目的本研究旨在结合传统MRI序列及增强检查,提取多模态高通量影像组学特征并联合语义特征,使用不同的机器学习分类器构建不同的模型并绘制列线图来鉴别高级别胶质瘤(high-grade glioma,HGG)和单发性脑转移瘤(solitary brain metastasis,...目的本研究旨在结合传统MRI序列及增强检查,提取多模态高通量影像组学特征并联合语义特征,使用不同的机器学习分类器构建不同的模型并绘制列线图来鉴别高级别胶质瘤(high-grade glioma,HGG)和单发性脑转移瘤(solitary brain metastasis,SBM)。材料与方法本研究对101名患者的多参数MR图像进行了回顾性分析,由两位资深医师标定肿瘤感兴趣区,然后对每个序列分别提取影像组学特征后进行组合,共提取428组影像组学特征。为消除人为标定差异,进行组内相关系数一致性检验,并运用最大相关最小冗余算法选取最具相关性的特征,然后进一步通过最小绝对收缩和选择算子算法筛除冗余特征。本研究采用支持向量机、逻辑回归、随机森林及K近邻四种算法建立分类模型。结合放射科医生评估的七项语义特征,通过卡方检验和多因素分析去除差异无统计学意义的语义特征。然后结合组学特征建立综合模型并绘制列线图。最终,评价各模型的诊断能力,以确定最优分类器。结果HGG及SBM患者建立的影像组学模型中LR的受试者工作特征曲线下面积(area under the curve,AUC)值最高,训练集与测试集分别为0.90和0.90。语义特征建立的模型中随机森林模型性能最好,训练集和测试集AUC分别为0.82和0.87。语义特征联合影像组学评分后采用逻辑回归建立的模型性能最好,训练集和测试集AUC分别为0.91和0.92。结论本研究使用影像组学机器学习分类器并联合其他图像语义特征绘制列线图对HGG及SBM进行鉴别,这是一种非侵入性方法,具有较好的准确性,为临床决策和实践提供了较大的帮助。展开更多
AIM:To investigate the relationship of solitary lymph node metastasis(SLNM)and age with patient survival in gastric cancer(GC).METHODS:The medical records databases of China’s Beijing Cancer Hospital at the Peking Un...AIM:To investigate the relationship of solitary lymph node metastasis(SLNM)and age with patient survival in gastric cancer(GC).METHODS:The medical records databases of China’s Beijing Cancer Hospital at the Peking University School of Oncology and Shanghai Tenth People’s Hospital affiliated to Tongji University were searched retrospectively to identify patients with histologically proven GC and SLNM who underwent surgical resection between October 2003 and December 2012.Patients with distant metastasis or gastric stump carcinoma following resection for benign disease were excluded from the analysis.In total,936 patients with GC+SLNM were selected for analysis and the recorded parameters of clinicopathological disease and follow-up(range:13-2925 d)were collected.The Kaplan-Meier method was used to stratify patients by age(≤50 years-old,n=198;50-64 years-old,n=321;≥65 years-old,n=446)and by metastatic lymph node ratio[MLR<0.04(1/25),n=180;0.04-0.06(1/25-1/15),n=687;≥0.06(1/15),n=98]for 5-year survival analysis.The significance of intergroup differences between the survival curves was assessed by a log-rank test. RESULTS:The 5-year survival rate of the entire GC+SLNM patient population was 49.9%.Stratification analysis showed significant differences in survival time(post-operative days)according to age:≤50 yearsold:950.7±79.0 vs 50-64 years-old:1697.8±65.9 vs≥65 years-old:1996.2±57.6,all P<0.05.In addition,younger age(≤50 years-old)correlated significantly with mean survival time(r=0.367,P<0.001).Stratification analysis also indicated an inverse relationship between increasing MLR and shorter survival time:<0.04:52.8%and 0.04-0.06:51.1%vs≥0.06:40.5%,P<0.05.The patients with the shortest survival times and rates were younger and had a high MLR(≥0.06):≤50 years-old:496.4±133.0 and 0.0%vs 50-65 years-old:1180.9±201.8 and 21.4%vs≥65 years-old:1538.4±72.4 and 37.3%,all P<0.05.The same significant trend in shorter survival times and rates for younger patients was seen with the mid-range MLR g展开更多
文摘目的本研究旨在结合传统MRI序列及增强检查,提取多模态高通量影像组学特征并联合语义特征,使用不同的机器学习分类器构建不同的模型并绘制列线图来鉴别高级别胶质瘤(high-grade glioma,HGG)和单发性脑转移瘤(solitary brain metastasis,SBM)。材料与方法本研究对101名患者的多参数MR图像进行了回顾性分析,由两位资深医师标定肿瘤感兴趣区,然后对每个序列分别提取影像组学特征后进行组合,共提取428组影像组学特征。为消除人为标定差异,进行组内相关系数一致性检验,并运用最大相关最小冗余算法选取最具相关性的特征,然后进一步通过最小绝对收缩和选择算子算法筛除冗余特征。本研究采用支持向量机、逻辑回归、随机森林及K近邻四种算法建立分类模型。结合放射科医生评估的七项语义特征,通过卡方检验和多因素分析去除差异无统计学意义的语义特征。然后结合组学特征建立综合模型并绘制列线图。最终,评价各模型的诊断能力,以确定最优分类器。结果HGG及SBM患者建立的影像组学模型中LR的受试者工作特征曲线下面积(area under the curve,AUC)值最高,训练集与测试集分别为0.90和0.90。语义特征建立的模型中随机森林模型性能最好,训练集和测试集AUC分别为0.82和0.87。语义特征联合影像组学评分后采用逻辑回归建立的模型性能最好,训练集和测试集AUC分别为0.91和0.92。结论本研究使用影像组学机器学习分类器并联合其他图像语义特征绘制列线图对HGG及SBM进行鉴别,这是一种非侵入性方法,具有较好的准确性,为临床决策和实践提供了较大的帮助。
基金Supported by Grants awarded to Dr.Chun-Qiu Chen from the National Science Foundation of China,No.81170345the Shanghai Tenth People’s Hospital Project for Cultivating Tutors of Doctors,No.12HBBD110
文摘AIM:To investigate the relationship of solitary lymph node metastasis(SLNM)and age with patient survival in gastric cancer(GC).METHODS:The medical records databases of China’s Beijing Cancer Hospital at the Peking University School of Oncology and Shanghai Tenth People’s Hospital affiliated to Tongji University were searched retrospectively to identify patients with histologically proven GC and SLNM who underwent surgical resection between October 2003 and December 2012.Patients with distant metastasis or gastric stump carcinoma following resection for benign disease were excluded from the analysis.In total,936 patients with GC+SLNM were selected for analysis and the recorded parameters of clinicopathological disease and follow-up(range:13-2925 d)were collected.The Kaplan-Meier method was used to stratify patients by age(≤50 years-old,n=198;50-64 years-old,n=321;≥65 years-old,n=446)and by metastatic lymph node ratio[MLR<0.04(1/25),n=180;0.04-0.06(1/25-1/15),n=687;≥0.06(1/15),n=98]for 5-year survival analysis.The significance of intergroup differences between the survival curves was assessed by a log-rank test. RESULTS:The 5-year survival rate of the entire GC+SLNM patient population was 49.9%.Stratification analysis showed significant differences in survival time(post-operative days)according to age:≤50 yearsold:950.7±79.0 vs 50-64 years-old:1697.8±65.9 vs≥65 years-old:1996.2±57.6,all P<0.05.In addition,younger age(≤50 years-old)correlated significantly with mean survival time(r=0.367,P<0.001).Stratification analysis also indicated an inverse relationship between increasing MLR and shorter survival time:<0.04:52.8%and 0.04-0.06:51.1%vs≥0.06:40.5%,P<0.05.The patients with the shortest survival times and rates were younger and had a high MLR(≥0.06):≤50 years-old:496.4±133.0 and 0.0%vs 50-65 years-old:1180.9±201.8 and 21.4%vs≥65 years-old:1538.4±72.4 and 37.3%,all P<0.05.The same significant trend in shorter survival times and rates for younger patients was seen with the mid-range MLR g
