医学
肝细胞癌
荟萃分析
磁共振成像
肝硬化
子群分析
放射科
置信区间
磁共振弥散成像
内科学
核医学
作者
Michael Chan,Ya Ruth Huo,Nelson Trieu,Amer Mitchelle,Jacob George,Emily He,Alice Unah Lee,Jeff Chang,Jessica Yang
标识
DOI:10.1016/j.cgh.2021.02.036
摘要
This meta-analysis investigates the diagnostic performance of non-contrast magnetic resonance imaging (MRI) for the detection of hepatocellular carcinoma (HCC).A systematic review was performed to May 2020 for studies which examined the diagnostic performance of non-contrast MRI (multi-sequence or diffusion-weighted imaging (DWI)- alone) for HCC detection in high risk patients. The primary outcome was accuracy for the detection of HCC. Random effects models were used to pool outcomes for sensitivity, specificity, positive likelihood ratio (LR) and negative LR. Subgroup analyses for cirrhosis and size of the lesion were performed.Twenty-two studies were included involving 1685 patients for per-patient analysis and 2128 lesions for per-lesion analysis. Multi-sequence non-contrast MRI (NC-MRI) using T2+DWI±T1 sequences had a pooled per-patient sensitivity of 86.8% (95%CI:83.9-89.4%), specificity of 90.3% (95%CI:87.3-92.7%), and negative LR of 0.17 (95%CI:0.14-0.20). DWI-only MRI (DW-MRI) had a pooled sensitivity of 79.2% (95%CI:71.8-85.4%), specificity of 96.5% (95%CI:94.3-98.1%) and negative LR of 0.24 (95%CI:1.62-0.34). In patients with cirrhosis, NC-MRI had a pooled per-patient sensitivity of 87.3% (95%CI:82.7-91.0%) and specificity of 81.6% (95%CI:75.3-86.8%), whilst DWI-MRI had a pooled sensitivity of 71.4% (95%CI:60.5-80.8%) and specificity of 97.1% (95%CI:91.9-99.4%). For lesions <2 cm, the pooled per-lesion sensitivity was 77.1% (95%CI:73.8-80.2%). For lesions >2 cm, pooled per-lesion sensitivity was 88.5% (95%CI:85.0-91.5%).Non-contrast MRI has a moderate negative LR and high specificity with acceptable sensitivity for the detection of HCC, even in patients with cirrhosis and with lesions <2 cm. Prospective trials to validate if non-contrast MRI can be used for HCC surveillance is warranted.
科研通智能强力驱动
Strongly Powered by AbleSci AI