CT影像组学在肺腺癌分化水平预测中的应用

Application of CT Radiomics in Predicting Differentiation Level of Lung Adenocarcinoma

目的 探讨高分辨CT平扫影像的特征提取和机器学习模型在预测肺腺癌分化水平中的应用价值。方法 回顾性分析507例经手术病理确诊为肺腺癌且分化水平明确的患者资料。根据分级标准,将入组病例分为低分化组和中高分化组。提取CT影像特征,并采用7种机器学习算法构建预测模型。计算模型在预测肺腺癌分化水平时的曲线下面积(AUC)、准确度、特异度和敏感度。结果 低分化组175例,中高分化组332例。XGBoost模型表现最佳,其在验证集的AUC、准确度、特异度和敏感度分别为0.878、0.829、0.667和0.727。结论 基于CT影像组学模型能有效预测肺腺癌的低分化与中高分化水平。

Objective To investigate the value of prediction of the differentiation level in lung adenocarcinoma based on CT radiomics model.Methods Data from 507 patients with postoperative pathological confirmed lung adenocarcinoma and clearly defined differentiation level of lung adenocarcinoma were retrospective analyzed.The enrolled cases were divided into poorly differentiation group and moderate-to-high differentiation group based on the grading criteria.CT image features were extracted,and seven machine learning algorithms were used to construct prediction models to obtain the AUC,accuracy,specificity,and sensitivity.Results The poorly differentiation group consisted of 175 cases,while the moderate-to-high differentiation group had 332 cases.The XGBoost model demonstrated the best performance,with the AUC,accuracy,specificity,and sensitivity of this model on the validation set being 0.878,0.829,0.667,and 0.727,respectively.Conclusion CT radiomics model can effectively predict the differentiation level of poorly differentiation and moderate-to-high differentiation in lung adenocarcinoma.