超低剂量胸部CT结合深度学习重建可用于肺结节评估

Ultra-low-dose chest CT combined with deep learning reconstruction can be used for pulmonary nodule assessment

目的探讨超低剂量CT(ULDCT)结合深度学习重建(DLIR)算法在肺结节评估中的可行性。方法于2023年6月~11月在陕西中医药大学附属医院前瞻性纳入142例因肺结节复查的患者,同时接受标准剂量CT检查(SDCT)和超低剂量CT检查(ULDCT)。SDCT采用基于多模型的迭代重建40%(ASIR-V40%)(A组),ULDCT分别采用ASIR-V40%(B组)和高强度深度学习重建(DLIR-H)(C组)。记录两种扫描方式的辐射剂量和3组图像人工检测结节数。测量3组图像肺组织、主动脉、肌肉的CT值和噪声值(SD),并计算各组织的信噪比(SNR)。采用双盲法对3组图像肺结节的恶性征象(毛刺、分叶、空洞或空泡、胸膜牵拉征、血管穿行)进行5分制主观评分。以病理诊断为金标准,对比分析超低剂量和常规剂量胸部CT检查对肺结节恶性征象的诊断效能。对3组图像的定量指标和主观评分进行统计学分析。结果ULDCT相较于SDCT辐射剂量降低约92.7%,差异有统计学意义(P<0.05)。C组图像肺组织、主动脉和肌肉的SD值低于B组,信噪比高于B组,肺结节恶性征象的显示能力优于B组(P<0.05),与A组的差异无统计学意义(P>0.05)。3组图像肺结节检出数量分别为187、179、187个。与病理结果对照,A组和C组诊断恶性肺结节的效能均高于B组,差异有统计学意义(P<0.05)。结论超低剂量胸部CT结合深度学习重建能够获得与标准剂量ASIR-V40%重建相当的图像质量,且对肺结节的检出及征象显示良好,临床可用于对肺结节的评估。

Objective To evaluate the feasibility of using deep learning reconstruction(DLIR)for pulmonary nodule assessment under ultra-low dose CT(ULDCT)scanning.Methods A total of 142 patients who underwent CT scans for pulmonary nodules re-examination included.All patients were examined by both standard-dose CT(SDCT)and ULDCT.SDCT images were reconstructed with adaptive statistical iterative reconstruction-V 40%(ASIR-V40%),ULDCT images were reconstructed with ASIR-V40%and DLIR-H,respectively.A total of three sets of images were obtained(Group A,group B,group C).The radiation dose of both scanning modes and the number of lung nodules were recorded manually.The CT values and noise values(SD)of lung tissue,aorta and muscle were measured in 3 groups images,and the signal-to-noise ratio(SNR)was calculated for each tissue.The malignant signs of lung nodules in the three groups were scored by double-blind method.Using the pathological diagnosis as the gold standard,the diagnostic efficacy of ULDCT and SDCT examination on the malignant signs(burr,lobular,pleural traction sign,vacuole or void,vascular perforation)of pulmonary nodules was analyzed by comparison.Statistical analysis was performed on the quantitative indicators and subjective scores of these three sets of images.Results The radiation dose of ULDCT was reduced by about 92.7%compared with SDCT,and the difference was statistically significant(P<0.05).The SD values of lung tissue,aorta and muscle in group C were lower than those in group B,and the SNR was higher than that in group B(P<0.05),and the ability to display malignant signs of nodules were better than those in group B,and there was no statistical difference between group C and group A(P>0.05).The number of pulmonary nodules detected in the three groups was 187,179 and 187,respectively.Compared with the pathological results,the efficacy of group A and group C in diagnosing malignant pulmonary nodules was higher than that of group B,and the difference was statistically significant(P<0.05).Conclusion Ultra-low-dose chest CT combined with deep learning reconstruction can obtain image quality comparable to ASIR-V40%of SDCT,and show good detection and signs of nodules,which can be used for clinical evaluation of pulmonary nodules.