基于心脏磁共振T1 mapping和逻辑回归方法建立影像组学模型鉴别肺动脉高压

Cardiac magnetic resonance native T1 mapping-based Logistic regression radiomic model for the distinction of pulmonary hypertension

目的评估基于心脏磁共振平扫T1 mapping的影像组学模型用于鉴别肺动脉高压(PH)的价值。方法回顾性分析2019年7月-2021年12月就诊于北京协和医院并接受心脏磁共振检查的175例患者,其中116例为PH患者,59例为非PH患者。将其中140例患者(93例PH患者和47例非PH患者)归入训练集,其余35例患者(23例PH患者和12例非PH患者)归入验证集。采用专用的影像组学研究平台对平扫T1 mapping图像进行勾画并提取影像组学特征。基于逻辑回归方法建立影像组学模型,鉴别PH患者和非PH患者。绘制受试者操作特征曲线并计算曲线下面积(AUC),评估基于平扫T1 mapping的逻辑回归模型用于诊断PH的准确性。结果基于逻辑回归方法的影像组学模型在训练集和验证集诊断PH的AUC分别为0.973(95%CI:[0.930,0.993])和0.841(95%CI:[0.678,0.942])。采用最佳截断值0.829,模型在训练集和验证集诊断PH的准确度分别为91.43%和80.00%。结论基于平扫T1 mapping的和逻辑回归方法的影像组学模型在诊断PH中具有一定的潜在应用价值。

Objective To explore the role of native T1 mapping-based radiomics in distinguishing patients with and without pulmonary hypertension(PH).Methods A total of 175 consecutive patients who had been referred to Peking Union Medical College Hospital and received cardiac magnetic resonance imaging examination between July 2019 and December 2021 were included for a retrospective analysis,including 116 patients with PH and 59 patients without PH.140 patients(including 93 PH patients and 47 without PH)were allocated to the training cohort and 35(including 23 PH patients and 12 without PH)allocated to the test cohort.Radiomic features were extracted from native T1 mapping images using a dedicated software platform.The radiomic model was constructed by using Logistic regression algorithm for differentiation between patients with and without PH.Receiver operating characteristic curves and area under the curve(AUC)were calculated to explore the diagnostic accuracy.Results The radiomic model showed a favorable performance for the PH distinction.The AUC of the Logistic regression model in the training cohort and test cohort were 0.973(95%CI:[0.930,0.993])and 0.841(95%CI:[0.678,0.942]),respectively.Using the optimal cutoff value 0.829,the radiomic model exhibited an accuracy of 91.43% in the training cohort and 80.00 %in the test cohort,respectively.Conclusion The radiomic models based on native T1 mapping hold promise for a discriminating between patients with and without PH.