【摘要】 目的 构建基于MR的组学模型并联合临床特征实现术前对前列腺特异性抗原（PSA）为4～10 ng/ml且前列腺影像报告与数据系统（PI-RADS）评分≥3分前列腺癌（PCa）的诊断。方法 回顾性分析两家医院经病理证实的PSA水平在4～10 ng/ml且PI-RADS评分≥3分的PCa 82例，良性前列腺增生150例。使用ITK-SNAP分别在T₂WI、ADC及动态增强扫描（DCE）手动勾画感兴趣区（ROI）,用python平台提取特征，FAE软件完成对数据预处理、特征选取以及组学模型构建。通过单因素及多因素Logistic回归分析临床特征确定独立预测因子来构建临床模型。通过曲线下面积（AUC）评价模型性能，选取最佳组学模型联合临床模型构建综合模型。并通过外部验证来验证模型的泛化能力。通过R软件绘制综合模型列线图，并采用校准曲线及决策曲线评估其拟合度以及临床应用价值。结果 综合模型、组学模型及临床模型的在训练集中AUC分别为0.933（95%CI:0.874～0.970）、0.915（0.852～0.957）、0.802（95%CI:0.722～0.867）;在测试集中AUC分别为0.887（95%CI:0.792～0.982）、0.872（95%CI:0.759～0.966）、0.677（95%CI:0.521～0.833）;在外部验证集中AUC分别为0.866（95%CI:0.740～0.946）、0.812（95%CI:0.676～0.908）、0.709（95%CI:0.564～0.829）,其中综合模型的诊断效能均为最佳。列线图的校准曲线结果显示预测结果与病理结果有较好的一致性。决策曲线表现出综合模型具有良好的临床应用价值。结论 基于MR的组学模型联合临床模型构建的综合模型，可以用来术前无创预测PSA为“灰区”且PI-RADS评分3分及以上的PCa,从而减少不必要的穿刺活检。更多还原

【Abstract】 Objective To construct a MR based omics model combined with clinical features to achieve preoperative non-invasive diagnosis of prostate cancer with PSA of 4-10 ng/ml and PI-RADS v2.1 score ≥3. Methods Retrospective analysis was performed on 82 cases of prostate cancer and 150 cases of benign prostatic hyperplasia with pathologically confirmed PSA levels between 4-10 ng/ml and PI-RADS score ≥3 in two hospitals.ITK-SNAP was used to manually delineate the areas of interest in T₂WI,ADC and DCE,respectively.Features were extracted using Python platform, and data preprocessing, feature selection and model construction were completed by FAE software.The clinical model was established by univariate and multivariate Logistic regression analysis to determine clinical independent predictors.The AUC value was used to evaluate the model performance, and the best omics model was combined with clinical model to construct the comprehensive model.The generalization ability of the model is verified by external validation.R software was used to draw the comprehensive model rosette, and the fitting degree and clinical application value of the rosette were evaluated by calibration curve and decision curve. Results The AUC of comprehensive model, omics model and clinical model in training set were 0.933（95%CI:0.874-0.970）,0.915（95%CI:0.852-0.957）,0.802（95%CI:0.722-0.867）,In the test set, AUC were 0.887（95%CI:0.792-0.982）,0.872（95%CI:0.755-0.947）,0.677（95%CI:0.521-0.833）,Inthe external validation set, the AUC was 0.866（95%CI:0.740-0.946）,0.812（95%CI:0.676-0.908） and 0.709（95%CI:0.564-0.829）,respectively, and the diagnostic efficiency of the comprehensive model was the best.The calibration curve of the line map was in good agreement with the pathological results.The decision curve shows good clinical application value. Conclusion The comprehensive model based on MR omics model and clinical model can be used to predict prostate cancer noninvasively with “grey area” PSA and PI-RADS V2.1 score of 3 or above, thus reducing unnecessary biopsy rate.更多还原