摘要
目的建立基于多模态MRI影像组学及临床-影像表型特征的预测模型,评价其治疗前预测局部进展期直肠癌(locally advanced rectal cancer,LARC)新辅助放化疗(neoadjuvant chemoradiotherapy,nCRT)疗效的效能及临床应用价值,并通过Shapley分析赋予多模态影像组学模型可解释性。材料与方法回顾性分析2018年1月至2024年12月中国科学技术大学附属第一医院(中心1)、中国科学院合肥肿瘤医院(中心2)共172例接受nCRT并进行手术切除的LARC患者,收集临床病理资料及nCRT前MRI图像资料。根据术后病理结果,按照第8版AJCC直肠癌肿瘤退缩分级(tumor regression grading,TRG)标准,将TRG 0~1级的患者分为反应良好(good responders,GR)组(n=77),TRG 2~3级分为反应不良(poor responders,PR)组(n=95)。中心1的患者按7∶3的比例随机分为训练集(n=92)和内部验证集(n=40),中心2的患者作为外部验证集(n=40)。选择高分辨率轴位T2WI、轴位弥散加权成像(diffusion-weighted imaging,DWI)、矢状位对比增强T1WI(contrast-enhanced T1WI,CE-T1WI)序列沿肿瘤边缘勾画感兴趣区(region of interest,ROI),对其进行图像预处理后提取组学特征。采用Spearman相关性分析及最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)算法进行特征降维,筛选最优特征。通过极限梯度提升(eXtreme gradient boosting,XGBoost)机器学习分类器分别获取T2WI、DWI、CE-T1WI、多模态(T2WI、DWI、CE-T1WI)影像组学评分(radiomics score,Rad-score),分别构建T2WI、DWI、CE-T1WI、多模态影像组学模型;通过单因素、多因素logistic回归筛选临床-影像表型独立预测因子构建临床-影像表型模型。最后,选择临床-影像表型独立预测因子联合多模态影像组学Rad-score建立列线图模型。受试者工作特征(receiver operating characteristic,ROC)曲线用于各模型预测效能的评估,其中,表现最优的影像组学模型进一步使用Shapley算法进行解释。结果单因素及多因素logistic回归分析显示,年龄、肿瘤长径、nCRT方案可独立预测疗效,临床-影像表型模型在训练集、内部验证集、外部验证集的ROC曲线下面积(area under the curve,AUC)为0.80(95%CI:0.75~0.85)、0.73(95%CI:0.68~0.78)、0.60(95%CI:0.55~0.65)。最佳影像组学模型为基于T2WI、DWI、CE-T1WI序列联合构建的多模态模型,其在训练集、内部验证集、外部验证集的AUC分别为0.98(95%CI:0.95~1.00)、0.95(95%CI:0.91~0.99)、0.86(95%CI:0.81~0.91)。列线图模型在训练集的AUC为0.99(95%CI:0.97~1.00)、精确度、敏感度及特异度分别为98%、95%、98%;在内部验证集中分别为0.98(95%CI:0.95~1.00)、98%、98%、98%;在外部验证集中分别为0.88(95%CI:0.83~0.93)、88%、87%、87%。DeLong检验显示列线图模型的效能显著优于临床模型、影像组学模型(P<0.05)。Shapley分析显示DWI序列中wavelet-LHL_glszm_SmallAreaEmphasis为多模态影像组学模型最重要的特征。结论基于临床-影像表型及多模态磁共振影像组学特征的列线图模型可作为一种准确、无创的方法预测直肠癌患者nCRT疗效,Shapley算法可提供影像组学模型可解释性,经独立的外部队列验证,进一步证明该模型可用于指导临床诊疗及决策。
Objective:To establish a prediction model based on multiparametric MRI radiomics and clinical-radiology features,and evaluate its efficacy in predicting neoadjuvant chemoradiotherapy(nCRT)in patients with locally advanced rectal cancer.The Shapley algorithm was employed to enhance model interpretability.Materials and Methods:A retrospective analysis was conducted on 172 patients who received nCRT and surgery from the First Affiliated Hospital of the University of Science and Technology of China(center 1)and the Hefei Cancer Hospital of the Chinese Academy of Sciences(center 2),and Clinical and MRI data were analyzed.According to the 8th edition AJCC tumor regression grading(TRG)criteria for rectal cancer,patients with TRG 0-1 were classified as good responders(GR),while those with TRG 2-3 were classified as poor responders(PR)based on postoperative pathological results.The GR group comprised 77 patients,and the PR group comprised 95 patients.Patients from center 1 were randomly divided into a training set(n=92)and an internal validation set(n=40),while the patients from center 2 were utilized as an independent external validation set(n=40).High-resolution axial T2WI,diffusion-weighted imaging(DWI)and sagittal contrast-enhanced T1WI(CE-T1WI)sequences were selected to delineate the region of interest(ROI)along the tumor margins.PyRadiomics software was used to extract all radiomics features after image preprocessing.Spearman correlation analysis and least absolute shrinkage and selection operator(LASSO)analysis were used to retain the radiomics features strongly associated with the efficacy of nCRT.T2WI,DWI,CE-T1WI and multiparametric radiomics score(Rad-score)were obtained by eXtreme gradient boosting(XGBoost)classifier.The independent clinical-radiology predictors were screened by single-multiple logistic regression to build the clinical-radiology model,and the multiparametric model Rad-score combined with independent clinical-radiology predictors was selected to build the nomogram model.The performance of the model was evaluated using receiver operating characteristic(ROC)curves.The best-performing radiomics model was explained by the Shapley algorithm.Results:Univariate and multivariate logistic regression analysis identified age,tumor longest diameter,and neoadjuvant treatment modalities as independent predictors for treatment efficacy.The clinical-radiology model demonstrated the area under the curve(AUC)of 0.80(95%CI:0.75 to 0.85)in the training set,0.73(95%CI:0.68 to 0.78)in the internal validation set,and 0.60(95%CI:0.55 to 0.65)in the external validation set.Among radiomics models,the multiparametric radiomics model(T2WI+DWI+CE-T1WI)achieved optimal performance,with AUCs of 0.98(95%CI:0.95 to 1.00),0.95(95%CI:0.91 to 0.99),0.86(95%CI:0.81 to 0.91)in the training,internal validation,and external validation sets,respectively.The nomogram model achieved the best predictive performance.The AUC,accuracy,sensitivity,and specificity of the training set of nomogram model were 0.99(95%CI:0.97 to 1.00),98%,95%,and 98%,respectively.The internal validation sets were 0.98(95%CI:0.95 to 1.00),98%,98% and 98%,respectively.The external validation sets were 0.88(95%CI:0.83 to 0.93),88%,87% and 87 %respectively.DeLong test indicated that the nomogram model's performance was superior to the clinical model and the radiomics models(P<0.05).Shapley analysis revealed that wavelet-LHL_glszm_SmallAreaEmphasis in DWI sequence was the most important feature in the radiomics model.Conclusions:The nomogram based on multiparametric MRI radiomics and clinical-radiology features may be used as an accurate and non-invasive method to predict the efficacy of nCRT in rectal cancer patients,and the Shapley algorithm can provide interpretability of radiomics model.This nomogram has been validated using an external validation set,suggesting its potential utility of providing important guidance for clinical diagnosis and treatment decision-making.
作者
李雪萌
周燕飞
王傲阳
赵敏
王璟
江丽
韦炜
高飞
LI Xuemeng;ZHOU Yanfei;WANG Aoyang;ZHAO Min;WANG Jing;JIANG Li;WEI Wei;GAO Fei(Graduate School,Bengbu Medical University,Bengbu 233030,China;Department of Radiology,Hefei Cancer Hospital,Chinese Academy of Sciences,Hefei 236000,China;Graduate School,Wannan Medical College,Wuhu 241000,China;Department of Pathology,The People's Hospital of Chizhou,Chizhou 247000,China;Department of Radiology,the First Affiliated Hospital of USTC,Anhui Provincial Hospital,Hefei 230001,China)
出处
《磁共振成像》
2026年第1期59-68,122,共11页
Chinese Journal of Magnetic Resonance Imaging
基金
安徽省重点研究与开发计划项目(编号:2022e07020007)
北京医学奖励基金项目(编号:YXJL-2025-0483-0275)
中国科学技术大学附属第一医院(安徽省立医院)医学人工智能联合基金项目(编号:MAI2023C006)。
关键词
局部进展期直肠癌
新辅助放化疗
磁共振成像
多模态
影像组学
列线图
可解释性
locally advanced rectal cancer
neoadjuvant chemoradiotherapy
magnetic resonance imaging
multiparametric
radiomics
nomogram
interpretability
