前列腺癌影像多模态AI:融合mpMRI、PSMA PET、超声与临床数据
本文综述工作流相关的多模态人工智能方法,融合多参数MRI、PSMA PET、超声(含经直肠超声和弹性成像)及临床或病理数据,用于前列腺癌诊断、局部分期与个体化治疗。主要结论为:MRI联合临床可改善临床显著前列腺癌分诊;MRI-TRUS融合可提高靶向活检定位;多模态模型可辅助包膜外侵犯预测及生化复发预测。最具备落地条件的是模块化多模态系统。
要点
- MRI联合临床的融合模型可改善临床显著前列腺癌分诊,阅片时可关注此类风险模型输出。
- MRI-TRUS融合模型在靶向活检定位上优于单模态AI和常规放射科MRI判读。
- 多模态策略可辅助包膜外侵犯量化,为保留神经手术规划提供参考。
核心结论
- 最具备落地条件的是模块化多模态系统,可改善分诊、引导活检靶向并量化局部侵犯风险,但需透明验证与人本化部署。
研究设计与样本
- 本文为综述,聚焦工作流相关的多模态AI方法,范围包括mpMRI、PSMA PET、超声及临床或病理数据融合用于诊断、局部分期与个体化治疗。
临床应用与阅片要点
- 阅片分诊时可参考MRI联合临床风险模型输出,不要仅凭单一序列或单一影像模态。
- 对需靶向活检患者,可关注MRI-TRUS融合模型的病灶定位提示,但仍需结合实时超声与操作者判断。
- 包膜外侵犯评估可借助多模态预测结果,结合PET/MRI或PET/CT联合MRI信息辅助神经保留规划。
- 治疗后随访中,融合影像与临床变量的模型可用于生化复发风险分层,但需验证后用于临床。
- 部署应选择模块化且验证路径透明的系统,并注意与现有影像工作流整合。
局限性
- 本文为综述,未报告系统检索策略、纳入研究质量评价及定量合并。
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)。
英文原文(摘要)· 点击展开
BACKGROUND: Prostate cancer imaging is inherently multimodal, yet many AI tools remain single-modality and therefore misaligned with real-world abdominal/genitourinary radiology decision-making.
PURPOSE: We review workflow-relevant multimodal AI methods that fuse mpMRI, PSMA PET, ultrasound (including TRUS and elastography), and clinical or pathology data for diagnosis, local staging, and treatment personalization.
CONTENT: MRI-plus-clinical fusion improves csPCa triage beyond imaging-only baselines and supports practical risk-model implementations. MRI-TRUS fusion models demonstrate improved lesion localization for targeted biopsy compared with unimodal AI and standard radiologist MRI interpretation in multicenter settings. For local staging, multimodal strategies for extraprostatic extension prediction are supported by meta-analytic evidence and emerging PET/MRI- or PET/CT-plus-MRI approaches that can assist radiologists and inform nerve-sparing planning. For treatment personalization, multimodal models predict biochemical recurrence after prostatectomy and extend toward systemic endpoints using imaging fused with clinical variables or pathology-derived features.
CONCLUSION: The most adoption-ready directions for Abdominal Radiology readers are modular multimodal systems that improve triage, guide biopsy targeting, and quantify local extension risk with transparent validation pathways and human-centered deployment design.
引用
- Doniyor, Jasur, Gulnihol, et al. Multimodal artificial intelligence for prostate cancer imaging: workflow-relevant fusion of mpMRI, PSMA PET, ultrasound, and clinical data for diagnosis, local staging, and treatment personalization. Abdom Radiol (NY). 2026. PMID: 42573772. doi:10.1007/s00261-026-05735-3
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。