基于梯度技术优化肺结核继发纤维纵隔炎引发的肺动脉高压的CT图像分析  

Optimization of CT image analysis of pulmonary hypertension due to pulmonary tuberculosis secondary to fibrous mediastinitis based on gradient technique

作　　者：纳建荣[1] 严梅 褚娇娇 王一帆[1] NA Jianrong;YAN Mei;CHU Jiaojiao;WANG Yifan(Respiratory and Critical Care Department and adiology Department of Ningxia Medical University General Hospital,Yinchuan 750004,China)

机构地区：[1]宁夏医科大学总医院呼吸与危急重症科及放射科,宁夏银川750004

出　　处：《中国病原生物学杂志》2024年第9期1053-1056,共4页Journal of Pathogen Biology

基　　金：宁夏自然科学基金项目(No.2023AAC03665,2023AAC03603)。

摘　　要：目的本研究旨在开发并评估基于卷积神经网络(CNN)的深度学习模型,用于从梯度优化后的CT图像中自动识别肺结核继发纤维纵隔炎引发的肺动脉高压。方法通过采用梯度优化算法对CT图像进行处理,结合使用梯度下降法进行模型训练以提高诊断的精确性和效率。开发过程中,首先对数据进行归一化处理并应用数据增强技术以提升模型的泛化能力。采用的U-Net网络架构为基础框架,模型通过Adam优化器进行训练,初步学习率设置为0.001,并通过早停机制防止过拟合。采用精确度、召回率、F1分数和接收者操作特征曲线(ROC)下的面积(AUC)等关键指标进行评估。在独立测试集上,模型展现出了优异的表现。结果在独立测试集上,该模型展现出优异的性能。戴斯系数、召回率和精准率的平均值分别为0.789、0.794和0.820。结果表明,该模型能有效识别肺动脉高压的CT图像特征。模型在不同程度的肺结核患者中表现稳定,重度和极重度肺结核患者中的平均戴斯系数、召回率和精准率均达到0.80以上,而在轻度患者中这些指标稍低。模型的敏感性为85%,特异性为88%,F1分数为86%,接收者操作特征曲线(AUC)下的面积为96%,显示出优秀的诊断能力。结论经过优化后的图像构建的模型能够有效地识别肺动脉高压的CT图像特征,对于临床诊断和治疗规划具有重要的参考价值。Objective The aim of this study was to develop and evaluate a deep learning model based on convolutional neural networks(CNNs)for the automatic identification of pulmonary hypertension due to pulmonary tuberculosis secondary to fibrosing mediastinitis from optimized CT images.Methods CT images are processed by using a gradient optimization algorithm,combined with model training using gradient descent to improve diagnostic accuracy and efficiency.During the development process,the data is first normalized and data enhancement techniques are applied to improve the generalization ability of the model.The U-Net network architecture was used as the base framework,and the model was trained by the Adam optimizer,with the initial learning rate set to 0.001 and an early stopping mechanism to prevent overfitting.Key metrics such as precision,recall,F1 score and area under the receiver operating characteristic curve(ROC)(AUC)are used for evaluation.The model demonstrates excellent performance on an independent test set.Results The model demonstrated excellent performance on the independent test set.The mean values of the Dace coefficient,recall,and precision were 0.789,0.794,and 0.820,respectively.These results indicate that the model is effective in recognizing CT image features of pulmonary hypertension.The model performed consistently in patients with different degrees of tuberculosis,with mean Dess coefficients,recalls,and precision rates reaching above 0.80 in patients with severe and very severe tuberculosis,while these metrics were slightly lower in patients with mild disease.The model had a sensitivity of 85%,a specificity of 88%,an F1 score of 86%,and an area under the receiver operating characteristic curve(AUC)of 96%,demonstrating excellent diagnostic capabilities.Conclusion The model constructed from the optimized images can effectively identify the CT image features of pulmonary arterial hypertension,which is an important reference value for clinical diagnosis and treatment planning.

关 键 词：梯度优化 深度学习 CT图像 肺结核 肺动脉高压 

分 类 号：R521[医药卫生—内科学] R544.1[医药卫生—临床医学]