PET/MRI影像组学特征预测肺腺癌与肺鳞癌病理分型价值

Application of PET/MRI radiomics in differentiating the pathological types of lung adenocarcinoma and squamous cell carcinoma

目的探讨PET/MRI影像组学特征预测肺腺癌与肺鳞癌病理分型的价值。方法回顾杭州市全景影像中心2018年10月至2020年12月初次行PET/MRI检查的53例肺癌患者临床、PET/MRI资料,其中肺腺癌36例,肺鳞癌17例。应用影像组学软件计算和选择与肺癌分型最相关的影像组学特征,通过五折交叉验证方式分为训练组及测试组,采用ROC曲线及AUC差异评价模型预测能力。结果从T2WI图像和PET图像中各提取了2600个影像组学特征,经最小绝对收缩和选择算子(Lasso)回归经过多次筛选及降维后进行特征选择,最终保留5个特征。PET/MRI模型的训练组及测试组AUC、F1评分法、召回率、精密度、灵敏度、准确度分别是0.881及0.826、0.882及0.781、0.931及0.829、0.838及0.768、0.931及0.829、0.831及0.714。结论基于PET/MRI影像组学特征构建的预测模型能对肺腺癌与肺鳞癌术前病理分型进行无创性、可重复性预测,为临床准确诊断和个体化治疗提供客观依据,对临床治疗具有重要的指导意义。

Objective To investigate the value of PET/MRI radiomics in differentiating pathological histological classification of lung adenocarcinoma and squamous cell carcinoma.Methods The clinical and imaging data of 53 patients with lung adenocarcinoma or lung squamous cell carcinoma who underwent PET/MRI examination in Hangzhou Panoramic Imaging Center from October 2018 to December 2020 were retrospectively analyzed.The pathological diagnosis was obtained from surgery specimens or fiberscopic biopsy,there were 36 cases of lung adenocarcinoma and 17 cases of squamous cell carcinoma.setting adenocarcinoma as positive(36 cases)and squamous cell carcinoma as negative(17 cases).The radiomics characteristics related to lung cancer pathological classification were calculated and selected with radiomic software,and the patients were divided into training set and testing set by cross-validation,and the prediction efficacy of the model was evaluated by ROC curve.Results A total of 2600 radiomic characteristics were extracted from T2WI images and PET images.After multiple screening and dimensionality reduction by mini mum absolute contraction and selection operator(Lasso)regression method,5 characteristics were finally retained.The AUC,F1 score,recall rate,precision,sensitivity and accuracy of the training set and the testing set were 0.881 and 0.826,0.882 and 0.781,0.931 and 0.829,0.838 and 0.768,0.931 and 0.829,0.831 and 0.714,respectively.Conclusion The established model based on PET/MRI radiomics can effectively differentiate lung adenocarcinoma and squamous cell carcinoma preoperatively,which may be used clinically.