超低剂量CT扫描结合深度学习图像重建对计算机辅助诊断系统定量分析肺结节的影响

目的 观察超低剂量CT（ULDCT）扫描结合深度学习图像重建（DLIR）对计算机辅助诊断系统（CAD）定量分析肺结节的影响。方法 前瞻性纳入56例复诊肺结节患者，行ULDCT及标准剂量CT（SDCT）检查。对ULDCT分别采用自适应统计迭代重建V40%（ASIR-V40%）及高强度DLIR（DLIR-H），对SDCT以ASIR-V40%进行重建，获得ULDCT-ASIR-V40%（A组）、ULDCT-DLIR-H（B组）及SDCT-ASIR-V40%（C组）图像。基于各组图像筛选长径4～30 mm肺结节为目标结节，由2名医师判断其为实性结节、钙化结节或实性结节。应用CAD分别基于3组图像评估结节类型，定量分析其长径、横径、密度、体积及恶性风险。结果 共纳入104个目标结节，医师判断为51个实性结节、26个钙化结节及27个非实性结节。CAD针对A、B组图像的评估结果为53个实性、24个钙化、27个非实性结节，其针对C组的评估结果与医师一致。相比C组相应结节类型，CAD判定的A组实性及钙化结节的密度、非实性结节的体积及恶性风险均降低，B组钙化结节的密度降低（P均＜0.05）；3组间结节其他CAD定量参数差异均无统计学意义（P均＞0.05）。结论 基于ULDCT行DLIR-H重建可能低估CAD判定的肺钙化结节的密度，但对其他CAD定量参数无明显影响。

Impact of ultra-low dose CT scanning combined with deep learning image reconstruction on quantitative analysis of pulmonary nodules using computer aided diagnostic system

Objective To investigate the impact of ultra-low dose CT （ULDCT） scanning combined with deep learning image reconstruction （DLIR） on quantitative analysis of pulmonary nodules using computer aided diagnostic system （CAD）. Methods Fifty-six further consultation patients with pulmonary nodules were prospectively enrolled. ULDCT and standard-dose CT （SDCT） were performed. The raw ULDCT images were reconstructed using adaptive statistical iterative reconstruction-V40% （ASIR-V40%） and high-strength DLIR （DLIR-H） to obtain ULDCT-ASIR-V40% （group A） and ULDCT-DLIR-H （group B） images， while SDCT images were reconstructed with ASIR-V40% to obtain SDCT-ASIR-V40% （group C） images. Pulmonary nodules with long diameter of 4—30 mm were selected as the target nodules based on reconstructed images. The nodules were divided into solid nodules， calcified nodules and non-solid nodules by 2 physicians. CAD software was used to evaluate the classification of nodules based on 3 groups of images， and the long diameter， transverse diameter， density， volume and malignant risk were quantitatively analyzed. Results Totally 104 target nodules were selected， including 51 solid nodules， 26 calcified nodules and 27 non-solid nodules according to physicians. CAD classified 53 solid， 24 calcified and 27 non-solid nodules based on group A and B， while based on group C， CAD classification was consistent with that of physicians’. Compared with group C， the density of solid and calcified nodules， the volume and malignant risk of non-solid nodules judged by CAD in group A decreased， so did the density of calcified nodules in group B （all P＜0.05）. No significant difference of the other CAD quantitative parameters of nodules was found among 3 groups （all P＞0.05）. Conclusion ULDCT scanning combined with DLIR might underestimate the density of calcified pulmonary nodules judged by CAD， but had no significant impact on the other CAD quantitative parameters.