Department of Radiology, Zhujiang Hospital, Southern Medical University, 253 Gongye Middle Avenue, Haizhu District, Guangzhou, Guangdong, 510282, China.
Department of Radiology, Zhujiang Hospital, Southern Medical University, 253 Gongye Middle Avenue, Haizhu District, Guangzhou, Guangdong, 510282, China.
Clin Radiol. 2023 Dec;78(12):e993-e1000. doi: 10.1016/j.crad.2023.08.018. Epub 2023 Sep 5.
To evaluate the feasibility of intranodular and perinodular computed tomography (CT) radiomics features for predicting the occurrence of pulmonary haemorrhage after percutaneous CT-guided transthoracic lung biopsy (PCTLB) in pulmonary nodules.
The data for 332 patients with pulmonary nodules who underwent PCTLB were reviewed retrospectively. Pulmonary haemorrhage after PCTLB was evaluated using CT (144 cases occurred). Radiomics features based on gross nodular (GNV) and perinodular volumes (PNV) were extracted from pre-biopsy CT images and features selection using least absolute shrinkage and selection operator (LASSO) regression, and three radiomics scores (rad-scores) were built. Rad-scores, clinical, and clinical-radiomic models were developed and evaluated to predict the occurrence of pulmonary haemorrhage.
Five, five, and six significant features were selected for prediction of pulmonary haemorrhage based on GNV, PNV, and GNV + PNV, respectively. Lesion depth was the only clinical characteristics related to pulmonary haemorrhage. Lesion depth and rad-score based on GNV, PNV, and GNV + PNV for predicting the pulmonary haemorrhage achieved areas under the curves (AUCs) of 0.656, 0.645, 0.651, and 0.635 in the validation group, respectively. Three clinical-radiomic models improved the AUCs to 0.743, 0.723, and 0.748. The performance of rad-score_GNV + PNV combined with lesion depth outperformed the clinical model (p=0.024) and the radiomics signature (p=0.038). In addition, the radiomics signatures were significantly associated with higher-grade pulmonary haemorrhage (p<0.05).
Radiomics features from intranodular and perinodular regions of pulmonary nodules have good predictive ability for pulmonary haemorrhage after PCTLB, which may provide additional predictive value for clinical practice.
评估肺结节经皮 CT 引导经胸肺活检（PCTLB）后发生肺出血的结节内和结节周 CT 放射组学特征的可行性。
回顾性分析了 332 例接受 PCTLB 的肺结节患者的数据。采用 CT 评估 PCTLB 后肺出血（144 例发生）。从术前 CT 图像中提取基于大体结节（GNV）和结节周体积（PNV）的放射组学特征，并使用最小绝对收缩和选择算子（LASSO）回归进行特征选择，构建三个放射组学评分（rad-score）。建立并评估 rad-score、临床和临床放射组学模型，以预测肺出血的发生。
基于 GNV、PNV 和 GNV+PNV 预测肺出血，分别选择了 5、5 和 6 个显著特征。病变深度是唯一与肺出血相关的临床特征。基于 GNV、PNV 和 GNV+PNV 的病变深度和 rad-score 预测肺出血的验证组 AUC 分别为 0.656、0.645、0.651 和 0.635。三种临床放射组学模型将 AUC 提高至 0.743、0.723 和 0.748。rad-score_GNV+PNV 结合病变深度的性能优于临床模型（p=0.024）和放射组学特征（p=0.038）。此外，放射组学特征与较高等级的肺出血显著相关（p<0.05）。
肺结节内和结节周区域的放射组学特征对 PCTLB 后肺出血具有良好的预测能力，可为临床实践提供额外的预测价值。
