【摘要】 目的 基于晚期肺腺癌患者吉非替尼治疗前肺部平扫CT影像组学特征，建立影像组学预测模型评估近期疗效。资料与方法 回顾性分析2020年1月—2022年5月哈尔滨医科大学附属第一医院接受吉非替尼治疗的71例晚期肺腺癌患者，治疗后第3个月评估疗效，依据实体肿瘤疗效评估标准分为治疗有效组41例和治疗无效组30例。所有患者按7∶3随机分为训练组49例、验证组22例，训练组采用最小绝对收缩和选择算子筛选最佳影像组学特征，建立K邻近、逻辑回归、支持向量机、自适应增强、梯度提升、随机森林、集成算法和高斯朴素贝叶斯8种模型，比较受试者工作特征曲线下面积和准确度，选择最佳模型。在验证组测试最佳模型性能，使用决策曲线检验最佳模型的临床可用性，使用校准曲线对预测模型的准确度进行可视化。结果 通过最小绝对收缩和选择算子回归筛选得到7个最佳影像组学特征，在多种模型中选择最佳模型逻辑回归，验证组中逻辑回归的曲线下面积为0.774(95%CI 0.536～0.951)，敏感度为0.846，特异度为0.556，准确度为0.727，召回率为0.846,F1-score为0.733。结论 基于CT影像组学模型可能筛选出晚期肺腺癌患者中靶向治疗的敏感人群。更多还原

【Abstract】 Purpose To evaluate the short-term efficacy of Gefitinib based on the non-contrast CT radiomics model in patients with advanced lung adenocarcinoma before treatment. Materials and Methods Seventy-one patients with advanced lung adenocarcinoma and treated with Gefitinib in the First Affiliated Hospital of Harbin Medical University from January 2020 to May 2022 were retrospectively analyzed, the efficacy of the treatment was evaluated in the third month after receiving the treatment, and all patients were divided into the treatment-effective group(n=41) and the treatment-ineffective group(n=30) according to the criteria for evaluating the efficacy of solid tumors.All patients were randomized into the training group and the validation group according to 7∶3, the least absolute shrinkage and selection operator were used in the training group to screen the best radiomics features, and eight models were established, including K-proximity,Logistic regression, support vector machine, adaptive enhancement, gradient boosting, random forest, ensemble algorithm and Gaussian naive Bayes. The optimal model was selected by comparing the area under the receiver operating characteristic curve(AUC) and accuracy. The optimal model performance was tested in the validation group, the clinical usability of the best model was tested using decision curves, the accuracy of the predictive models was visualized using calibration curves. Results The seven best radiomics features were obtained by least absolute shrinkage and selection operator regression screening, the best model Logistic regression was selected in a variety of models, the AUC value of Logistic regression in the validation group was 0.774(95% CI 0.536-0.951), with a sensitivity of 0.846, a specificity of 0.556,an accuracy of 0.727, a recall of 0.846, and an F1-score of 0.733. Conclusion It is possible to screen out the sensitive population of advanced lung adenocarcinoma for targeted therapy based on CT radiomics models.更多还原