基于腹部低keV图像的能谱CT影像组学模型评估骨质状态  

Evaluation on bone status by radiomics model based on low keV abdominal images of spectral CT

作　　者：程启烨 刘义军[1] 王诗耕 童小雨 范勇 魏巍 陈安良[1] 胡梦婷 张竞颐 CHENG Qi-ye;LIU Yi-jun;WANG Shi-geng(Department of Radiology,the First Affiliated Hospital of Dalian Medical University,Dalian 116011,China)

机构地区：[1]大连医科大学附属第一医院放射科,辽宁大连116011

出　　处：《放射学实践》2025年第4期529-534,共6页Radiologic Practice

摘　　要：目的:探究基于腹部能谱40~70 keV单能量图像构建的影像组学模型评估骨质状态的可行性及其效能。方法:回顾性将2022年2月-2023年11月因疑似结直肠癌而在本院行CT结肠造影(CTC)及全腹部能谱CT平扫和三期增强扫描的407例患者纳入本研究。将CTC图像(120 kVp)传至定量CT(QCT)Pro工作站,测量3个椎体(L1~L3)骨密度(BMD)的平均值,并根据测量结果将患者分为3组:骨质疏松组(BMD<80 mg/cm^(3))、骨量减少组(80 mg/cm^(3)≤BMD≤120 mg/cm^(3))及骨量正常组(BMD>120 mg/cm^(3))。基于全腹部能谱CT扫描的平扫期数据,重建40~70 keV(间隔10 keV)4组单能量CT图像,每组图像按7∶3的比例随机分为训练集和测试集。分别在4组图像上于L1~L3椎体内勾画全体积ROI并提取影像组学特征。然后,采用最小冗余最大相关性(mRMR)、递归特征消除(RFE)和最小绝对收缩和选择算子(LASSO)筛选最佳组学特征。以QCT测量结果为标准,采用随机森林(RF)分类器分别基于4组单能量CT图像构建4个评估骨质状态的组学模型,绘制受试者工作特征(ROC)曲线并计算Macro-AUC以评估4个模型的诊断效能。结果:基于4组单能量CT图像,分别提取了10、9、3和9个最佳组学特征。在测试集中各模型的Macro-AUC随keV的降低呈逐渐上升的趋势,分别为0.887、0.893、0.894和0.897,均具有良好的诊断效能。结论:基于腹部能谱CT低keV单能量(40~70 keV,间隔10 keV)图像构建的骨质状态评估模型具有良好的诊断效能。Objective:To explore the efficacy and feasibility of radiomics models based on 40~70keV(with an interval of 10keV)monoenergetic abdominal images of spectral CT to evaluate bone status.Methods:407 patients with suspected colorectal cancer who underwent CT colography(CTC)examination and full-abdomen spectral CT scanning in our hospital from February 2022 to November 2023 were retrospectively collected.The CTC images(120kVp)were transferred to quantitative CT(QCT)pro workstation to measure the mean bone mineral density(BMD)of L1~L3 vertebrae.According to the BMD measured on QCT,the patients were classified into osteoporosis(BMD<80mg/cm^(3)),osteopenia(80mg/cm^(3)≤BMD≤120mg/cm^(3)),and normal bone mass(BMD>120mg/cm^(3)).Four groups of monoenergetic CT images of 40~70keV(with 10keV interval)were reconstructed from the full-abdomen CT scan.Each group of images was randomly divided into the training set and the test set at a ratio of 7∶3.The whole-volume region of interest(ROI)was delineated from L1~L3 vertebral bodies in four groups of images,and the radiomic features were extracted.The best radiomics features were selected by minimum redundancy maximum relevance(mRMR),recursive feature elimination(RFE),least absolute shrinkage and selection operator(LASSO).Using the measurement results of QCT as the reference standard,four radiomics models for bone status assessment were constructed by random forest(RF)classifier based on four monoenergetic monoenergetic CT images.The receiver operating characteristic(ROC)curve was plotted and the macro area under curve(Macro-AUC)was calculated to evaluate the diagnostic performance of the four models.Results:Based on 40~70keV(with an interval of 10keV)monoenergetic CT images,10,9,3 and 9 optimal radiomics features were extracted,respectively.The AUC values of each model in test set increased gradually with the decrease of keV,and they were 0.887,0.893,0.894 and 0.897,respectively,which all achieved good diagnostic performance.Conclusion:The bone status assessment models based o

关 键 词：腰椎 骨密度 能谱CT 单能量图像 影像组学 随机森林 

分 类 号：R814.42[医药卫生—影像医学与核医学] R681.5[医药卫生—放射医学]