基于多参数MRI影像组学模型预测直肠腺瘤癌变的研究

目的 基于多参数MRI影像组学模型预测直肠腺瘤癌变.方法 回顾性分析山东省千佛山医院2016年11月至2018年12月46例经病理证实为直肠腺瘤(n=25)和直肠腺瘤癌变患者(n=21).所有患者均在术前2周接受盆腔MRI检查,包括高分辨率T2WI序列及弥散加权成像序列(DWI).通过RadCloud v2.2平台分别...

A Multiparametric MRI-Based Machine Learning Model for Preoperatively Prediction of canceration ofrectal adenoma based on multi-parameter MRI imaging omics model

Objective To predictthe canceration of rectal adenoma based on multi-parameter MRI imaging omics model.Methods A total of 46 patients with rectal adenoma(n=25)and cancerous rectal adenoma(n=21)confirmed by pathology from November 2016 to December 2018 in Shandong Provincial Qianfoshan Hospital were retrospectively analyzed.All patients underwent pelvic MRI examination 2 weeks before surgery,including high-resolution T2WI and diffusion-weighted imaging(DWI).A total of 1396 image omics features were extracted from the high-resolutionT2WI and DWI sequences respectively by RadCloud v2.2 platform.The least absolute shrinkageand selection operator(LASSO)wasused toscreen cancer-related features ofrectal adenomafrom 1396 T2WI features,1396 DWI features and 2792 combined features(T2WI sequence and DWI sequence).Logistic regression(LR)algorithm and5-fold cross-validation were used to construct three imaging omics predictionmodels:Model 1(T2WI),Model 2(DWI)and Model 3(T2WI+DWI).The diagnostic performance of the imaging omics modelwas evaluated by accuracy,sensitivity,specificity and area under the curve(AUC).Results The AUC of Model 1,Model 2 and Model 3was0.80,0.84,0.92 respectively.Model 3 showed the best predictive performance.The accuracy,sensitivity and specificity of Model 3were0.85,0.81,0.88 respectively.Conclusion Maging omics model based on multi-parameterMRI hasthe potential to predict canceration of rectal adenoma,and combination of high resolutionT2WI and DWI sequence is more effective than single sequence in predicting canceration.