使用基于CT的放射组学早期识别儿童难治性肺炎支原体：一项多中心研究

摘要(中文译文)

开发并验证一个利用临床、影像及影像组学特征早期预测儿童难治性肺炎支原体肺炎（RMPP）的模型。这项多中心回顾性研究共纳入419例患儿，分为训练队列（248例）、测试队列（62例）和外部验证队列（109例）。根据临床指南将患者分为非RMPP组和RMPP组。使用PyRadiomics从胸部CT扫描中提取影像组学特征，随后采用SelectKBest和最小绝对收缩和选择算子回归进行特征筛选。构建了三种基于随机森林的预测模型：临床-影像模型、影像组学模型和整合模型。采用受试者工作特征曲线下面积（AUC）、McNemar检验和净重分类改善指数（NRI）评估预测性能。在验证队列中，整合模型表现出最高的预测性能（AUC：0.811，95% CI：0.704-0.917），优于影像组学模型（AUC：0.784，95% CI：0.683-0.885）和临床-影像模型（AUC：0.675，95% CI：0.603-0.833）。McNemar检验显示，在验证队列中，影像组学模型与临床-影像模型（P=0.001）、影像组学模型与整合模型（P=0.013）、临床-影像模型与整合模型（P<0.001）之间的分类差异均有统计学意义。在验证队列中，整合模型的NRI高于影像组学模型和临床-影像模型（均P<0.001），但影像组学模型与临床-影像模型之间无差异（P=0.070）。关键预测因素包括D-二聚体、发热类型和全身免疫炎症指数，以及影像组学特征如灰度共生矩阵和小波峰度。整合模型结合了临床、影像和影像组学特征，增强了对RMPP的风险分层能力。

Abstract

To develop and validate a model that utilizing clinical, imaging, and radiomics characteristics for early predicting refractory pneumonia (RMPP) in children. This multicenter retrospective study included a total of 419 children, divided into training ( = 248), testing ( = 62), and external validation ( = 109) cohorts. Patients were classified into non-RMPP and RMPP groups based on clinical guidelines. Radiomics features were extracted from chest CT scans using PyRadiomics, followed by SelectKBest and least absolute shrinkage and selection operator regression for feature selection. Three random forest-based predictive models were developed: clinical-imaging, radiomics, and integrated. Predictive performance was evaluated using the area under the receiver operating characteristic curve (AUC), McNemar tests, and net reclassification improvement (NRI). The integrated model demonstrated the highest predictive performance (AUC: 0.811, 95% CI: 0.704-0.917), compared with both the radiomics (AUC: 0.784, 95% CI: 0.683-0.885) and clinical-imaging (AUC: 0.675, 95% CI: 0.603-0.833) models in the validation cohort. McNemar tests revealed significant differences in classification between the radiomics and clinical-imaging models ( = 0.001), radiomics and integrated models ( = 0.013), and clinical-imaging and integrated models ( < 0.001) in the validation cohort. In the validation cohort, the NRI was higher for the integrated model than for the radiomics and clinical-imaging models (both < 0.001) but did not differ between the radiomics and clinical-imaging models ( = 0.070). Key predictors included D-dimer, type of fever, and the systemic immune-inflammation index, along with radiomics features such as gray-level co-occurrence matrix and wavelet kurtosis. The integrated model, combining clinical, imaging, and radiomics features, enhances risk stratification for RMPP.

如何引用

Frontiers in medicine 的其他文献

• EGCG对多种消化系统疾病的化学预防作用：系统综述和荟萃分析
• 高原地区深静脉血栓形成风险评估量表的构建与验证
• 新冠肺炎后持续性肺结节的多模式预测：放射组学特征与临床和流行病学变量的整合
• 不同核心肌训练方案对慢性非特异性腰痛的比较有效性：系统综述和荟萃分析
• 纵隔肿瘤感染引起的梗阻性休克：病例报告和文献综述
• 成人心脏骤停后综合征基础研究进展：综合综述
• 产后子宫活动的客观监测：系统范围综述
• 全-RAS癌症疫苗设计和临床前景的免疫信息学见解
• 脑出血患者肺部感染风险预测模型及护理干预策略优化
• {"_":"半导体激光与次氯酸钠冲洗清除根管内Enterococcus faecalis的比较疗效。","i":["粪肠球菌"]}