机器学习模型联合PSMA PET影像组学与临床特征预测mCRPC患者接受177Lu-PSMA RLT的总生存期
本研究回顾性纳入102例接受[177Lu]Lu-PSMA RLT的mCRPC患者,基于治疗前[68Ga]Ga-PSMA-11 PET/CT提取影像组学特征,构建多模态机器学习模型预测总生存期。最佳模型12个月AUC为0.778,单用PET数据模型AUC为0.715,与临床模型0.766相当。
要点
- 基于治疗前PSMA PET影像组学的机器学习模型可预测mCRPC患者RLT后总生存期。
- 最佳多模态模型12个月AUC为0.778,单用PET模型AUC为0.715。
- 病灶内异质性特征与更短总生存期相关(p=0.025)。
核心结论
- 基于[68Ga]Ga-PSMA-11 PET/CT影像组学的机器学习模型可无创预测mCRPC患者接受[177Lu]Lu-PSMA RLT的总生存期。
研究设计与样本
- 回顾性单中心研究,纳入2022年1月至2024年3月接受[177Lu]Lu-PSMA治疗的102例mCRPC患者。
- 每例患者治疗前[68Ga]Ga-PSMA-11 PET/CT勾画5个病灶,按IBSI指南提取影像组学特征。
- 构建整合临床、基因组、生化及影像特征的多模态机器学习模型预测总生存期。
主要发现
- 总队列中位总生存期为11.2个月。
- 病灶内异质性(二阶影像组学特征归一化逆差矩)与更短总生存期相关(p=0.025)。
- 最佳机器学习模型整合22个临床、生化、基因组及影像组学特征,12个月AUC为0.778。
- 单用PET数据的模型AUC为0.715,与临床模型AUC 0.766相当。
临床怎么用
- 治疗前[68Ga]Ga-PSMA-11 PET/CT影像组学可用于mCRPC患者接受[177Lu]Lu-PSMA RLT前的无创风险分层。
- 该模型在真实世界环境中具有预测潜力,但需外部验证后方可用于临床决策。
局限性
- 回顾性单中心设计,样本量102例。
- 需外部验证确认模型的泛化能力。
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: Metastatic castration-resistant prostate cancer (mCRPC) remains incurable despite recent therapeutic advances. [177Lu]Lu-PSMA RLT (lutetium-177 labelled prostate-specific membrane antigen radioligand therapy) provides meaningful clinical benefit in selected patients. However, ∼30% of the patients do not respond to [177Lu]Lu-PSMA RLT. Although several baseline prognostic factors are well known, reliable pretherapeutic predictive biomarkers are still lacking. We developed and compared several predictive machine learning (ML) models integrating clinical, biochemical, and imaging data, including [68Ga]Ga-PSMA-11 PET (gallium-68 labelled prostate-specific membrane antigen ligand positron emission tomography) radiomic features, to estimate overall survival (OS) in patients with mCRPC treated with [177Lu]Lu-PSMA RLT. METHODS: This retrospective monocentric study included 102 patients with mCRPC treated with [177Lu]Lu-PSMA between January 2022 and March 2024. Five lesions per patient were segmented on their pretherapeutic [68Ga]Ga-PSMA-11 PET/computed tomography (CT) scans. Radiomic features were extracted following Image Biomarker Standardization Initiative guidelines. Multimodal ML models integrating clinical, genomic, biochemical, and imaging features (including radiomics) were developed to predict OS. RESULTS: Median OS of the total cohort was 11.2 mo. Intralesional homogeneity, characterized by the second-order radiomic feature normalized inverse difference moment, was associated with shorter OS (p = 0.025). The best-performing ML model achieved a 12-mo area under the curve (AUC) of 0.778 by combining 22 clinical, biochemical, genomic, and radiomic features. The model leveraging PET data alone achieved strong performance, with an AUC comparable to the clinical model (0.715 vs 0.766). CONCLUSIONS: ML models based on [68Ga]Ga-PSMA-11 PET/CT, including radiomic features, offer a noninvasive, standardized, and widely accessible approach to predict OS in patients with mCRPC treated with [177Lu]Lu-PSMA RLT. This study highlights the strong predictive potential of PSMA PET imaging for patient stratification in a real-world setting.
引用
- Edouard L, Papillon L, Ferrer L, et al. Machine Learning Models Combining [68Ga]Ga-PSMA-11 PET Radiomics and Clinical Features to Predict Overall Survival for Metastatic Castration-resistant Prostate Cancer Treated with [177Lu]Lu-PSMA Radioligand Therapy. Eur Urol Oncol. 2026. PMID: 42881113. doi:10.1016/j.euo.2026.09.015
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。