基于X线的纹理分析在诊断跟距联合畸形中的临床应用价值  

The clinical application value of texture analysis based on X-ray in the diagnosis of talocalcaneal coalition

作　　者：郝海凤 张卜天[1] 滕佩宏 祖莅惠 刘畅[1] 刘桂锋[1] HAO Haifeng;ZHANG Butian;TENG Peihong;ZU Lihui;LIU Chang;LIU Guifeng(Department of Radiology,China-Japan Union Hospital of Jilin University,Changchun 130033,China)

机构地区：[1]吉林大学中日联谊医院放射线科,吉林长春130033

出　　处：《中国实验诊断学》2024年第9期1021-1025,共5页Chinese Journal of Laboratory Diagnosis

基　　金：吴阶平医学基金会(320.6750.2022-11-50);吉林省科技厅科技发展计划项目(20220203113SF)。

摘　　要：目的构建跟距联合畸形(talocalcaneal coalition)的X线影像组学模型,并检验其对跟距联合畸形的筛查诊断能力。方法回顾性分析2019年1月至2023年3月吉林大学中日联谊医院放射线科200例行踝关节或足部X线检查的患者临床放射资料(跟距联合阳性及阴性各100例),手动勾画跟距联合畸形所在影像学区域,基于Python-pyradiomics库初步提取影像组学特征,通过曼-惠特尼U检验及最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)算法实现数据降维和特征筛选,用支持向量机(support vector machine,SVM)对筛选得到的影像组学特征分类建模,最终以受试者工作特征(receiver operating characteristic,ROC)曲线的曲线下面积(area under the curve,AUC)、精确度、召回率、敏感度、特异度及F1分数评价模型的诊断效能。结果从X线图像中初步提取到105个组学特征,经曼-惠特尼U检验及LASSO算法筛选出7个强相关性特征,最终以SVM分类器所构建模型的测试集AUC值为0.93,精确度、召回率、敏感度、特异度和F1分数分别为88%、85%、93%、92%、88%,对跟距联合畸形有良好的筛查诊断能力。结论基于X线的影像组学模型可作为筛查诊断跟距联合畸形的一种准确高效的无创性工具,帮助临床医师诊断跟距联合畸形。Objective To construct a X-ray-based radiomics model to test the screen and diagnostic ability of talocalcaneal coalition.Methods The retrospective study was performed on 200 patients who had the X-ray examinations of ankle or foot(100 cases each positive and negative for talocalcaneal coalition)in the Radiology Department of China-Japan Union Hospital of Jilin University from January 2019 to March 2023.The segmentation of talocalcaneal coalition was manually sketched to extract the radiomics features of each patient.Data reduction and feature selection can be realized through Mann-Whitney U rank sum test and least absolute shrinkage and selection operator(LASSO)algorithm.Through the Mann-Whitney U test and LASSO algorithm realized data reduction and feature screening.Support vector machine(SVM)was used to built the prediction model of those screened radiomics features.Finally,the screen and diagnostic ability of the model was evaluated by the area under the receiver operating characteristic(ROC)curve(AUC),accuracy,recall rate,sensitivity,specificity and F1-score.Results The initial 105 features were extracted from each X-ray film.Seven strongly correlated features were selected by Mann-Whitney U test and LASSO algorithm.The AUC value was 0.93,and the accuracy,recall rate,sensitivity,specificity and F1 scores were 88%,85%,93%,92%and 88%,respectively.The model has good diagnostic efficiency in the identification of talocalcaneal coalition.Conclusion The X-ray-based radiomics model provide a more accurate and efficient noninvasive tool for the identification of talocalcaneal coalition,which can effectively improve the detection rate of it.

关 键 词：跟距联合畸形 影像组学 X线成像 最小绝对收缩和选择算子 支持向量机 

分 类 号：R682.16[医药卫生—骨科学]