Objective: To investigate the value of CT-based radiomics and clinical data in predicting the efficacy of non-vascularized bone grafting (NVBG) in hip preservation, and to construct a visual, quantifiable, and effective method for decision-making of hip preservation.
Methods: Between June 2009 and June 2019, 153 patients (182 hips) with osteonecrosis of the femoral head (ONFH) who underwent NVBG for hip preservation were included, and the training and testing sets were divided in a 7∶3 ratio to define hip preservation success or failure according to the 3-year postoperative follow-up. The radiomic features of the region of interest in the CT images were extracted, and the radiomics-scores were calculated by the linear weighting and coefficients of the radiomic features after dimensionality reduction. The clinical predictors were screened using univariate and multivariate Cox regression analysis. The radiomics model, clinical model, and clinical-radiomics (C-R) model were constructed respectively. Their predictive performance for the efficacy of hip preservation was compared in the training and testing sets, with evaluation indexes including area under the curve, C-Index, sensitivity, specificity, and calibration curve, etc. The best model was visualised using nomogram, and its clinical utility was assessed by decision curves.
Results: At the 3-year postoperative follow-up, the cumulative survival rate of hip preservation was 70.33%. Continued exposure to risk factors postoperative and Japanese Investigation Committee (JIC) staging were clinical predictors of the efficacy of hip preservation, and 13 radiomic features derived from least absolute shrinkage and selection operator downscaling were used to calculate Rad-scores. The C-R model outperformed both the clinical and radiomics models in predicting the efficacy of hip preservation 1, 2, 3 years postoperative in both the training and testing sets ( P<0.05), with good agreement between the predicted and observed values. A nomogram constructed based on the C-R model showed that patients with lower Rad-scores, no further postoperative exposure to risk factors, and B or C1 types of JIC staging had a higher probability of femoral survival at 1, 2, 3 years postoperatively. The decision curve analysis showed that the C-R model had a higher total net benefit than both the clinical and radiomics models with a single predictor, and it could bring more net benefit to patients within a larger probability threshold.
Conclusion: The prediction model and nomogram constructed by CT-based radiomics combined with clinical data is a visual, quantifiable, and effective method for decision-making of hip preservation, which can predict the efficacy of NVBG before surgery and has a high value of clinical application.
目的: 探讨CT影像组学和临床资料在预测非血管化植骨（non-vascularized bone grafting，NVBG）保髋疗效中的价值，构建一种可视量化、有效的保髋决策方法。.
方法: 纳入2009年6月—2019年6月接受NVBG进行保髋的153例（182髋）股骨头坏死（osteonecrosis of the femoral head，ONFH）患者，以7∶3比例划分训练集（130髋）和测试集（52髋），根据术后3年时的随访结果定义保髋成败。提取CT影像中感兴趣区的影像组学特征，根据降维后的影像组学特征及其系数的线性加权获得影像组学评分，采用单因素、多因素Cox回归分析筛选临床预测因子，分别构建影像组学模型、临床模型和综合模型。在训练集及测试集中比较3种模型对保髋疗效的预测性能，评价指标包括曲线下面积、C-Index、敏感度、特异度和校准曲线等，采用列线图对最佳模型可视量化并通过决策曲线评估其临床效用。.
结果: 术后3年随访时，保髋累积生存率达70.33%。术后是否继续接触致病因素和日本骨坏死研究会（JIC）分型是影响保髋疗效的临床预测因子，最小绝对收缩和选择算子降维所得的13个影像组学特征被用于构建Rad-scores。在训练集和测试集中，综合模型预测术后1、2、3年保髋疗效的性能均优于临床模型和影像组学模型（ P<0.05），预测值和实际观察值间具有良好一致性。基于综合模型所构建的列线图显示，较低的Rad-scores、术后不再接触致病因素和JIC分型中B型及C1型患者，术后1、2、3年股骨头生存概率较高。决策曲线分析结果表明，综合模型的总净收益均高于单一预测因子的临床模型和影像组学模型，可在较大概率阈值范围内带给患者更多净获益。.
结论: CT影像组学联合临床资料所构建的预测模型和列线图，是一种可视量化、有效的保髋决策方法，能在术前预测NVBG的保髋疗效，具备较高的临床应用价值。.
Keywords: CT-based radiomics; Osteonecrosis of the femoral head; efficacy of hip preservation; nomogram; non-vascularized bone grafting; visual and quantitative assessment.