| 王明月,吴艳,周悦,等.深度学习重建算法结合低剂量CT在机会性骨质疏松症筛查中的应用研究[J].中华放射医学与防护杂志,2023,43(11):923-928.Wang Mingyue,Wu Yan,Zhou Yue,et al.Application of deep learning reconstruction algorithm combined with low-dose CT for screening opportunistic osteoporosis[J].Chin J Radiol Med Prot,2023,43(11):923-928 |
| 深度学习重建算法结合低剂量CT在机会性骨质疏松症筛查中的应用研究 |
| Application of deep learning reconstruction algorithm combined with low-dose CT for screening opportunistic osteoporosis |
| 投稿时间：2023-05-06 |
| DOI：10.3760/cma.j.cn112271-20230506-00224 |
| 中文关键词: 定量CT 骨密度 深度学习 图像质量 |
| 英文关键词:Quantitative CT Bone mineral density Deep learning Image quality |
| 基金项目:河南省高端外国专家引进计划项目(HNGD2022033) |
|
| 摘要点击次数: 1887 |
| 全文下载次数: 921 |
| 中文摘要: |
| 目的 将深度学习重建算法与低剂量CT相结合,探究其对图像质量的影响及对骨密度测量的影响,及其在机会性骨质疏松筛查中的应用价值。方法 前瞻性收集同时接受胸上腹部联合低剂量扫描的患者119例(年龄≥40岁)。所有图像分别使用滤波反投影(FBP)算法、基于混合模型的自适应统计迭代重建(ASIR-V)50%和3个水平深度学习算法进行重建。使用非同步定量CT软件进行骨密度测量,比较不同重建条件下的骨密度(BMD)。分别计算降主动脉、肝脏、脾脏的信噪比(SNR)和对比噪声比(CNR),将前腹壁脂肪的标准差代表图像的噪声,并使用5分制主观评价法,进行图像质量客观评价。比较不同重建方法下,不同部位的客观和主观图像质量。结果 在不同的重建方法下,BMD的差异无统计学意义(P>0.05)。高级别的深度学习重建算法(DLIR-H)较ASIR-V 50%在降主动脉、肝脏和脾脏的SNR分别提高了103.88%、125.09%、136.13%,图像噪声降低了55.98%,DLIR-H的CNR和主观评分(肺部病变显示能力除外)均优于DLIR-L和ASIR-V 50%(χ2=158.31~275.35,P＜0.001)。结论 深度学习算法不影响骨密度测量的准确性,图像质量优于ASIR-V 50%。深度学习算法联合低剂量CT可用于机会性骨质疏松筛查。 |
| 英文摘要: |
| Objective To explore the influence of deep learning reconstruction algorithm combined with low-dose CT on image quality and bone mineral density measurement and the application value in opportunistic osteoporosis screening.Methods A total of 119 patients (aged ≥40 years) who underwent a combined chest and upper abdominal low-dose scan were prospectively included. All the images were reconstructed using filtered back projection(FBP) alogrithm, hybrid model-based adaptive statistical iterative reconstruction (ASIR-V) 50% and three levels of deep learning reconstruction algorithm respectively. Bone mineral density (BMD) values for different reconstruction conditions were measured and compared using asynchronous quantitative CT software. The signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of descending aorta, liver and spleen were calculated, and the image noise was the standard deviation of anterior abdominal wall fat and the image quality was objectively evaluated by using the five-point subjective evaluation method. The objective and subjective image quality of different body parts with different reconstruction method was compared.Results There was no statistical difference in BMD with different reconstruction method (P > 0.05). Compared with ASIR-V 50%, the SNRs of high level deep learning image reconstruction (DLIR-H)in descending aorta, latissimus dorsi, liver and spleen were increased by 103.88%, 125.09% and 136.13% respectively, and the image noise was decreased by 55.98%. Both the CNR and subjective scores (except the ability to display lung lesions) of DLIR-H were better than those of DLIR-L and ASIR-V 50% (χ2 =158.31-275.35, P<0.001).Conclusions The deep learning algorithm does not affect the accuracy of bone mineral density measurement, and the image quality is better than that of ASIR-V 5%. Deep learning algorithm combined with low-dose CT can be used for opportunistic osteoporosis screening. |
| HTML 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|