基于对照CT影像的股骨头坏死预后预测  

Prognostic prediction of the femoral head using osteonecrosis and contralateral regions based on CT images

作　　者：赵文哲 胡守业 赵益童 刘毅楠 李瑶[3,4] 李达宁 黄定兴 孙健 ZHAO Wenzhe;HU Shouye;ZHAO Yitong;LIU Yinan;LI Yao;LI Daning;HUANG Dingxing;SUN Jian(School of Public Health,Xi’an Jiaotong University Health Science Center,Xi’an 710061;Department of Medical Imaging,The First Affiliated Hospital of Xi’an Jiaotong University,Xi’an 710061;Xi’an Honghui Hospital,Xi’an Jiaotong University Health Science Center,Xi’an 710054;School of Basic Medicine,Xi’an Jiaotong University,Xi’an 710061;Key Laboratory of Environment and Genes Related to Diseases,Ministry of Education,Xi’an 710061;National Health Commission Key Laboratory of Trace Elements and Endemic Diseases,Xi’an 710061,China)

机构地区：[1]西安交通大学医学部公共卫生学院,陕西西安710061 [2]西安交通大学第一附属医院医学影像科,陕西西安710061 [3]西安交通大学附属红会医院,陕西西安710054 [4]西安交通大学医学部基础医学院,陕西西安710061 [5]教育部环境与疾病相关基因重点实验室,环境与地方病研究室,陕西西安710061 [6]国家卫生健康委员会微量元素与地方病研究重点实验室,陕西西安710061

出　　处：《西安交通大学学报（医学版）》2022年第5期732-736,共5页Journal of Xi’an Jiaotong University（Medical Sciences）

摘　　要：目的利用影像组学方法开发一种基于CT影像的股骨头坏死患者预后预测模型,探究对侧正常股骨头区域的影像特征对预后预测的附加价值。方法本回顾性研究共纳入患者51例,所有患者均在术前行CT扫描。对每位患者勾画坏死股骨头和对侧正常股骨头两部分作为感兴趣区域。对患者的感兴趣区域共提取968个影像组学特征。综合利用单变量和多变量分析,根据股骨头坏死区域、对侧正常股骨头区域和综合两个区域的特征开发了3个预测模型。使用10次随机实验的方式进行模型构建及验证,计算10次实验的平均结果作为最终结果。结果对于坏死股骨头的影像特征,37个特征显示出对预后的预测价值,平均AUC值为0.7082±0.0299。预测模型在训练集和验证集的AUC分别为0.9110±0.0294和0.6886±0.0893。对于对侧正常股骨头的影像特征,14个特征显示出对预后的预测价值,这些特征的平均AUC为0.7036±0.0069,基于正常股骨头的影像特征构建的预测模型在训练集和验证集的AUC分别为0.8672±0.0395和0.6690±0.0726。综合预测模型在训练集和验证集的AUC值均高于基于坏死股骨头的影像特征构建的预测模型(训练集:0.9358±0.0166;验证集:0.7379±0.0908)。结论基于CT影像的影像组学方法可对股骨头坏死患者的预后进行预测;此外,患者的对侧正常股骨头的影像特征对患者预后预测具有额外价值;综合坏死股骨头和对侧正常股骨头的影像特征可对患者预后实现更准确的预测。Objective To develop a prognosis model based on CT images using radiomics method for patients with osteonecrosis of the femoral head(ONFH)and to investigate the additional prediction value of the imaging features of the contralateral normal femoral head regions for the prognosis prediction.Methods A total of 51 patients were included in this retrospective study.All the patients had preoperative CT images.For each patient,two regions of interest(ROIs)were involved,including the osteonecrosis region and the contralateral normal femoral head region.A total of 968 radiomics features were extracted for each patient.We made both the univariate and multivariable analyses.Three models were developed based on the features of osteonecrosis region,contralateral normal femoral head region,and both regions.The 10 times of repeated random experiments were used for model construction and validation.The average performance of the 10 times of experiments was reported as the results.Results For the features of osteonecrosis region,37 features showed significant predictive value,with the mean AUC value of 0.7082±0.0299.The AUC of the constructed prediction model was 0.9110±0.0294 and0.6886±0.0893 for the training set and validation set,respectively.For the features of contralateral normal femoral head region,14 features showed significant predictive value,with the mean AUC value of 0.7036±0.0069.The AUC value of the constructed model for the training set and validation set was 0.8672±0.0395 and 0.6690±0.0726,respectively.For the models developed based on combined features,the AUC value was higher than that of the models developed based on osteonecrosis region features(training set:0.9358±0.0166 vs.validation set:0.7379±0.0908).Conclusion We developed a novel CT images-based radiomics method to predict postoperative prognosis in patients with ONFH.Furthermore,the features of contralateral normal femoral head region has additional prediction value.Combining the imaging features of osteonecrosis region and contralateral normal

关 键 词：股骨头坏死 影像组学 对照CT影像 预后预测 

分 类 号：S857.16[农业科学—临床兽医学]