基于改进Boosting集成模型的肺炎感染诊断方法  

A method for diagnosis of pneumonia infection based on improved Boosting ensemble model

作　　者：杨倩[1] 王莉[1] 张萍[1] 宫艳艳[1] 付玉叶 YANG Qian;WANG Li;ZHANG Ping;GONG Yanyan;FU Yuye(Clinical Laboratory,Shaanxi Provincial People's Hospital,Xi'an 710068,China)

机构地区：[1]陕西省人民医院检验科,陕西西安710068

出　　处：《分子影像学杂志》2025年第4期435-440,共6页Journal of Molecular Imaging

基　　金：陕西省自然科学基础研究计划项目(2024JC-YBQN-0820)。

摘　　要：目的提出一种基于改进Boosting集成模型的肺炎感染诊断方法。方法选取2023年9月~2024年5月在陕西省人民医院经CT检查的肺炎感染患者315例,对所有患者进行CT诊断。在CT图像的预处理阶段,应用图像增强技术提升图像质量,确保模型在特征提取时获取更加清晰的影像信息。在特征提取过程中通过XGBoost框架提取纹理特征、形状特征和像素强度信息,利用主成分分析法减少特征维度。通过引入聚焦损失函数解决样本不平衡问题,确保模型对良性和恶性样本有更均衡的关注。采用贝叶斯优化方法,构建高斯过程回归模型,对超参数进行调整。结果本研究所提出的诊断方法在测试集中的平均曲线下面积值为0.9649,F1分数为0.9423,显著优于轻量级梯度提升机、随机森林和K最近邻等对比模型。结论本研究所提出的诊断方法为提高肺炎感染的识别和早期干预提供了有效工具,有助于医生更早识别高风险患者,制定个性化的治疗方案。Objective To propose a diagnosis method for pneumonia infection based on improved Boosting integration model.Methods A total of 315 patients with pneumonia infection who were examined by CT in Shaanxi Provincial People's Hospital from September 2023 to May 2024 were selected,and CT diagnosis was carried out for all patients.In the preprocessing stage of CT images,image enhancement technology was applied to improve the image quality and ensure that the model acquired clearer image information during feature extraction.In the feature extraction process,texture features,shape features and pixel intensity information are extracted through the XGBoost framework,and the principal component analysis is used to reduce the feature dimensions.In addition,the sample imbalance problem is solved by introducing a focus loss function to ensure that the model has a more balanced focus on benign and malignant samples.Meanwhile,Bayesian optimisation is used in the hyperparameter optimisation process to construct a Gaussian process regression model to adjust the hyperparameters,thus ensuring that the optimal parameter combinations are selected to further improve the prediction accuracy of the model.Results The diagnostic method proposed in this study has a mean area under the curve(mAUC)value of 0.9649 and an F1 score of 0.9423 in the test set,which significantly outperforms the comparative models such as lightweight gradient booster,random forest,and K-nearest neighbour.Conclusion The diagnostic method proposed in this study provides an effective tool to improve the identification and early intervention of pneumonia infections,helping physicians to identify high-risk patients earlier and develop personalised treatment plans.

关 键 词：肺炎感染 CT图像:XGBoost 贝叶斯优化 

分 类 号：R563.1[医药卫生—呼吸系统]