| 唐丽,刘星,侯平,王明月,魏一娟,吕培杰,高剑波.深度学习重建联合Smart去金属伪影算法对颈部CT图像中口腔金属植入物伪影的影响[J].中国医学影像技术,2023,39(11):1731~1735 |
| 深度学习重建联合Smart去金属伪影算法对颈部CT图像中口腔金属植入物伪影的影响 |
| Impact of deep learning reconstruction combined with Smart metal artifact reduction algorithm on oral metallic implant artifacts on neck CT images |
| 投稿时间：2023-06-15 修订日期：2023-09-28 |
| DOI：10.13929/j.issn.1003-3289.2023.11.031 |
| 中文关键词: 口腔 伪影 深度学习 体层摄影术,X线计算机 |
| 英文关键词:mouth artifacts deep learning tomography, X-ray computed |
| 基金项目:河南省医学科技攻关计划联合共建项目(LHGJ20210327)。 |
|
| 摘要点击次数: 1637 |
| 全文下载次数: 571 |
| 中文摘要: |
| 目的 观察深度学习重建(DLR)联合Smart去金属伪影(MAR)算法对颈部CT图像中口腔金属植入物伪影的影响。方法 回顾性分析60例口腔内存在金属植入物患者的颈部CT资料,其中19例存在颈部病变且受金属伪影干扰;分别以自适应迭代重建(ASIR-V,重建百分比为50%)联合Smart MAR(IR+S组)、DLR-H(重建强度为高水平)联合Smart MAR(DH+S组)、DLR-M(重建强度为中水平)联合Smart MAR(DM+S组)、DLR-H(DH组)及DLR-M(DM组)重建静脉期图像,通过计算各组病变/舌部软组织噪声(SD1)、头夹肌噪声(SD2)及伪影指数(AI)对图像进行客观评估;以Likert量表对图像整体及显示病灶质量进行主观评分;比较各组主、客观评估结果的差异。结果 5组图像SD1、SD2及AI差异均有统计学意义(P均＜0.05)。SD1及AI在DH+S组、DM+S组、IR+S组、DH组及DM组依次升高(P均＜0.05);SD2在DH+S组、DM+S组及IR+S组依次升高(P均＜0.05),而在DH+S组与DH组、DM+S组与DM组均无统计学差异(P均＞0.05)。存在颈部病变的6例(6/19,31.58%)及无颈部病变的4例(4/41,9.76%)可于IR+S组、DH+S组及DM+S组中发现DH组及DM组中不存在的伪影。5组图像整体及显示病灶质量评分差异均有统计学意义(P均＜0.05),在DH+S组、DM+S组及IR+S组依次降低(P均＜0.05),但均高于DH组及DM组(P均＜0.05),而DM组与DH组间差异均无统计学意义(P均＞0.05)。结论 以DLR联合Smart MAR算法重建口腔金属植入物患者颈部CT图像的噪声、AI、图像整体质量及显示病灶质量均较好,但存在无法去除金属伪影的可能。 |
| 英文摘要: |
| Objective To observe the impact of deep learning reconstruction (DLR) combined with Smart metal artifact reduction (MAR) algorithm on oral metallic implants artifacts on neck CT images. Methods Neck CT images of 60 patients with oral metallic implants were retrospectively analyzed.Among them, 19 cases were found with cervical lesions interfered by metal artifacts. Venous phase CT images were reconstructed by using adaptive statistical iterative reconstruction V (ASIR-V, reconstruction percentage was 50%) combined with Smart MAR (IR+S group), DLR-H (high-resolution) combined with Smart MAR (DH+S group), DLR-M (medium-resolution) combined with Smart MAR (DM+S group), DLR-H (DH group) and DLR-M (DM group), respectively.The noise of lesion/tongue tissue (SD1), noise of head clamp muscle (SD2) and artifact index (AI) of each group were calculated to evaluate the images objectively. Likert scale was performed to evaluate the overall image quality and lesion demonstration. Both subjective and objective results were compared among groups. Results Significant differences of SD1, SD2 and AI were found among 5 groups (all P＜0.05). SD1 and AI increased gradually in DH+S group, DM+S group, IR+S group, DH group and DM group (all P＜0.05). SD2 increased gradually in DH+S group, DM+S group and IR+S group (all P＜0.05), while no significant difference of SD2 was found between DH+S group and DH group, nor between DM+S group and DM group (all P＞0.05). Artifacts of 6 cases (6/19, 31.58%) with and 4 cases (4/41, 9.76%) without cervical lesions were observed in IR+S group, DH+S group and DM+S group, which didn't exist in DH group and DM group. Significant differences of overall image quality and lesion demonstration were found among 5 groups (both P＜0.05). The overall image quality and lesion demonstration decreased gradually in DH+S, DM+S and IR+S groups (all P＜0.05), but all higher than those in DH group and DM group, respectively (all P＜0.05). No significant difference of overall image quality nor lesion demonstration was found between DM group and DH group (both P＞0.05). Conclusion Noise, AI, overall image quality and lesion demonstration of neck CT images reconstructed using DLR combined with Smart MAR algorithm were all good, but there was a possibility that metal artifact couldn't be removed. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
