曹文静,孙昊桦,赵立毅,李翔,全国涛.深度学习全模型迭代算法用于重建CT图像:体模实验[J].中国医学影像技术,2025,41(4):557~562
深度学习全模型迭代算法用于重建CT图像:体模实验
Artificial intelligence iterative reconstruction of CT images: Phantom experiment
投稿时间:2024-10-21  修订日期:2024-12-16
DOI:10.13929/j.issn.1003-3289.2025.04.011
中文关键词:  体层摄影术,X线计算机  图像处理,计算机辅助  深度学习  体模,显像术
英文关键词:tomography, X-ray computed  image processing, computer-assisted  deep learning  phantoms, imaging
基金项目:
作者单位E-mail
曹文静 上海联影医疗科技股份有限公司, 上海 201800  
孙昊桦 上海联影医疗科技股份有限公司, 上海 201800  
赵立毅 上海联影医疗科技股份有限公司, 上海 201800  
李翔 上海联影医疗科技股份有限公司, 上海 201800  
全国涛 上海联影医疗科技股份有限公司, 上海 201800 guotao.quan@united-imaging.com 
摘要点击次数: 333
全文下载次数: 105
中文摘要:
      目的 利用体模评估深度学习全模型迭代算法(AIIR)用于重建CT图像的价值。方法 以模型迭代重建(MBIR)结合深度学习(DL)生成AIIR,对CCT MITA IQ体模、CT ACR 464体模、Catphan 700体模、光盘叠加体模及CT PBU-60全身体模进行CT扫描;分别采用滤波反投影法(FBP)、KARL 3D迭代重建等常规算法及AIIR重建CT图像,比较各种重建图像的噪声、X线剂量水平,以及低对比度分辨力、高对比度空间分辨力、锥束伪影及条形伪影等。结果 相比常规算法,利用AIIR重建CT图像可使噪声降低61.74%~99.76%、X线剂量降低60.00%~90.00%,同时将低密度分辨能力提升至1.99~4.86倍、高对比度空间分辨力提升至1.55~2.57倍,且能明显减轻锥束伪影及条形伪影。结论 AIIR用于重建CT图像较常规算法具有明显优势。
英文摘要:
      Objective To observe the value of artificial intelligence iterative reconstruction (AIIR) for reconstruction CT images of phantoms. Methods AIIR was developed through combining model-based iterative reconstruction (MBIR) with deep learning (DL) techniques. CT scanning of CCT MITA IQ phantom, CT ACR 464 phantom, Catphan 700 phantom, disc stack phantom and CT PBU-60 whole body phantom were performed, and the images were reconstructed with conventional algorithms like filtered back projection (FBP) and KARL 3D iterative reconstruction, as well as AIIR, respectively. Then the noise, X-ray dosage, as well as low contrast resolution, high contrast spatial resolution, cone-beam artifacts and streaking artifacts of various reconstructed images were compared. Results Compared to CT images reconstructed with conventional algorithms, those reconstructed with AIIR showed 61.74%—99.76% reduction of image noise and 60.00%—90.00% reduction of X-ray dosage, while increased image low contrast resolution to 1.99—4.86 times and high contrast spatial resolution to 1.55—2.57 times. Additionally, AIIR significantly reduced cone-beam artifacts and streaking artifacts. Conclusion AIIR showed obvious advantages for reconstruction CT images of phantoms compared with conventional algorithms.
查看全文  查看/发表评论  下载PDF阅读器