Additional value of metabolic parameters to PET/CT-based radiomics nomogram in predicting lymphovascular invasion and outcome in lung adenocarcinoma
Additional value of metabolic parameters to PET/CT-based radiomics nomogram in predicting lymphovascular invasion and outcome in lung adenocarcinoma
复制标题
基于 PET/CT 的放射组学列线图的代谢参数在预测肺腺癌淋巴管侵犯和预后方面的附加价值
DOI：
10.1007/s00259-020-04747-5
复制
发表时间：
2020-05-25
影响因子：
9.1
通讯作者：
Wang, Zhenguang
中科院分区：
文献类型：
--
作者：
Nie, Pei;Yang, Guangjie;Wang, Zhenguang
目的肺腺癌（LAC）患者的淋巴血管侵犯（LVI）影响手术效果。术前预测LVI是具有挑战性的使用传统的临床和影像学参数。本研究的目的是探讨结合临床因素、CT特征和最大标准化摄取值（SUVmax）的放射组学图在预测LAC的LVI和预后方面的价值，并评估SUVmax对基于PET/CT的放射组学图的附加价值。方法回顾性分析272例LAC患者（有lvi者87例，无lvi者185例）的PET/CT扫描结果，其中160例SUVmax≥2.5的患者进行PET放射组学分析。分析临床资料和CT特征，选择独立的LVI预测因子。对独立LVI预测因子和SUVmax的性能进行了评价。采用最小绝对收缩和选择算子算法构建二维（2D）和三维（3D） CT放射组学特征（RSs）和PET-RS，并计算放射组学评分（Rad-scores）。建立了SUVmax （RNWS）和不SUVmax （RNWOS）的放射组学图，包括rad评分和独立的临床和CT因素。对模型的性能进行了校准、鉴别和临床实用性评估。评估所有临床、PET/CT、病理、治疗和放射组学参数，以确定无进展生存（PFS）的独立预测因子。结果sct形态学是LVI的独立预测因子。与CT形态学相比，SUVmax在训练集（P< 0.001）和测试集（P= 0.042）中具有更好的识别能力。共提取1409个CT和PET放射组学特征，分别精简为8个、8个和10个特征，构建2D CT- rs、3D CT- rs和PET- rs。2D- rs与3D- rs的AUC差异无统计学意义（P < 0.05）， 2D CT-RS的AUC高于3D CT-RS。CT-RS、CT-RNWOS和CT-RNWS在训练集（AUC分别为0.799、0.796和0.851）和测试集（AUC分别为0.818、0.822和0.838）上具有良好的辨别能力。CT-RNWS与CT-RNWOS在训练集中的AUC差异有统计学意义（P= 0.044）。决策曲线分析（DCA）显示CT-RNWS在临床有用性方面优于CT-RS和CT-RNWOS。此外，DCA显示PETCT-RNWS比PET-RNWS和CT-RNWS提供最高的净效益。病理和rnws预测的lvi存在和lvi不存在患者的PFS有显著差异（P< 0.001）。在244例ct - rnws预测队列中，碳水化合物抗原125 （CA125）、癌胚抗原（CEA）、神经元特异性烯醇化酶（NSE）、病理LVI、组织学亚型和SUVmax是PFS的独立预测因子；在141例petct - rnws预测队列中，CA125、NSE、病理性LVI和SUVmax是PFS的独立预测因子。结论放射组学特征图，包括rad评分，临床和PET/CT参数，对LAC的LVI状态具有良好的预测作用。病理性LVI和SUVmax与LAC预后相关。
PurposeLymphovascular invasion (LVI) impairs surgical outcomes in lung adenocarcinoma (LAC) patients. Preoperative prediction of LVI is challenging by using traditional clinical and imaging parameters. The purpose of this study was to investigate the value of the radiomics nomogram integrating clinical factors, CT features, and maximum standardized uptake value (SUVmax) to predict LVI and outcome in LAC and to evaluate the additional value of the SUVmax to the PET/CT-based radiomics nomogram.MethodsA total of 272 LAC patients (87 LVI-present LACs and 185 LVI-absent LACs) with PET/CT scans were retrospectively enrolled, and 160 patients with SUVmax ≥ 2.5 of them were used for PET radiomics analysis. Clinical data and CT features were analyzed to select independent LVI predictors. The performance of the independent LVI predictors and SUVmax was evaluated. Two-dimensional (2D) and three-dimensional (3D) CT radiomics signatures (RSs) and PET-RS were constructed with the least absolute shrinkage and selection operator algorithm and radiomics scores (Rad-scores) were calculated. The radiomics nomograms, incorporating Rad-score and independent clinical and CT factors, with SUVmax (RNWS) or without SUVmax (RNWOS) were built. The performance of the models was assessed with respect to calibration, discrimination, and clinical usefulness. All the clinical, PET/CT, pathologic, therapeutic, and radiomics parameters were assessed to identify independent predictors of progression-free survival (PFS).ResultsCT morphology was the independent LVI predictor. SUVmax provided better discrimination capability compared with CT morphology in the training set (P< 0.001) and test set (P= 0.042). A total of 1409 CT and PET radiomics features were extracted and reduced to 8, 8, and 10 features to build the 2D CT-RS, 3D CT-RS, and the PET-RS, respectively. There was no significant difference in AUC between the 2D-RS and 3D-RS (P> 0.05), and 2D CT-RS showed a relatively higher AUC than 3D CT-RS. The CT-RS, the CT-RNWOS, and the CT-RNWS showed good discrimination in the training set (AUC [area under the curve], 0.799, 0.796, and 0.851, respectively) and the test set (AUC, 0.818, 0.822, and 0.838, respectively). There was significant difference in AUC between the CT-RNWS and CT-RNWOS (P= 0.044) in the training set. Decision curve analysis (DCA) demonstrated the CT-RNWS outperformed the CT-RS and the CT-RNWOS in terms of clinical usefulness. Furthermore, DCA showed the PETCT-RNWS provided the highest net benefit compared with the PET-RNWS and CT-RNWS. PFS was significantly different between the pathologic and RNWS-predicted LVI-present and LVI-absent patients (P< 0.001). Carbohydrate antigen 125 (CA125), carcinoembryonic antigen (CEA), neuron-specific enolase (NSE), pathologic LVI, histologic subtype, and SUVmax were independent predictors of PFS in the 244 CT-RNWS-predicted cohort; and CA125, NSE, pathologic LVI, and SUVmax were the independent predictors of PFS in the 141 PETCT-RNWS-predicted cohort.ConclusionsThe radiomics nomogram, incorporating Rad-score, clinical and PET/CT parameters, shows favorable predictive efficacy for LVI status in LAC. Pathologic LVI and SUVmax are associated with LAC prognosis.