Abstract: Objective　To explore the application value of radiomics in the differential diagnosis of benign and malignant lesions classified as PI-RADS 3 in prostate imaging reports, providing an objective basis for clinical decision-making. Methods　A retrospective study included 96 patients who underwent prostate biopsy or surgery at Shantou Central Hospital from January 2018 to December 2024, all with preoperative magnetic resonance imaging (MRI) reports classified as PI-RADS 3 and confirmed by pathology, with an average age of (70.60±7.41) years. All patients underwent multiparametric MRI. Using 3D-Slicer software, regions of interest (ROIs) of the lesions were manually delineated on axial T2-weighted images (T2WI), and 1 127 radiomic features were extracted using open-source software FAE. In the training group (67 cases), recursive feature elimination (RFE) was employed to select key features, and a radiomic score (Radscore) model was constructed based on the coefficients of these features. The model's performance was evaluated in the validation group (29 cases) and compared with traditional indicators such as prostate-specific antigen (PSA) and apparent diffusion coefficient (ADC). Independent samples t tests, Mann-Whitney U tests, and χ² tests were used for intergroup comparisons, logistic regression analysis was conducted to build the model, and receiver operating characteristic (ROC) curves were plotted to assess diagnostic performance. Results　Six key features were selected from the 1 127 features to construct the Radscore model. In the training group, the area under the curve (AUC) of the Radscore model for distinguishing benign from malignant lesions was 0.974 [95% confidence interval (CI) 0.942-1.000], with a sensitivity of 92.9% and specificity of 100.0%. In the validation group, the AUC was 0.933 (95% CI 0.813-1.000), with a sensitivity of 95.8% and specificity of 80.0%. The diagnostic performance (AUC) of the Radscore model was significantly higher than that of the PSA and ADC value models in both the training and validation groups. Multivariable logistic regression analysis indicated that Radscore was an independent factor for predicting prostate cancer (OR=1.80, P=0.003). Conclusion　The T2WI-based radiomics model significantly improves diagnostic accuracy for PI-RADS 3 lesions by quantifying tumor heterogeneity, surpassing traditional indicators (PSA/ADC). This model provides an objective basis for clinical decision-making, helping to optimize biopsy strategies and reduce unnecessary invasive procedures.

Key words: Prostate cancer, PI-RADS 3 score, Radiomics

摘要： 目的　探讨影像组学在前列腺影像报告和数据系统（PI-RADS）3分病变良恶性鉴别诊断中的应用价值，为临床决策提供客观依据。方法　回顾性纳入2018年1月至2024年12月于汕头市中心医院行前列腺穿刺活检或手术，术前磁共振成像（MRI）报告为PI-RADS 3分且经病理证实的96例患者，年龄（70.60±7.41）岁。所有患者均行多参数MRI检查。使用3D-Slicer软件在T2加权成像（T2WI）轴位图像上手动勾画病灶的感兴趣区（ROI），通过开源软件FAE提取1 127个影像组学特征。在训练组（67例）中采用递归特征消除（RFE）筛选关键特征，并基于这些特征的系数加权构建影像组学评分（Radscore）模型。在验证组（29例）中评估模型性能，并与前列腺特异性抗原（PSA）、表观扩散系数（ADC）等传统指标进行比较。采用独立样本t检验、Mann-Whitney U检验、χ²检验进行组间比较，通过logistic回归分析构建模型，并绘制受试者操作特征曲线（ROC）评估诊断效能。结果　从1 127个特征中筛选出6个关键特征构建Radscore模型。在训练组中，Radscore模型鉴别良恶性的受试者工作特征曲线下面积（AUC）为0.974［95%置信区间（CI）0.942～1.000］，灵敏度92.9%，特异度100.0%；在验证组中，AUC为0.933（95%CI 0.813～1.000），灵敏度95.8%，特异度80.0%。训练组和验证组中Radscore模型的诊断效能（AUC）均高于PSA和ADC值模型。多因素logistic回归分析显示，Radscore是预测前列腺癌的独立因素［优势比（OR）=1.80，P=0.003］。结论　基于T2WI构建的影像组学模型，通过高通量提取并分析肿瘤的异质性特征，显著提升了PI-RADS 3分病变中前列腺癌的良恶性鉴别诊断准确性，其效能优于传统指标PSA和ADC值。该模型为临床决策提供了客观依据，有助于优化穿刺策略，减少不必要的有创检查。

关键词: 前列腺癌, PI-RADS 3分, 影像组学