深度学习算法优化儿童超低剂量气道CT的可行性  

Deep learning image reconstruction for evaluating large airways in children with ultra-low dose chest CT

作　　者：孙记航[1] 高军[1] 王帆宁[1] 王蓓[1] 霍爱华[1] 刘志敏[1] 彭芸[1] SUN Ji-hang;GAO Jun;WANG Fan-ning;WANG Bei;HUO Ai-hua;LIU Zhi-min;PENG Yun(Department of Radiology,Beijing Children’s Hospital,Capital Medical University/National Center for Children’s Health,Beijing 100045,China)

机构地区：[1]首都医科大学附属北京儿童医院/国家儿童医学中心影像中心,北京100045

出　　处：《影像诊断与介入放射学》2021年第1期10-14,共5页Diagnostic Imaging & Interventional Radiology

摘　　要：目的评价利用深度学习重组算法(DLIR)对儿童超低剂量胸部CT大气道的显示能力。方法连续选取17例[平均年龄(6.5±3.0)岁,范围3~12岁]需要复查胸部CT平扫的患儿入组,采用超低剂量CT扫描方案,管电压80 kV(3~6岁)或100 kV(7~12岁),管电流10 mA,转速0.35 s,重组为0.625 mm的50%自适应迭代重组(ASIR-V)图像,低强度DLIR(DL-L),中强度DLIR(DL-M)及高强度DLIR(DL-H)图像,作为观察组图像。患儿3个月内常规低剂量胸部CT作为对照组。采用5分制主观评价气道显示质量(5分最好,3分合格)。客观测量5组图像肺野的CT值及噪声值,统计两组患儿的有效剂量并比较。结果超低剂量组的CTDIvol和有效剂量为(0.07±0.03)mGy及(0.04±0.01)mSv,较低剂量组降低了98%。超低剂量组图像噪声值50%ASIR-V图像为50.6±10.1,DL-L为57.5±11.5,DL-M为52.2±10.8,DL-H为43.2±10.0,较对照组(27.5±8.2)明显增高。主观评价所有大气道评分均高于3分,其中DL-H为(4.7±0.3)分,接近对照组的(4.9±0.1)分。结论使用高强度DLIR优化超低剂量胸部CT的大气道显示,可在降低98%放射剂量的基础上,得到可供满意诊断的气道图像。Objective To assess the ability of a deep learning image reconstruction(DLIR)algorithm in evaluating large airways with ultra-low radiation dose chest CT in children.Methods 17 children(age range:3-12 years,average:6.5±3.0 years)underwent plain chest CT in routine radiation dose initially and ultra-low-dose plain chest CT with 80 kV(3-6 years)or 100 kV(7-12 years),10 mA,0.35 s rotation time within 3 months.The routine radiation dose CT images were reconstructed with 50%ASIR-V.The ultra-low-dose CT images were reconstructed with the adaptive statistical iterative reconstruction-V at 50%strength(50%ASIR-V)with standard kernel and DLIR at low(DL-L),medium(DL-M)and high(DL-H)settings at 0.625 mm slice thickness.Two radiologists independently evaluated the large airways in the lungs on a 5-point scales(5:excellent;4:good;3:measurable and acceptable;2:detectable;1:not acceptable).The image quality of the lung parenchyma was measured and statistically compared among the 5 reconstruction groups.The radiation dose was recorded for the two patient groups.Results The volume CT dose index[(0.07±0.03)mGy]and effective dose[(0.04±0.01)mSv]of the ultra-low dose CT were 98%lower than that of routine CT[(3.2±0.8)mGy,(2.1±0.5)mSv].The image noise of lung parenchyma on ultra-low dose CT with 50%ASIR-V[(50.6±10.1)HU,DL-L(57.5±11.5)HU],DL-M[(52.2±10.8)HU],and DL-H[(43.2±10.0)HU]was significantly higher than that on routine CT[(27.5±8.2)HU].However,the image quality scores for evaluating large airways were all higher than 3,with DL-H having the highest value on ultra-low dose CT at 4.7±0.3 similar to the score of 4.9±0.1 on routine CT.Conclusion Ultra-low dose chest CT reconstructed with DL-H provides image quality comparable to routine radiation dose CT for evaluating large airways in the lungs with 98%radiation dose reduction.

关 键 词：体层摄影术 X线计算机 儿童 气管 图像重组 深度学习 

分 类 号：R816.92[医药卫生—放射医学]