去噪-重建联合算法BM3D-GAMP在欠采样LDCT肺癌筛查中的应用价值  

Application value of denoising-reconstruction joint algorithm BM3D-GAMP in undersampling LDCT lung cancer screening

作　　者：成明峰 孙希子 夏黎明[3] CHENG Ming-feng;SUN Xi-zi;XIA Li-ming(School of cyber science and engineering,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]华中科技大学,武汉430074 [2]武汉软件工程职业学院,武汉430205 [3]华中科技大学同济医学院附属同济医院放射科,武汉430030 [4]华中科技大学出版社,武汉430223

出　　处：《放射学实践》2025年第3期395-402,共8页Radiologic Practice

基　　金：国家“新一代人工智能”重大项目(2021ZD0111104);中央高校基本科研业务费资助(HUST,2024JYCXJJ038);国家自然科学基金资助项目(82272109)。

摘　　要：目的:针对少视角的低剂量CT图像重建问题,提出一种BM3D-GAMP稀疏重建算法,并探究其在低剂量肺癌筛查中的潜在应用价值。方法:回顾性收集298例发现结节的常规剂量CT平扫病例的DICOM资料和投影数据,选取各例最大结节最大横径对应层面的投影数据。以0.1的压缩比例均匀选取投影角度以模拟稀疏均匀采样策略对压缩的投影数据进行重建,并使用Matlab实现算法。通过客观评价指标和主观图像质量评分比较提出的压缩重建算法与另两种算法的重建性能,并将重建图像与原始图像质量进行对比。3名高年资放射科医师对3种不同算法重建出的最大结节层面的图像进行评分。结果:客观指标和3名医生得到的统一临床评价均表明提出的联合算法在模拟肺癌筛查LDCT欠采样图像的去噪重建中性能最优(P<0.001)。亚组分析表明这种图像重建质量的优越性仅在对实性结节的重建中不再显著(BM3D-IT vs.BM3D-GAMP,P=0.808),且这种差异在小结节(5~<15 mm)上更显著。结论:去噪-重建联合算法在少角度采样的肺癌筛查LDCT图像重建上有较高的应用价值。Objective:A BM3D-GAMP sparse reconstruction algorithm is proposed to address the problem of low-dose CT image reconstruction with limited-angle projections,and its potential application value in low-dose lung cancer screening was investigated.Methods:The DICOM data and projection data of 298 cases with nodules detected in conventional-dose CT plain scans were retrospectively collected.The projection data corresponding to the maximum transverse diameter of the largest nodule in each case were selected.A compression ratio of 0.1 was uniformly used to select projection angles to simulate a sparse uniform sampling strategy.The compressed projection data were then reconstructed using an algorithm implemented using Matlab.The reconstruction performance of the proposed compression reconstruction algorithm and the other two algorithms were compared through objective evaluation indices and subjective image quality scores,and the quality of the reconstructed images was compared with that of the original images.Three senior radiologists scored the images of the largest nodule layer reconstructed by the three different algorithms.Results:Both objective indices and the unified clinical evaluations by the three doctors indicated that the proposed hybrid algorithm had the best performance in the denoising and reconstruction of undersampled LDCT images in simulated lung cancer screening(P<0.001).Subgroup analysis demonstrated that the superiority in image reconstruction quality was no longer significant in the reconstruction of solid nodules(BM3D-IT vs.BM3D-GAMP,P=0.808),and this difference was more significant in small nodules(5~<15mm).Conclusion:The proposed novel denoising-reconstruction joint algorithm has a considerable application value in the LDCT image reconstruction for lung cancer screening with sparse-angle samplings.

关 键 词：体层摄影术 X线计算机 图像去噪 压缩重建 低剂量CT 肺肿瘤 肺癌筛查 

分 类 号：R814.42[医药卫生—影像医学与核医学] R734.2[医药卫生—放射医学]