智能识别辅助工具在肺部磨玻璃结节诊断中的应用研究  

Study on the application of the assisted tool of intelligent recognition in the diagnosis of pulmonary GGN

作　　者：卢杰伦 王慧贤[2] 刘天柱[2] 黄东晖[2] LU Jie-lun;WANG Hui-xian;LIU Tian-zhu(The Second Clinical Medical College of Guangzhou University of Chinese Medicine,Guangzhou 510000,China;不详)

机构地区：[1]广州中医药大学第二临床医学院,广东广州510000 [2]广东省中医院珠海医院呼吸内科,广东珠海519000

出　　处：《中国医学装备》2021年第10期19-23,共5页China Medical Equipment

基　　金：国家中医临床研究基地业务建设科研专项(JDZX2015230)“清金化痰汤通过免疫调节炎症反应治疗慢性阻塞性肺疾病急性加重期(痰热壅肺证)临床研究”;广东省中医药管理局科研项目(20182140)“清热化痰法治疗AECOPD痰热壅肺证的系统评价及临床研究”;广东省第二批中医临床优秀人才研修项目(粤中医办函[2017]267号)。

摘　　要：目的:在肺部磨玻璃结节(GGN)早期CT图像诊断中,建立智能识别辅助工具,实现降低肺部GGN检测人工投入、提高诊断效率并最终扩大筛查范围。方法:构建肺部GGN智能识别辅助诊断策略,采用肺结节分割方法,从CT图像中获取肺结节轮廓,进而提取轮廓区域内的结节形状、纹理特征;基于轮廓特征构建毛玻璃甄别模型,通过甄别模型鉴定结节是否为毛玻璃结节,提取、量化并评估“毛刺”等恶性征象特征;基于结节区域内形状、纹理、恶性征象等特征建立恶性度评估模型建立恶性度评估机制,实现恶性肿瘤辅助检测功能。结果:通过构建肺部GGN智能识别辅助诊断策略,在测试中可实现肺结节轮廓分割、结节区域特征提取、肺部GGN诊断、“毛刺”等恶性征象检测功能,并可实现结节恶性度评估功能。在性能测试中,34例样本中30例检测正确,整体诊断准确率为88.23%;单样本检测时间为4.5 s,速度明显优于医师单样本平均诊断的23.6 s。结论:构建肺部GGN智能识别辅助诊断策略,可提高临床诊断效率,其整体策略具有实现简单、计算量稳定、查全率高等优点,有助于为临床提供辅助诊断工具,降低临床工作量,在肺癌辅助筛查中具有实践意义和研究价值。Objective:To establish an assisted tool of intelligent recognition in the diagnosis of computed tomography(CT)image at early stage of pulmonary ground glass nodules(GGN),so as to reduce the investment of manual work in the detection of pulmonary GGN,and improve the diagnosis efficiency,and finally expand the screening scope.Methods:The paper established a strategy of assisted diagnosis of intelligent recognition of pulmonary GGN.The segmentation method of pulmonary nodule was adopted to obtain the outline of pulmonary nodule from CT image,and then the shape and textural features of nodules within the field of outline were further extracted.Based on the features of outline,the screening model of ground glass was constructed.Whether the nodule was GGN was determined by screening model,and the characteristics of malignant sign such as“burr”was further extracted,quantized and assessed.And the assessment model and mechanism of malignant degree were established on the basis of the shape,texture and malignant signs in the region of nodule so as to realize the function of auxiliary detection for malignant tumor.Results:The strategy of constructing assisted diagnosis of intelligent recognition of pulmonary GGN could realize series of detection functions included the segmentation of the outline of pulmonary nodule,the feature extraction of nodule region,the diagnosis of pulmonary GGN,“burr”and other malignant signs in the detection.And it could realize the assessment function of malignant degree of nodule.In the test of performance,the detections of 30 cases of 34 samples were correct,and the overall accuracy of diagnosis was 88.23%,and the detection time of single sample was only 4.5s,which was significantly better than the average diagnosis time(23.6s)of physician for single sample.Conclusion:The constructed strategy of assisted diagnosis of intelligent recognition of pulmonary GGN can increase the diagnostic efficiency in clinical work,and its overall strategy has the advantages include the simplicity of realiza

关 键 词：肺癌 磨玻璃结节(GGN) 诊断 人工智能 智能识别辅助工具 计算机断层扫描(CT) 

分 类 号：R816.4[医药卫生—放射医学]