目的:构建基于磁共振成像(magnetic resonance imaging,MRI)技术的影像组学特征及临床风险因素为基础的膀胱癌(bladder cancer,BCa)肌层浸润预测模型,用以术前准确且无创评估肿瘤肌层浸润情况.方法:本研究采用回顾性方法,汇集了 76例接受根治性膀胱切除术(radical cystectomy,RC)的病例,所有患者术前30 d内均行3.0T MRI扫描,入院检查至手术日的等待期间不存在外来干预措施,且术后病理均证实为BC.在MRI 诊断过程中,采取了 T2 加权成像(T2-weighted imaging,T2WI)和弥散加权成像(diffusion-weighted ima-ging,DWI)2 种序列,研究者于每例患者的T2WI和相应的表观扩散系数图(apparent diffusion coefficient,ADC)上勾画出肿瘤最大占位区域,提取影像组学特征,并运用最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)对特征进行筛选以达到降维目的.通过单因素与多因素分析同步进行,筛选出与肿瘤肌层侵袭有关的临床独立风险因素,进而共同创建影像组学与临床相关的列线图.结果:本次研究提取出影像组学特征属性共计2 286个.最终建立起的影像组学-临床融合模型指标包括影像组学特征10个和临床独立危险因子2个.曲线下面积(area under the curve,AUC)分别为0.97(训练集)和0.88(验证集),表现出良好的校准和鉴别能力.相比于单纯的影像组学模型和临床模型,影像组学-临床融合模型在校正曲线中更加贴合理想的预测情况(即贴合对角虚线代表的模型预测概率等同实际发生概率).训练集和验证集在决策曲线中净收益值均高于全干预线和无干预线,具有更高的临床净效益和使用价值.结论:相较于单纯的影像组学或临床因素模型,将两者结合的融合模型在膀胱癌肌层浸润情况上表现出更好的预测效能,有助于术前对患者进行准确、无创的评估.
Construction of a bladder cancer muscle invading prediction model based on MRI imaging radiomics
Objective:Constructing a muscle invading prediction model for bladder cancer(BCa)based on ra-diomics attributes derived from magnetic resonance imaging(MRI)technology and clinical risk factors,aimed at preoperative,accurate,and non-invasive assessment of tumor muscle infiltration.Methods:This retrospective study finally included 76 bladder cancer patients who underwent radical cystectomy(RC).All patients underwent 3.0T MRI scans within 30 days prior to surgery,with no external interventions between imaging and surgery,and postoperative pathology confirmed bladder cancer.MRI examinations included T2-weighted imaging(T2WI)and diffusion-weighted imaging(DWI)sequences.The researchers delineated the maximum tumor occupancy area on T2WI and the corresponding apparent diffusion coefficient(ADC)maps for each patient,from which radiomic fea-tures were extracted.Feature selection and dimensionality reduction were achieved using the least absolute shrink-age and selection operator(LASSO)method.Additionally,clinical risk factors associated with tumor muscle inva-sion were identified through univariate and multivariate analyses,and an integrated radiomic-clinical nomogram was constructed.Results:In this study,a total of 2 286 radiomic features were extracted.Following selection,the radiomics-clinical model constructed with 10 radiomic features and 2 clinical risk factors exhibited excellent calibra-tion and discrimination abilities,with area under the curve(AUC)of 0.97 and 0.88 in the training and validation sets,respectively.Compared to either the standalone radiomics or clinical models,the radiomics-clinical fusion model demonstrated a closer fit to the ideal prediction scenario on the calibration curve(i.e.alignment with the di-agonal dashed line representing model prediction probability equivalent to actual occurrence probability).Both the training and validation sets showed higher net benefit values on the decision curve compared to either all or none intervention lines,indicating superior clinical net benefit and utility.Conclusion:The predictive model that com-bines radiomics attributes with clinical risk factors further enhances predictive efficacy compared to models that re-ly solely on radiomics or clinical factors.It demonstrates good predictive utility for the muscle invading of bladder tumors,aiding in the preoperative,accurate,and non-invasive assessment of bladder cancer.
NETL
NSTL
万方数据
