基于CT影像组学模型预测骨巨细胞瘤术后复发

Prediction of Postoperative Recurrence of Giant Cell Tumor of Bone Based on CT Radiomics Models

目的 探讨基于术前CT平扫影像组学特征模型预测骨巨细胞瘤3年内复发的价值。资料与方法 回顾性分析2007年2月—2018年5月北京大学人民医院经组织学证实的95例骨巨细胞瘤的临床及影像学资料,以3∶1随机分为训练组71例和测试组24例。基于术前CT平扫提取的影像组学特征,于训练组中使用最小绝对收缩与选择算子算法进行降维后建立预测骨巨细胞瘤手术后3年内复发的影像组学标签。使用Spearman相关分析算法、最小绝对收缩与选择算子回归及梯度提升迭代决策树进行特征降维。使用多变量Logistic回归及随机森林纳入影像组学标签构建影像组学模型预测骨巨细胞瘤复发;使用受试者工作特征曲线评估影像组学标签在训练组中的准确性,并通过验证组进行验证。结果 提取术前CT平扫图像,共提取12个与骨巨细胞瘤术后3年内复发相关的影像组学特征构成影像组学标签,使用Logistic回归显示在训练组和测试组中预测骨巨细胞瘤术后3年内复发的曲线下面积分别为0.962和0.924,使用随机森林训练组及测试组的曲线下面积分别为0.991和0.917。结论基于术前CT影像组学模型作为非侵入性量化工具,预测骨巨细胞瘤复发具有良好的效能。

Purpose To determine if radiomics analysis based on preoperative CT can predict postoperative recurrence within three years of giant cell tumor of bone. Materials and Methods A total of 95 patients from Peking University People’s Hospital with pathologically confirmed giant cell tumor of bone from February 2007 to May 2018 were retrospectively selected and reviewed. CT images and clinical information acquired before the operation were retrieved for radiomics analysis. All patients were randomly divided into training group(n=71)and test group(n=24) at a ratio of 3∶1. Dimensionality reduction of radiomics features was performed using univariate analysis, least absolute shrinkage and selection operator and gradient boosting decision tree. Then, radiomics prediction models were constructed using Logistic regression and random forest. Diagnostic performance of the prediction model was evaluated through receiver operating characteristic curve and quantified by area under the curve. Independent validation was performed using test group data and receiver operating characteristic curve analysis. Results The final radiomics model was built using 12 selected features. Logistic regression showed that the area under the curve for predicting the recurrence of giant cell tumor within three years after surgery in the training group and the validation group was 0.962 and 0.924, respectively;and the area under the curve of random forest in the training group and test group was 0.991 and 0.917, respectively.Conclusion CT-based radiomics model, as a non-invasive quantitative tool, has good performance for predicting the recurrence of giant cell tumor of bone within three years before treatment.