose simulation experiments revealed that DPR-reconstructed images with acquisition time reduced by 1/2, equivalent to images acquired under half doses, still showed significantly higher quality than images reconstructed using OSEM under full acquisition time. Conclusions Compared to the conventional OSEM algorithm, DPR significantly improves the quality of 18F-FDG PET/MR images of the liver and pancreas and maintains high imaging stability under reduced acquisition time or equivalent low-dose conditions. Among the DPR algorithms at all levels, the level-1 algorithm delivers the optimal overall performance, holding potential value in acting as a parameter for the clinical standardized reconstruction of abdominal PET/MR images. |
| HTML 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|