基于计划CT的影像组学预测高危局限期前列腺癌生化失败与远处转移
该回顾性研究纳入608例接受调强放疗的高危和极高危局限期前列腺癌患者,从计划CT提取1316个影像组学特征,构建临床、影像组学及混合模型预测生化失败(BF)和远处转移(DM)。中位随访8.2年,BF和DM分别发生178例(29.3%)和52例(8.6%)。测试队列中混合模型预测BF的C指数为0.687,优于临床模型(0.637)和影像组学模型(0.640)(P < .05);预测DM的C指数分别为0.586、0.588和0.599,均无统计学显著改善。
要点
- 混合模型预测BF的C指数0.687,优于单一临床或影像组学模型(P < .05)。
- 影像组学模型预测DM的C指数0.588,混合模型0.599,均未达显著改善。
- 计划CT影像组学联合临床变量可提升高危前列腺癌BF预测。
核心结论
- 整合临床与计划CT影像组学特征可改善高危和极高危局限期前列腺癌的生化失败预测。
- 对远处转移预测未见显著改善。
研究设计与样本
- 回顾性研究纳入608例接受调强放疗的高危和极高危局限期前列腺癌患者。
- 收集5个临床变量(年龄、临床T分期、Gleason评分、前列腺特异性抗原水平、处方剂量),从计划CT提取1316个影像组学特征(形状、一阶、纹理)。
- 患者按70%和30%随机分为训练-验证队列和测试队列,按BF和DM发生率分层。
- 采用随机生存森林结合欠采样构建模型,将死亡视为竞争风险。
主要发现
- 中位随访8.2年,BF发生178例(29.3%),DM发生52例(8.6%)。
- 测试队列中预测BF的C指数:临床模型0.637、影像组学模型0.640、混合模型0.687(P < .05)。
- 测试队列中预测DM的C指数:临床模型0.586(P = .203)、影像组学模型0.588(P = .153)、混合模型0.599(P = .099)。
关键数值与推荐
- 模型性能以C指数评价,0.5为随机预测,1.0为完美预测。
- 按中位风险评分将测试队列患者分为高分组和低分组,采用Gray检验评估组间生存差异。
临床怎么用
- 计划CT影像组学联合临床变量可用于高危和极高危局限期前列腺癌生化失败的风险分层。
- 远处转移预测方面现有模型性能有限,需谨慎用于临床决策。
局限性
- 回顾性研究设计。
- 远处转移事件数较少(52例)。
- 影像组学模型预测DM的C指数未达统计学显著改善。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
PURPOSE: This study aims to assess the utility of radiomics features extracted from planning computed tomography (pCT) images in predicting biochemical failure (BF) and distant metastasis (DM) in patients with localized prostate cancer (PCa). METHODS AND MATERIALS: This retrospective study included 608 patients with high-risk and very high-risk localized PCa who received intensity modulated radiation therapy. Five clinical variables-age, clinical T-stage, Gleason score, prostate-specific antigen level, and prescribed dose-were collected. Additionally, 1316 radiomics features (shape, first-order, and texture) were extracted from pCT images. Patients were randomly classified into training-validation (70%) and test (30%) cohorts, with stratification by BF and DM incidence. Predictive models were developed to estimate BF and DM risk using 3 strategies: a clinical model based on the 5 clinical variables; a radiomics model using 5 selected radiomics features; and a hybrid model combining both. All models were constructed using random survival forest with undersampling, considering death as a competing risk. Models were then applied to the test cohort, stratifying patients into high- and low-score groups by the median risk score. Model performance was evaluated using the concordance index (C-index), where values of 0.5 and 1.0 indicate random and perfect predictions, respectively, with survival differences between the 2 groups assessed using Gray's test. RESULTS: During a median follow-up of 8.2 years, BF and DM occurred in 178 (29.3%) and 52 (8.6%) patients, respectively. In the test cohort, the C-indices for BF prediction were 0.637, 0.640, and 0.687 for the clinical, radiomics, and hybrid models, respectively (P < .05). For DM prediction, the C-indices were 0.586 (P = .203), 0.588 (P = .153), and 0.599 (P = .099), respectively. CONCLUSIONS: Integrating clinical and pCT-based radiomics features enhanced BF prediction in patients with high-risk and very high-risk localized PCa when accounting for competing risks using random survival forest. However, no significant improvement was observed for DM prediction.
引用
- Adachi T, Hirata K, Aizawa R, et al. Computed Tomography-Based Radiomics Prediction of Biochemical Failure and Distant Metastasis in Patients With High- and Very High-Risk Localized Prostate Cancer. Adv Radiat Oncol. 2025. PMID: 41635296. doi:10.1016/j.adro.2025.101916
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。