【摘要】 目的构建胃肠道间质瘤（GISTs）病理危险度分级的CT放射学模型并验证其效果。方法回顾性分析沧州市中心医院收治的140例GISTs患者术前动脉期CT图像。将患者依据2008年美国国立卫生研究院（NIH）分类标准分为极低危组（n=8）、低危组（n=47）、中危组（n=33）和高危组（n=52） 4组,并将患者按随机数字表法随机分为训练组（n=100）和验证组（n=40）。比较训练组和验证组的临床特点。从肿瘤全部感兴趣体积（VOI）中提取396个CT放射组学特征,采用随机森林算法选择5个关键特征,构建放射组学模型,并分析放射组学模型性能。结果训练组和验证组的年龄和性别在GISTs危险度分级上差异无统计学意义（P₁=0.844和0.464; P₂=0.807和0.464; P₃=0.638和0.392; P₄=0.619和0.334）。放射组学模型性能结果显示,训练组准确性为0.621;特异度为0.812;敏感度为0.615;阳性预测值（PPV）为0.619;阴性预测值（NPV）为0.845;曲线下面积（AUC）为0.829（95%CI:0.756～0.902）。验证组准确性为0.633;特异性为0.824;敏感性为0.767; PPV为0.609; NPV为0.856; AUC为0.859（95%CI:0.792～0.926）。结论基于CT放射组学模型可于术前对GISTs危险度分级进行有效鉴别,对临床医师指导治疗,评估预后具有很大帮助。更多还原

【Abstract】 Objective Construct a CT radiology model for pathological risk classification of gastrointestinal stromal tumors（GISTs） and verify its effect.Methods The preoperative arterial CT images of 140 patients with GISTs admitted to Cangzhou Central Hospital were retrospectively analyzed.The patients were divided into four groups according to the classification criteria of 2008 National Institutes of Health（NIH） : very low risk group（n = 8）,low risk group（n = 47）,medium risk group（n = 33） and high risk group（n = 52）.The patients were randomly divided into training group（n = 100） and verification group（n = 40） according to the random number table method.The clinical characteristics of the training group and the validation group were compared.396 CT radiological features were extracted from all tumor volume of interest（VOI）,5 key features were selected by random forest algorithm to construct radiological model,and the radioomics model performance was analyzed.Results Of training and validation set GISTs risk classification difference between age and gender on no statistical significance（P₁= 0.844 and 0.464; P₂=0.807 and 0.464; P₃= 0.638 and 0.392; P₄= 0.619 and 0.334）.The performance results of the radiology model showed that the accuracy of the training group was 0.621.Specificity was 0.812; Sensitivity was 0.615; Positive predictive value（PPV） was 0.619; Negative predictive value（NPV） was 0.845.Area under the Curve（AUC） was 0.829（95% CI: 0.756-0.902）.Verify group accuracy was 0.633; Specificity was 0.824;Sensitivity was 0.767; PPV was 0.609; NPV was 0.856; AUC was 0.859（95% CI: 0.792-0.926）.Conclusion The CT radiomics model can effectively distinguish the risk classification of GISTs before surgery,which is very helpful for clinicians to guide treatment and evaluate prognosis.更多还原