【摘要】 目的：探究计算机断层扫描（computed tomography,CT）影像组学多分类联合模型在磨玻璃结节病理分析的预测价值。方法：回顾性收集2019年2月至2023年3月新乡医学院第一附属医院收治的影像表现为磨玻璃结节并经病理证实的早期肺腺癌患者资料，共285例患者的324个结节按照浸润性腺癌（invasive adenocarcinoma,IAC）、微浸润性腺癌（minimally invasive adenocarcinoma,MIA）及侵袭前病变（preinvasive lesions,PILs）分为三组，行递归消除及单变量逻辑回归对组学及临床-影像特征选择后，使用逻辑回归（Logistic regression,LR）、支持向量机（support vector machine,SVM）、随机森林（random forest,RF）和集成学习（stacking）构建七个模型，评估模型预测效能。结果：基于结合临床-影像-组学特征和集成机器学习策略的混合联合模型相比其他六个模型具有更准确的预测性能，准确度、精确度、特异性、召回率、F1评分分别为0.791、0.788、0.857、0.790、0.789。结论：基于CT影像组学建立的多分类联合模型能够较好地预测肺磨玻璃结节病理分型，有利于影像准确诊断及为临床制定治疗方案提供依据。更多还原

【Abstract】 Objective: To assess the predictive value of a combined multiclassification model for computed tomography(CT) in the pathological analysis of ground-glass nodules(GGN). Methods: Pulmonary GGN lesions that were pathologically confirmed as invasive adenocarcinoma(IAC), minimally invasive adenocarcinoma(MIA), adenocarcinoma in situ(AIS), and preinvasive lesions(PILs), were collected from patients who were treated at The First Affiliated Hospital of Xinxiang Medical University between February 2019 and March 2023. A total of324 nodules were retrospectively collected from 285 patients, and divided into three groups: infiltrating IAC, MIA, and PILs. Radiomics and clinical-CT features were selected through recursive feature elimination and univariate Logistic regression. Seven models were constructed using Logistic regression(LR), support vector machine(SVM), random forest(RF), and integrative learning(stacking). Results: The hybrid model combining clinical-CT-radiomics features and an integrative strategy showed superior predictive performance, with an accuracy of0.791, precision of 0.788, specificity of 0.857, recall of 0.790, and F1-Score of 0.789. Conclusions: The multiclassification joint model based on CT-radiomics is effective in predicting pathological classification of pulmonary GGNs. This model aids in accurate imaging diagnosis and can provide a basis for optimizing treatment plans.更多还原