CT引导下低辐射剂量扫描联合深度学习重建算法在肺穿刺活检术中的应用  

Application of low radiation dose scanning guided by CT combined with deep learning reconstruction algorithms in lung puncture biopsies

作　　者：徐龙 李鑫 张丽 于楠 段海峰 XU Long;LI Xin;ZHANG Li;YU Nan;DUAN Haifeng(School of Medical Technology,Shaanxi University of Chinese Medicine,Xianyang 712046,China;Department of Medical Imaging,Affiliated Hospital of Shaanxi University of Chinese Medicine,Xianyang 712000,China)

机构地区：[1]陕西中医药大学医学技术学院,陕西咸阳712046 [2]陕西中医药大学附属医院医学影像科,陕西咸阳712000

出　　处：《分子影像学杂志》2025年第1期44-50,共7页Journal of Molecular Imaging

基　　金：陕西省教育厅青年创新团队科研计划项目(23JP036)。

摘　　要：目的探讨低辐射剂量扫描联合深度学习重建(DLIR)算法在CT引导下肺穿刺活检中应用的可行性及临床价值。方法选取2023年9月~2024年3月在陕西中医药大学附属医院行CT引导下肺穿刺患者,根据扫描方案不同,将60例肺穿刺活检患者分为常规剂量组(A组)和低剂量组(B组)。A组为100 kV,噪声指数(NI)=15;B组NI=45,其余扫描参数均相同。在常规剂量组中首次和末次全肺扫描分别采用A、B组参数扫描,用于评价深度学习重建算法(DLIR)改善图像质量潜能。A组中首次全肺扫描采用滤波反投影(FBP)和权重为50%自适应统计迭代重建-V(50%ASIR-V)重建,末次全肺扫描采用深度学习重建算法的3种强度(DLIR-L、DLIR-M、DLIR-H)重建图像。分别测量脊柱旁肌肉、皮下脂肪及主动脉血管CT值和SD值,计算信号噪声比(SNR)和对比噪声比(CNR)。对比A、B两组的患者基线特征、穿刺过程中总的辐射剂量、病理阳性率及并发症发生率。结果5种不同条件下重建图像在肌肉、皮下脂肪及主动脉血管处CT值差异无统计学意义(P>0.05),SD值、SNR和CNR值差异有统计学意义(P<0.05),组间两两比较分析显示,DLIR-H图像与50%ASIR-V图像在肌肉、脂肪、血管SD和SNR的差异无统计学意义(P>0.05);FBP vs DLIR-H和DLIR-L vs DLIR-H组的CNR值差异有统计学意义(P<0.05)。与A组总辐射剂量相比,B组总辐射剂量减少约93.6%(P<0.001)。两组图像质量均能满足临床穿刺需要,两组患者的基线特征、病理阳性率及并发症发生率的差异均无统计学意义(P>0.05)。结论低剂量CT扫描结合DLIR重建,可以显著降低图像噪声,提高图像质量,且不影响穿刺安全性和病理阳性率。Objective To explore the feasibility and clinical value of low radiation dose scanning combined with deep learning reconstruction(DLIR)algorithm in CT-guided lung puncture biopsy.Methods Patients who underwent CT-guided lung puncture at the Affiliated Hospital of Shaanxi University of Traditional Chinese Medicine from September 2023 to March 2024 were selected,and according to the different scanning protocols,60 lung puncture biopsy patients were divided into a conventional dose group(group A)and a low-dose group(group B).Group A was 100 kV,with a noise index(NI)=15;Group B had an NI=45;the rest of the scanning parameters were the same.The first and last whole-lung scans in the conventional dose group were scanned with the parameters of group A and B,respectively.They were used to evaluate the image quality improvement potential of the deep learning reconstruction algorithm(DLIR).The first whole-lung scan in group A was reconstructed with filtered back projection(FBP)and weighted 50%adaptive statistical iterative reconstruction-V(50%ASIRV),and the last whole-lung scan was reconstructed with the three intensities of the deep learning reconstruction algorithm(DLIR-L,DLIR-M,DLIR-H)reconstructed images.The CT and SD values of paraspinal muscles,subcutaneous fat,and aortic vessels were measured,and the signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)were calculated.The baseline characteristics of the patients,the total radiation dose during puncture,the pathological positivity rate,and the complication rate were compared between group A and B.Results The differences in CT values at muscle,subcutaneous fat,and aortic vessels in the reconstructed images under the five different conditions were not statistically significant(P>0.05).The differences in SD,SNR,and CNR values were statistically significant(P<0.05).The two-by-two comparative analyses between the groups showed that there were no statistically significant differences between the DLIR-H images and the 50%ASIR-V images in muscle,fat,and vessel SD and SN

关 键 词：深度学习 穿刺活检 低剂量 辐射剂量 肺 

分 类 号：R563[医药卫生—呼吸系统] R816.4[医药卫生—内科学]