Page 86 - 南京医科大学自然版

南京医科大学学报（自然科学版） 第46卷第9期
·1356 · Journal of Nanjing Medical University（Natural Sciences） 2026年9月

·临床研究·

基于mp⁃MRI和临床特征构建前列腺癌根治术的切缘预测模型

李奕博 ，于 磊 ，丁 磊 ，梁 超 ，张国巍 ，邓 鑫 1*
1
2
2
1
1，2
江苏省人民医院宿迁医院泌尿外科，江苏 宿迁 223800；南京医科大学第一附属医院泌尿外科，江苏 南京 210003
1 2
［摘 要］ 目的：基于多参数核磁共振（multiparametric magnetic resonance imaging，mp⁃MRI）联合其他具有潜在预测能力的临
床特征筛选预测因子并构建前列腺癌根治术后患者的手术切缘情况的预测模型，为手术患者提供一个可靠的预测工具，从而
辅助临床决策。方法：回顾性分析2018年1月—2024年6月于南京医科大学第一附属医院接受前列腺癌根治术的927例患者
的临床资料，包括基线特征、术前血清生化指标、mp⁃MRI影像资料、术前穿刺病理结果、根治术后病理结果等，按照7∶3的比例
随机划分为训练集和验证集。对训练集数据进行单因素、多因素Logistic回归分析和向后法逐步回归分析筛选出能预测前列
腺癌切缘情况的独立预测因子，建立预测模型并进行决策曲线分析。。在验证集数据中检验预测模型的区分能力、准确性及临
床实用性。结果：Logistic回归分析筛选出前列腺特异性抗原密度（prostate⁃specific antigen density，PSAD）、PI⁃RADS评分、病灶
数量、包膜可疑侵犯、可疑淋巴结转移5个独立预测因子（P < 0.05），以此构建预测模型，绘制列线图。训练集和验证集的受试
者工作特征（receiver operating characteristic curve，ROC）曲线下面积分别为0.897（95%CI：0.874~0.921）和0.841（95%CI：0.793~
0.888）。训练集和验证集的校准曲线均紧贴对角线，平均绝对误差（mean absolute error，MAE）分别为0.011、0.009，表明模型未
出现明显过拟合且具有较好的泛化能力。决策曲线分析结果表明模型具有临床净获益。结论：mp⁃MRI在预测前列腺癌手术
切缘方面具有显著价值，本研究构建的手术切缘预测模型具有较好的预测能力，可辅助临床决策。
［关键词］ 前列腺癌；多参数磁共振；前列腺癌根治术；手术切缘阳性；列线图；预测模型
［中图分类号］ R691 ［文献标志码］ A ［文章编号］ 1007⁃4368（2026）09⁃1356⁃10
doi：10.7655/NYDXBNSN251200

Predictive model for surgical margin status in radical prostatectomy using mp ⁃ MRI and
clinical characteristics

1
1，2
2
1
2
LI Yibo ，YU Lei ，DING Lei ，LIANG Chao ，ZHANG Guowei ，DENG Xin 1*
1 Department of Urology，Jiangsu Province（Suqian）Hospital，Suqian 223800；Department of Urology，the First
2
Affiliated Hospital of Nanjing Medical University，Nanjing 210003，China
［Abstract］ Objective：Predictors were selected and a predictive model for surgical margin status in patients after radical
prostatectomy for prostate cancer was constructed based on multiparametric magnetic resonance imaging（mp ⁃ MRI）combined with
other clinical features with potential predictive capabilities. This provides a reliable predictive tool for surgical patients，thereby
assisting clinical decision⁃making. Methods：A retrospective analysis was conducted on clinical data from 927 patients who underwent
radical prostatectomy at the First Affiliated Hospital of Nanjing Medical University between January 2018 and June 2024. The data
included baseline characteristics，preoperative serum biochemical indicators，mp ⁃ MRI imaging data，preoperative biopsy pathology
results，and postoperative pathology results. The patients were randomly divided into a training set and a validation set in a 7∶3 ratio.
Univariate logistic regression analysis，multivariate logistic regression analysis，and backward stepwise regression analysis were
performed on the training set data to identify independent predictors of prostate cancer margin status and to establish a predictive
model. The predictive model was tested in the validation set for its discriminative ability，accuracy，and clinical utility. Results：
Logistic regression analysis identified five independent predictors（P < 0.05）：prostate⁃specific antigen density（PSAD），estradiol，PI⁃
RADS score，number of lesions，suspected capsular invasion，and suspected lymph node metastasis. These predictors were used to
construct a predictive model，and a nomogram was developed. Receiver operating characteristic（ROC）curves，calibration curves，and

［基金项目］ 宿迁市科技计划项目（SY202424）
通信作者（Corresponding author），E⁃mail：dx20191128@126.com（ORCID：0009⁃0002⁃5614⁃6597）
∗