基于MR影像组学的骨巨细胞瘤分期诊断模型的构建与评估  

Construction and Evaluation of A Diagnostic Model for Staging Giant Cell Tumor of Bone Based on MR Imaging Omics

作　　者：李瑶 李海燕[3] 马晓文[3] 唐清华[4] 李勇亮 孙健 Li Yao;Li Haiyan;Ma Xiaowen;Tang Qinghua;Li Yongliang;Sun Jian(Institute of Endemic Diseases,School of Public Health,Xi’an Jiaotong University Health Science Center,Xi’an 710061;Department of Pathology,Red Society Hospital Affiliated to Xi’an Jiaotong University,Xi’an 710054;Image center nuclear magnetic chamber,Red Society Hospital Affiliated to Xi’an Jiaotong University,Xi’an 710054;Department of Osteoporosis,Red Society Hospital Affiliated to Xi’an Jiaotong University,Xi’an 710054;Rehabilitation of Traditional Chinese Medicine,Red Society Hospital Affiliated to Xi’an Jiaotong University,Xi’an 710054)

机构地区：[1]西安交通大学医学部公共卫生学院地方病研究所,陕西西安710061 [2]西安交通大学附属红会医院病理科,陕西西安710054 [3]西安交通大学附属红会医院影像中心核磁室,陕西西安710054 [4]西安交通大学附属红会医院骨质疏松科,陕西西安710054 [5]西安交通大学附属红会医院中医康复科,陕西西安710054

出　　处：《现代医用影像学》2024年第8期1498-1503,共6页Modern Medical Imageology

摘　　要：目的:探讨MR影像组学对骨巨细胞瘤(GCTB)分期诊断的预测价值。方法:回顾性分析2017年1月至2023年11月西安市红会医院GCTB患者的MRI数据,其中1期18例,2期67例,3期42例,分别在T_(1)WI、T_(2)WI序列上逐层勾画感兴趣区(ROI)获得三维ROI,每个ROI提取组学特征。将数据集经随机分层抽样法按照7∶3的比例分为训练集和测试集。通过深度学习网络构建GCTB分期诊断模型,探究哪些影像特征与GCTB分期诊断具有相关性。结果:经过特征筛选,构建GCTB分期影像诊断、临床特征以及两者融合模型,共建立3个相关模型。在训练集上,影像组学模型AUC达到了0.79(95%CI:0.75-0.83),临床特征模型达到了0.75(95%CI:0.72-0.79),融合模型取得了0.85(95%CI:0.82-0.91);在测试集中,影像组学模型和临床特征模型的AUC分别为0.77(95%CI:0.72-0.85)、0.84(95%CI:0.82-0.90),融合模型的AUC值为0.86(95%CI:0.82-0.92),均高于单独构建的模型。训练集和测试集中,融合模型的AUC与影像组学模型的AUC差异无统计学意义(P均>0.05)。结论:影像组学模型具有较好的预测GCTB分期诊断的能力,融合模型可进一步提高预测性能,可以较准确的识别GCTB的不同分期,使得临床医师能够尽早制订治疗决策。Objective:To investigate the predictive value of MR imaging histology for staging diagnosis of giant.Methods:MRI data of GCTB patients in Xi’an Honghui Hospital from January 2017 to November 2023 were retrospectively analyzed,of which 18 cases were stage 1,67 cases were stage 2,and 42 cases were stage 3.Three⁃dimensional ROI were obtained by layer⁃by⁃layer outlining the regions of interest(ROI)on the T1 WI and T2 WI sequences,and histological features were extracted from each ROI.The dataset was divided into training set and test set by random stratified sampling method in the ratio of 7∶3.A GCTB staging diagnosis model was constructed by deep learning network to explore which image features are correlated with GCTB staging diagnosis.Results:After feature screening,GCTB staging image diagnosis,clinical features,and the fusion of the two models were constructed,and a total of three relevant models were built.In the training set,the AUC of the imaging histology model reached 0.79(95%CI:0.75-0.83),the clinical feature model reached 0.75(95%CI:0.72-0.79),and the fusion model achieved 0.85(95%CI:0.82-0.91);in the test set,the AUC of the imaging histology model and the clinical feature model were 0.77(95%CI:0.72-0.85),0.84(95%CI:0.82-0.90),and the AUC value of the fusion model was 0.86(95%CI:0.82-0.92),which were higher than those of the separately constructed models.The difference between the AUC of the fusion model and the AUC of the imaging histology model was not statistically significant in the training and test sets(both P>0.05).Conclusion:The imaging histology model has a better ability to predict the diagnosis of GCTB staging,and the fusion model can further improve the prediction performance,which can more accurately identify the different staging of GCTB,enabling clinicians to formulate treatment decisions as early as possible.

关 键 词：骨巨细胞瘤 MRI ROC曲线 影像组学 纹理特征 

分 类 号：R445.2[医药卫生—影像医学与核医学] R738[医药卫生—诊断学]