Zhao Xiaoqian, Zhao Yan, Zhang Jingmian, Zhang Zhaoqi, Liu Lihua, Zhao Xinming
Department of Nuclear Medicine, The Fourth Hospital of Hebei Medical University, 12 Jiankang Road, Shijiazhuang, 050011, Hebei, China.
Department of Oncology, The Fourth Hospital of Hebei Medical University and Hebei Provincial Tumor Hospital, Shijiazhuang, Hebei, China.
EJNMMI Res. 2023 Jan 22;13(1):4. doi: 10.1186/s13550-023-00956-9.
In recent years, immune checkpoint inhibitor (ICI) therapy has greatly changed the treatment prospects of patients with non-small cell lung cancer (NSCLC). Among the available ICI therapy strategies, programmed death-1 (PD-1)/programmed death ligand-1 (PD-L1) inhibitors are the most widely used worldwide. At present, immunohistochemistry (IHC) is the main method to detect PD-L1 expression levels in clinical practice. However, given that IHC is invasive and cannot reflect the expression of PD-L1 dynamically and in real time, it is of great clinical significance to develop a new noninvasive, accurate radiomics method to evaluate PD-L1 expression levels and predict and filter patients who will benefit from immunotherapy. Therefore, the aim of our study was to assess the predictive power of pretherapy [F]-fluorodeoxyglucose ([F]FDG) positron emission tomography/computed tomography (PET/CT)-based radiomics features for PD-L1 expression status in patients with NSCLC.
A total of 334 patients with NSCLC who underwent [F]FDG PET/CT imaging prior to treatment were analyzed retrospectively from September 2016 to July 2021. The LIFEx7.0.0 package was applied to extract 63 PET and 61 CT radiomics features. In the training group, the least absolute shrinkage and selection operator (LASSO) regression model was employed to select the most predictive radiomics features. We constructed and validated a radiomics model, clinical model and combined model. Receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC) were used to evaluate the predictive performance of the three models in the training group and validation group. In addition, a radiomics nomogram to predict PD-L1 expression status was established based on the optimal predictive model.
Patients were randomly assigned to a training group (n = 233) and a validation group (n = 101). Two radiomics features were selected to construct the radiomics signature model. Multivariate analysis showed that the clinical stage (odds ratio [OR] 1.579, 95% confidence interval [CI] 0.220-0.703, P < 0.001) was a significant predictor of different PD-L1 expression statuses. The AUC of the radiomics model was higher than that of the clinical model in the training group (0.706 vs. 0.638) and the validation group (0.761 vs. 0.640). The AUCs in the training group and validation group of the combined model were 0.718 and 0.769, respectively.
PET/CT-based radiomics features demonstrated strong potential in predicting PD-L1 expression status and thus could be used to preselect patients who may benefit from PD-1/PD-L1-based immunotherapy.
近年来，免疫检查点抑制剂（ICI）疗法极大地改变了非小细胞肺癌（NSCLC）患者的治疗前景。在现有的ICI治疗策略中，程序性死亡-1（PD-1）/程序性死亡配体-1（PD-L1）抑制剂在全球范围内应用最为广泛。目前，免疫组织化学（IHC）是临床实践中检测PD-L1表达水平的主要方法。然而，鉴于IHC具有侵入性，且无法动态、实时地反映PD-L1的表达情况，因此开发一种新的非侵入性、准确的放射组学方法来评估PD-L1表达水平，并预测和筛选能从免疫治疗中获益的患者具有重要的临床意义。因此，我们研究的目的是评估基于治疗前[F]氟脱氧葡萄糖（[F]FDG）正电子发射断层扫描/计算机断层扫描（PET/CT）的放射组学特征对NSCLC患者PD-L1表达状态的预测能力。
回顾性分析2016年9月至2021年7月期间334例治疗前行[F]FDG PET/CT成像的NSCLC患者。应用LIFEx7.0.0软件包提取63个PET和61个CT放射组学特征。在训练组中，采用最小绝对收缩和选择算子（LASSO）回归模型选择最具预测性的放射组学特征。我们构建并验证了放射组学模型、临床模型和联合模型。采用受试者工作特征（ROC）曲线及曲线下面积（AUC）评估这三种模型在训练组和验证组中的预测性能。此外，基于最优预测模型建立了预测PD-L1表达状态的放射组学列线图。
患者被随机分为训练组（n = 233）和验证组（n = 101）。选择两个放射组学特征构建放射组学特征模型。多因素分析显示，临床分期（优势比[OR] 1.579，95%置信区间[CI] 0.220 - 0.703，P < 0.001）是不同PD-L1表达状态的显著预测因素。放射组学模型在训练组（0.706对0.638）和验证组（0.761对0.640）中的AUC均高于临床模型。联合模型在训练组和验证组中的AUC分别为0.718和0.769。
基于PET/CT的放射组学特征在预测PD-L1表达状态方面显示出强大潜力，因此可用于预选可能从基于PD-1/PD-L1的免疫治疗中获益的患者。
