| 张玉荣,苑倩倩,高剑波,刘杰.深度学习重建算法在腹部CT成像中应用的体模研究[J].中华放射医学与防护杂志,2023,43(8):645-652 |
| 深度学习重建算法在腹部CT成像中应用的体模研究 |
| Application value of deep learning reconstruction algorithm in CT imaging of abdominal phantoms |
| 投稿时间：2022-12-26 |
| DOI：10.3760/cma.j.cn112271-20221226-00498 |
| 中文关键词: 辐射剂量 体层摄影术,X射线计算机 腹部 体模 |
| 英文关键词:Radiation dosage Tomography, X-Ray Computer Abdomen Phantom |
| 基金项目:河南省卫生健康委员会科技攻关项目（212102310142） |
|
| 摘要点击次数: 800 |
| 全文下载次数: 302 |
| 中文摘要: |
| 目的 探讨不同辐射剂量下深度学习图像重建算法（DLIR）相对于常规迭代重建算法（ASIR-V）对腹部体模CT图像质量的改善价值。方法 根据管电压设置100 kV组与120 kV组，每组按照容积剂量指数（CTDIvol）不同（2、4、6、8、10、15 mGy）分为6组进行常规扫描，获得基于滤波反投影（FBP）算法的CT图像，并使用不同权重迭代重建算法（ASIR-V 50%、80%、100%）及不同等级深度学习重建算法（DLIR-L、DLIR-M、DLIR-H）进行图像重建，共获得84组图像。对比分析不同重建方式下各CTDIvol组图像各部位CT值、噪声、信噪比（SNR）、对比噪声比（CNR）及主观评分的变化规律。图像质量主观评分比较采用Kruskal-Wallis H检验，客观指标和辐射剂量比较采用单因素方差分析及配对样本t检验。结果 同一管电压下，各CTDIvol组不同重建条件下各部位的噪声、SNR、CNR差异均有统计学意义（F=415.39、315.30，P<0.001），且ASIR-V 50%与DLIR-L图像的噪声、SNR、CNR差异无统计学意义（P>0.05）；主观评分之间差异均有统计学意义（100 kV组：H=13.47，P=0.036；120 kV组：H=12.99，P=0.043），且两名医师的主观评分一致性较高（Kappa>0.70），其中DLIR-H图像质量评分最高，DLIR-M与ASIR-V 50%图像质量主观评分基本一致；100 kV组图像质量主观评分整体较120 kV略高。以CTDIvol为15 mGy组ASIR-V 50%图像作为参照，在满足诊断需求的前提下，低中高等级的DLIR可以分别降低辐射剂量超过30%、70%、85%。结论 DLIR算法不仅能够显著降低图像噪声、提高图像质量，而且可以在满足诊断需求的前提下有效降低辐射剂量；推荐临床应用100 kV结合中、高等级DLIR行腹部低剂量CT扫描。 |
| 英文摘要: |
| Objective To explore the value of the deep learning image reconstruction (DLIR) algorithm in improving the CT image quality of abdominal phantoms under different radiation doses by comparing the DLIR algorithm with the conventional Adaptive Statistical Iterative Reconstruction-V (ASIR-V) technique.Methods Two groups with tube voltages of 100 kV and 120 kV (also referred to as the 100 kV and 120 kV groups, respectively) were involved. Each group was further divided into six subgroups based on different volumetric CT dose indices (CTDIvol:2, 4, 6, 8, 10 and 15 mGy). Subsequently, CT images based on the filtered back projection (FBP) algorithm were obtained and were then reconstructed using the ASIR-V algorithm with different weights (ASIR-V 50%, 80%, and 100%) and the DLIR algorithm with different levels (DLIR-L, M, and -H). As a result, 84 groups of images were obtained in total. Afterward, this study compared and analyzed the variations in CT values, noise, signal-to-noise ratios (SNRs), contrast-to-noise ratios (CNRs), and subjective scores of various parts in various CTDIvol subgroups under different reconstruction conditions. In addition, the subjective scores of the image quality were compared using the Kruskal-Wallis H test, while objective indices and radiation doses were compared through the univariate analysis of variance (ANOVA) and the paired t test.Results Under the same tube voltage, there were statistically significant differences in the noise, SNRs, and CNRs of various parts in various CTDIvol subgroups under different reconstruction conditions (F=415.39, 315.30, P < 0.001), while there was no statistically significant difference in the noise, SNRs, and CNRs of images constructed using ASIR-V 50% and DLIR-L (P > 0.05). Under different tube voltages, the subjective scores of both groups show statistically significant differences (100 kV group: H=13.47, P =0.036; 120 kV group:H=12.99, P=0.043). Moreover, two physicians offered consistent subjective scores, with Kappa values > 0.70. Among these images, DLIR-H images showed the highest subjective scores, followed by DLIR-M and ASIR-V 50% images, which had roughly consistent subjective scores. Moreover, the subjective scores of the 100 kV group were slightly higher than those of the 120 kV group. With the ASIR-V 50% images of the subgroup with a CTDIvol of 15 mGy as references, the DLIR-L, -M, and -H reduced radiation doses by more than 30%, 70% and 85%, respectively on the premise that diagnostic requirements were met.Conclusions The DLIR algorithm can not only significantly reduce the image noise and improve the image quality, but also effectively decrease the radiation doses on the premise of meeting the diagnostic requirements. It is recommended that 100 kV tube voltage combined with a medium- or high-level DLIR algorithm should be applied to low-dose abdominal CT scans in clinical applications. |
| HTML 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|